Udvalgte lsninger til

Probability
(Jim Pitman)
http://www2.imm.dtu.dk/courses/02405/ 17. december 2006

1 IMM - DTU
02405 Probability 2004-2-2 BFN/bfn

Solution for exercise 1.1.1 in Pitman

Question a)
2 3

Question b) 67%. Question c) 0.667 Question a.2)

4 7

Question b.2) 57%. Question c.2) 0.571

1 IMM - DTU
02405 Probability 2004-2-2 BFN/bfn

Solution for exercise 1.1.2 in Pitman

Question a) 8 of 11 words has four or more letters: Question b) 4 words have two or more vowels: Question c) The same words qualify (4):
4 11 4 11 8 11

1 IMM - DTU
02405 Probability 2003-9-10 BFN/bfn

Solution for exercise 1.2.4 in Pitman

It may be useful to read the denition of Odds and payo odds in Pitman pp. 6 in order to solve this exercise Question a) We dene the prot pr pr = 10(8 + 1) 100 1 = 10 Question b) The average gain pr. game is dened as the prot divided by the number of games 10 pr = = 0.1 n 100

1 IMM - DTU
02405 Probability 2004-2-4 BFN/bfn

Solution for exercise 1.3.1 in Pitman

Denote the fraction the neighbor gets by x. Then your friend gets 2x and you get 4x. 1 4 The total is one, thus x = 7 and you get 7 .

1 IMM - DTU
02405 Probability 2004-9-3 BFN/bfn

Solution for exercise 1.3.2 in Pitman

Question a) The event which occurs if exactly one of the events A and B occurs (A B c ) (Ac B) Question b) The event which occurs if none of the events A, B, or C occurs. (Ac B c C c ) Question c) The events obtained by replacing none in the previous question by exactly one, exactly two, and three Exactly one (A B c C c ) (Ac B C c ) (Ac B c C)

Exactly three (A B C)

Exactly two (A B C c ) (A B c C) (Ac B C)

1 IMM - DTU
02405 Probability 2004-2-4 KKA,BFN/bfn,kka

Solution for exercise 1.3.4 in Pitman

We dene the outcome space = {0, 1, 2} Question a) yes, {0, 1} Question b) yes, {1} Question c) no, (we have no information on the sequence) Question d) yes, {1, 2}

1 IMM - DTU
02405 Probability 2003-9-13 KKA/bfn,kka

Solution for exercise 1.3.8 in Pitman

It may be useful to make a sketch similar to the one given at page 22 in Pitman. From the text the following probabilities are given: P (A) = 0.6 P (Ac ) = 1 P (A) = 0.4 P (B) = 0.4 P (B c ) = 1 P (B) = 0.6 P (AB) = P (A B) = 0.2 Question a) P (A B) = P (A) + P (B) P (AB) = 0.6 + 0.4 0.2 = 0.8 Question b) P (Ac ) = 1 P (A) = 1 0.6 = 0.4 Question c) P (B c ) = 1 P (B) = 1 0.4 = 0.6 Question d) P (Ac B) = P (B) P (AB) = 0.4 0.2 = 0.2 Question e) P (A B c ) = 1 P (B) + P (AB) = 1 0.4 + 0.2 = 0.8 Question f ) P (Ac B c ) = 1 P (A) P (B) + P (AB) = 1 0.6 0.4 + 0.2 = 0.2

1 IMM - DTU
02405 Probability 2003-9-18 BFN/bfn

Solution for exercise 1.3.9 in Pitman

Question a) P (F G) = P (F ) + P (G) P (F G) = 0.7 + 0.6 0.4 = 0.9 using exclusion-inclusion. Question b) P (F GH) = P (F )+P (G)+P (H)P (F G)P (F H)P (GH)+P (F GH) = 0.7 + 0.6 + 0.5 0.4 0.3 0.2 + 0.1 = 1.0 using the general version of exclusion-inclusion (see exercise 1.3.11 and 1.3.12). Question c) P (H) = P ((F G)c H) + P ((F G) H) The latter probability is P ((F G) H) = P ((F H) (G H)) = P (F H) + P (G H) P (F G H) = 0.3 + 0.2 0.1 = 0.4 such that P (F c Gc H) = 0.5 0.4 = 0.1 P (F c Gc H) = P ((F G)c H)

1 IMM - DTU
02405 Probability 2003-9-11 BFN/bfn

Solution for exercise 1.3.11 in Pitman

P (A B C) = P (A (B C)) now applying inclusion-exclusion P (A(BC)) = P (A)+P (BC)P (A(BC)) = P (A)+P (BC)P ((AB)(AC)) once again we aplly inclusion-exclusion (the second and the third time) to get P (A(BC)) = P (A)+P (B)+P (C)P (BC)(P (AB)+P (AC)P ((AB)(AC))) = P (A) + P (B) + P (C) P (B C) P (A B) P (A C) + P (A B C)

1 IMM - DTU
02405 Probability 2003-9-11 BFN/bfn

Solution for exercise 1.3.12 in Pitman

We know from exercise 1.3.11 that the formula is valid for n = 3 and consider P n+1 Ai = P ((n Ai ) An+1 ) . i=1 i=1 Using exclusion-inclusion for two events we get the formula stated p.32. Since the exclusion-inclusion formula is assumed valid for n events we can use this formula for the rst term. To get through we realize that the last term P (n Ai An+1 ) i=1 is of the form P (n Bi ) i=1 with Bi = Ai An+1 , implying that we can use the inclusion-exclusion formula for this term too. The proof is completed by writing down the expansion explicitly.

1 IMM - DTU
02405 Probability 2004-2-4 BFN/bfn

Solution for exercise 1.4.1 in Pitman

Question a) Cant be decided we need to know the proportions of women and men (related to the averaging of conditional probabilities p. 41) Question b) True, deduced from the rule of averaged conditional probabilities Question c) True Question d) True Question e) 3 1 0.92 + 0.88 = 0.91 4 4

true

1 IMM - DTU
02405 Probability 2003-9-13 KKA/bfn,kka

Solution for exercise 1.4.2 in Pitman

We dene the events A The light bulb is not defect B The light bulb is produced in city B From the text the following probabilities are given: P (A|B) = 0.99 P (Ac |B) = 1 P (A|B) = 0.01 P (B) = 1/3 P (B c ) = 2/3 solution P (A B) = P (B)P (A|B) = 0.99/3 = 0.33

1 IMM - DTU
02405 Probability 2003-9-24 BFN/bfn

Solution for exercise 1.4.9 in Pitman

1 Question a) In scheme A all 1000 students have the same probability ( 1000 ) of being chosen. In scheme B the probability of being chosen depends on the school. A 1 student from the rst school will be chosen with probability 300 , from the second 1 1 with probability 1200 , and from the third with probability 1500 . The probability 1 1 of chosing a student from school 1 is p1 100 , thus p1 = 10 . Similarly we nd p2 = 2 and p3 = 1 . 5 2

1 IMM - DTU
02405 Probability 2003-9-11 BFN/bfn

Solution for exercise 1.5.3 in Pitman

C The event that the chip is ok A The event that a chip is accepted by the cheap test Question a) P (C|A) = P (A|C)P (C) 1 0.8 = P (A|C)P (C) + P (A|C c )P (C)c 0.8 + 0.1 0.2

Question b) We introduce the event S Chip sold P (S) = 0.8 + 0.2 0.1 = 0.82 The probability in question is P (C c |S) = P (S|C c )P (C c ) 0.1 0.2 1 = = P (S|C c )P (C c ) + P (S|C)P (C) 0.02 + 1 0.8 41

1 IMM - DTU
02405 Probability 2003-9-13 BFN/bfn

Solution for exercise 1.5.5 in Pitman

Dene the events H A randomly selected person is healthy D A randomly selected person is diagnosed with the disease Question a) From the text we have the following quantities P (H) = 0.99 P (D|H) = 0.05 P (D|H c ) = 0.8

and from the law of averaged conditional probabilities we get P (D) = P (H)P (D|H) + P (H c )P (D|H c ) = 0.99 0.05 + 0.01 0.8 = 0.0575 Question b) The proability in question P (H c Dc ) = P (H c )P (Dc |H c ) = 0.01 0.2 = 0.002 using the multiplication (chain) rule Question c) The proability in question P (H Dc ) = P (H)P (D c |H) = 0.99 0.95 = 0.9405 using the multiplication (chain) rule Question d) The probability in question is P (H c |D). We use Bayes rule to interchange the conditioning P (H c |D) = P (D|H c )P (H c ) = 0.8 0.010.008 + 0.05 0.99 = 0.139 P (D|H c )P (H c ) + P (D|H)P (H)

Question e) The probabilities are estimated as the percentage of a large group of people, which is indeed the frequency interpretation.

1 IMM - DTU
02405 Probability 2004-2-10 BFN/bfn

Solution for exercise 1.5.9 in Pitman

Denote the event that a shape of type i is picked by Ti , the event that it lands at by 1 F and the event that the number rolled is six by S. We have P (T i ) = 3 , i = 1, 2, 3, 2 1 P (F |T1 ) = 1 , P (F |T2 ) = 1 , and P (F |T3 ) = 3 P (S|F ) = 2 , and P (S|F c ) = 0. 3 2 Question a) We rst note that the six events Ti F and Ti F c (i = 1, 2, 3) is a partition of the outcome space. Now using The Rule of Averaged Conditional Probabilities (The Law of Total Probability) page 41

P (S) = P (S|T1 F )P (T1 F )+P (S|T2 F )P (T2 F )+P (S|T3 F )P (T3 F )+P (S|T1 F c )P (T1 F c )+P The last three terms are zero. We apply The Multiplication Rule for the probabilities P (Ti F ) leading to P (S) = P (S|T1 F )P (F |T1 )P (T1 )+P (S|T2 F )P (F |T2 )P (T2 )+P (S|T3 F )P (F |T3 )P (T3 ) a special case of The Multiplication Rule for n Events page 56. Inserting numbers P (S) = 1 111 111 121 + + = 233 223 233 4

Question b) The probability in question is P (T1 |S). Applying Bayes rule page 49 P (T1 |S) = P (S|T1 )P (T1 ) = P (S)
11 63 1 4

2 9

1 IMM - DTU
02405 Probability 2004-2-7 BFN/bfn

Solution for exercise 1.6.1 in Pitman

This is another version of the birthday problem. We denote the event that the rst n persons are born under dierent signs, exactly as in example 5 page 62. Correspondingly, Rn denotes the event that the nth person is the rst person born under the same sign as one of the previous n 1 persons. We nd
n

P (Dn ) =
i=1

i1 12

n 13

We nd P (D4 ) = 0.57 and P (D5 ) = 0.38.

1 IMM - DTU
02405 Probability 2003-9-24 BFN/bfn

Solution for exercise 1.6.5 in Pitman

Question a) We will calculate the complementary probability, the no student has the same birthday and do this sequentially. The probability that the rst student has 364 a dierent birthday is 365 , the same is true for all the remaining n 2 students. The probability in question is P (All other n 1 students has a dierent birthday than no.1) = 1 Question b) 1 364 365
n1

364 365

n1

ln (2) 1 n + 1 = 253.7 2 ln (365) ln (364)

Question c) In the birthday problem we only ask for two arbitrary birthdays to be the same, while the question in this exercise is that at least one out of n 1 has a certain birthday.

1 IMM - DTU
02405 Probability 2003-9-18 BFN/bfn

Solution for exercise 1.6.6 in Pitman

Question a) By considering a sequence of throws we get P (1) = 0 1 P (2) = 6 52 P (3) = 66 543 P (4) = 666 5434 P (5) = 6666 54325 P (6) = 66666 54321 P (7) = 66666 Question b) The sum of the probabilities p2 to p6 must be one, thus the sum in question is 1. Question c) Can be seen immediately.

1 IMM - DTU
02405 Probability 2004-9-24 BFN/bfn

Solution for exercise 1.6.7 in Pitman

Question a) The exercise is closely related to example 7 p.68. Using the same notation and approach
c c P (Current ows) = P ((S1 S2 ) S3 ) = (1 P (S1 S2 ))P (S3 ) = (1 q1 q2 )q3

(use 1 = p1 p2 + q1 p2 + p1 q2 + q1 q2 to get the result in Pitman) Question b) P (Current ows) = P (((S1 S2 ) S3 )cupS4 ) = 1 (1 q1 q2 )q3 q4 (or use exclusion/inclusion like Pitman)

1 IMM - DTU
02405 Probability 2003-9-13 BFN/bfn

Solution for exercise 1.6.8 in Pitman

question a) The events Bij occur with probability P (Bij ) = It is immediately clear that P (B12 B23 ) = 1 = P (B12 )P (B23 ). 3652 1 365

implying independence. The following is a formal and lengthy argument. Dene Aij as the the event that the ith person is born the jth day of the year.
1 We have P (Aij ) = 365 and that A1i , A2,j , A3,k , and A4,l are independent. The event Bij can be expressed by

Bij = 365 (Ai,k Aj,k ) k=1

1 such that P (Bij ) = 365 by the independence of Ai,k and Aj,k . The event B12 B23 can be expressed by

B12 B23 = 365 (A1,k A2,k A3,k ) k=1 and by the independence of the As we get P (B12 B23 ) = question b) The probability P (B13 |B12 B23 ) = 1 = P (B13 ) thus, the events B12 , B13 , B23 are not independent. question c) Pairwise independence follows from a)
1 3652

1 IMM - DTU
02405 Probability 2004-2-10 BFN/bfn

Solution for exercise 2.1.1 in Pitman

Question a) We use the formula for the number of combinations - appendix 1, page 512 (the binomial coecient) 7 4 = 7 3 = 7! 765 = = 35 4!3! 321

Question b) The probability in question is given by the binomial distribution, see eg. page 81. 3 4 5 1 35 125 35 = = 0.0156 6 6 67

1 IMM - DTU
02405 Probability 2003-9-13 BFN/bfn

Solution for exercise 2.1.2 in Pitman

We dene the events Gi: i girls in family. The probabilities P (Gi) is given by the binomial distribution due to the assumptions that the probabilities that each child is a girl do not change with the number or sexes of previous children. P (Gi) = 4 i 1 i 1 41 , 2 2 P (G2) = 6 5 8 1 3 = 16 8

P (G2c ) = 1 P (G2) =

1 IMM - DTU
02405 Probability 2004-2-10 BFN/bfn

Solution for exercise 2.1.4 in Pitman

We denote the event that there are 3 sixes in 8 rolls by A, the event that there are 2 sixes in the rst 5 rolls by B. The probability in question is P (B|A). Using the general formula for conditional probabilities page 36 P (B|A) = P (B A) P (A)

The probability P (B A) = P (A|B)P (B) by the multiplication rule, thus as a speical case of Bayes Rule page 49 we get P (B|A) = P (A|B)P (B) P (B A) = P (A) P (A)

Now the probability of P (A) is given by the binomial distribution page 81, as is P (B) and P (A|B) (the latter is the probability of getting 1 six in 3 rolls). Finally P (2 sixes in 5 rolls)P (1 six in 3 rolls) = P (B|A) = P (3 sixes in 8 rolls) 5 2
53 65

3 1
55 68

52 63

5 2

5 2 8 3

3 1

a hypergeometric probability. The result generalizes. If we have x successes in n trials then the probability of having y x successes in m n trials is given by m y n x The probabilities do not depend on p. nm xy

1 IMM - DTU
02405 Probability 2005-5-30 BFN/bfn

Solution for exercise 2.1.6 in Pitman

We dene events Bi that the man hits the bulls eye exactly i times. The probabilities of the events Bi is given by the Binomial distribution P (Bi) = 8 i 0.7i0.38i

Question a) The probability of the event P (B4) = Question b) P (B4| 8 Bi ) = i=2 Question c) 6 2 0.72 0.34 = 0.0595 P ((B4 (8 Bi )) P (B4) i=2 = == 0.1363 8 P (i=2 Bi ) 1 P (B0) P (B1) 8765 4 4 0.7 0.3 = 0.1361 4321

1 IMM - DTU
02405 Probability 2004-2-10 BFN/bfn

Solution for exercise 2.2.1 in Pitman

All questions are answered by applying The Normal Approximation to the Binomial 1 1 Distribution page 99 (131). We have = n p = 400 2 = 200, = npq = 400 2 1 = 2 10. The questions dier only in the choice of a and b in the formula. Question a) a = 190, b = 210 P (190 to 210 successes) = 210.5 200 10 189.5 200 10

= (1.05) (1.05) = 0.8531 (1 0.8531)0.7062 Question b) a = 210, b = 220 P (210 to 220 successes) = 220.5 200 10 209.5 200 10

= (2.05) (0.95) = 0.9798 0.8289 = 0.1509 Question c) a = 200, b = 200 P (200 successes) = 200.5 200 10 199.5 200 10

= (0.05) (0.05) = 0.5199 (1 0.5199) = 0.0398 Question d) a = 210, b = 210 P (210 successes) = 210.5 200 10 209.5 200 10

= (1.05) (0.95) = 0.8531 0.8289 = 0.0242

1 IMM - DTU
02405 Probability 2004-2-10 BFN/bfn

Solution for exercise 2.2.4 in Pitman

We apply The Normal Approximation to the Binomial Distribution page 99. Note 1 that b= such that the rst term is 1. We have = n p = 300 3 = 100 and
2 300 1 3 = 10 3 2 . 3

The value of a in the formula is 121 (more than 120). We get 120.5 100 8.165

P (More than 120 patients helped = 1

= 1(2.51) = 10.994 = 0.006

1 IMM - DTU
02405 Probability 2003-9-19 BFN/bfn

Solution for exercise 2.2.14 in Pitman

Question a) We dene the events W i that a box contains i working devices. The probability in question can be established by P (W 390W 391W 392W 393W 394W 395W 396W 397W 398W 399W 400) since the event W i are mutually exclusive. The probabilities P (W i) are given by the binomial distribution P (i) = 400 i 0.95i 0.05400i ,

= P (W 390)+P (W 391)+P (W 392)+P (W 393)+P (W 394)+P (W 395)+P (W 396)+P (W 397)+P (W 398

we prefer to use the normal approximation, which is 1P (less than 390 working)=1
1 390 2 400 0.95 400 0.95 0.05

= 1(2.18) = 10.9854 = 0.0146

Without continuity correction we get 1 (2.29) = 0.0110 The skewness correction is: 1 1 1 2 0.95 1 2 (2.182 1) e 2 2.18 = 0.0048 6 400 0.95 0.95 2 The skewness correction is quite signicant and should be applied. Finally we approximate the probability in question with 0.00098, which is still somewhat dierent from the exact value of 0.0092. Question b) P (at least k)=1 With
1 k + 2 400 0.95 400 0.95 0.05

0.95

we nd k = 373.

1 k + 2 400 0.95 1.645 400 0.95 0.05

1 IMM - DTU
02405 Probability 2003-9-24 BFN/bfn

Solution for exercise 2.4.7 in Pitman

Question a) From page 90 top we know that m is the largest integer less than equal to (n + 1) p = 2.6, thus m = 2. Question b) 25 2 Question c) Question d) 2.52 2.5 e = 0.2566 2! Question e) Normal m is now 250
1 250 + 2 250 2500 0.09 1 2 + 2 2.5 25 0.09

0.12 0.923 = 0.2659

1 + 1 2.5 2 25 0.09

= (0) (0.667) = 0.2475

1 250 2 250 2500 0.09

= (

1 1 ) ( ) = 0.0266 30 30

Question f ) Poisson - as above 0.2566.

1 IMM - DTU
02405 Probability 2003-9-19 BFN/bfn

Solution for exercise 2.4.8 in Pitman

The Poisson probabilities P (k) are k P (k) = e k! We use odds ratio for the probabilities P (k + 1) = P (k)
k+1 e (k+1)! k e k!

k+1

The ratio is strictly decreasing in k. For < 1 maximum will be P (0), otherwise the probabilities will increase for all k such that > k, and decrease whenever < k. For non-integer the maximum of P (k) (the mode of the distribution) is obtained for the largest k < . For intger the value of P () = P ( + 1).

1 IMM - DTU
02405 Probability 2003-9-13 BFN/bfn

Solution for exercise 2.4.10 in Pitman

The probability of the event that there is at least one success can be calculated using the Binomial distribution. The probability of the complentary event that there is no successes in n trials can be evaluated by the Poisson approximation. P (0) = e N 3 N = 0.5134 Similarly for n = 5 N 3 P (0) + P (1) = e N 3 N
1 5 1 2

1+

5 3

= 0.5037

1 IMM - DTU
02405 Probability 2004-2-10 BFN/bfn

Solution for exercise 2.5.1 in Pitman

Question a) We use the hypergeometric distribution page 125 since we are dealing with sampling without replacement 20 4 50 10 30 6

P (Exactly 4 red tickets) =

Question b) We apply the binomial distribution (sampling with replacement page 123) 4 6 20 30 24 36 10 P (Exactly 4 red tickets) = = 210 10 4 50 50 5

1 IMM - DTU
02405 Probability 2003-10-3 BFN/bfn

Solution for exercise 2.5.9 in Pitman

Question a) The probability that the second sample is drawn is the probability that the rst sample contains exactly one bad item, which occurs with probability 10 1 50 5 40 4

p1 =

(the hypergeometric distribution page 125). The probability that the second sample contains more than one bad item is calculated via the probability of the complementary event, i.e. that the second sample contains one or two bad items, which is 36 9 36 9 10 0 9 1 p2 = + 45 45 10 10 The answer to the question is the product of these two probabilities p 1 (1 p2 ) = 0.2804. Question b) The lot is accepted if we have no bad items in the rst sample or the event described under a) 10 0 50 5 40 5 10 1 50 5 40 4 9 0 45 10 36 10 9 1 45 10 36 9 = 0.4595

1 IMM - DTU
02405 Probability 2003-10-5 BFN/bfn

Solution for exercise 3.1.5 in Pitman

The random variable Z = X1 X2 has range {1, 2, 3, 4, 5, 6, 8, 9, 10, 12, 15, 16, 18, 20, 24, 25, 36}. We nd the probability of Z = i by counting the combinations of X 1 , X2 for which X1 X2 = i. we get: Z = i P (Z = i) 1 1 36 2 2 36 2 3 36 3 4 36 2 5 36 4 6 36 2 8 36 1 9 36 2 10 36 4 12 36 2 15 36 1 16 36 2 18 36 2 20 36 2 24 36 1 25 36 2 30 36 1 36 36

1 IMM - DTU
02405 Probability 2003-11-30 BFN/bfn

Solution for exercise 3.1.14 in Pitman

Question a) We dene the events Gg as the the events that team A wins in g games. The probabilities P (Gg) can be found by thinking of the game series as a sequence of Bernoulli experiments. The event Gg is the event that the fourth succes (win by team A) occurs at game g. These probabiliites are given by the negative binomial distribution (page 213 or page 482). Using the notation of the distribution summary page 482, we identify r = 4, n = g 4 (i.e. counting only the games that team A loses). We get P (Gg) = Question b)
7

g1 41

p4 q g4

g = 4, 5, 6, 7

4 g=4

g1 3

q g4

Question c) The easiest way is rst answering question d) then using 1binocdf (3, 7, 2/3) in MATLAB. 0.8267 Question d) Imagine that all games are played etc. From the binomial formula p7 + 7p6 q + 21p5 q 2 + 35p4 q 3 = p7 + p6 q + 6p6 q + 6p5 q 2 + 15p5 q 2 + 35p4 q 3 = p6 + 6p5 q + 15p4 q 2 + 20p4 q 3 = p6 + p5 q + 5p5 q + 15p4 q 2 + 20p4 q 3 etc. Question e) P (G = 4) = p4 + q 4 P (G = 6) = 10p2 q 2 (p2 + q 2 ) Independence for p = q =
1 2

P (G = 5) = 4pq(p3 + q 3 ) P (G = 7) = 20p3 q 3 (p + q)

1 IMM - DTU
02405 Probability 2003-11-21 BFN/bfn

Solution for exercise 3.1.16 in Pitman

Question a) Using the law of averaged conditional probabilities we get
n n

P (X +Y = n) =
i=0

P (X = i)P (X +Y = n|X = i) =
i=0

P (X = i)P (Y = ni)

where the last equality is due to the independence of X and Y . Question b) The marginal distribution of X and Y is P (X = 2) = 1 , 36 P (X = 3) = P (X = 6) = 1 , 18 5 , 36 P (X = 4) = P (X = 7) = 1 12 1 6

1 P (X = 5) = , 9 We get

P (X + Y = 8) = 2

1 1 1 5 + 36 36 18 9

35 1 1 = 12 12 16 81

1 IMM - DTU
02405 Probability 2003-10-6 BFN/bfn

Solution for exercise 3.1.24 in Pitman

Question a) We dene P (X even) = P (Y even) = p, and introduce the random variable W = X + Y . The probability pw of the event that W is even is pw = p2 + (1 p)(1 p) = 2p2 + 1 2p = (1 p)2 + p2 with minimum
1 2 1 for p = 2 .

Question b) We introduce p0 = P (X mod 3 = 0), p1 = P (X mod 3 = 1), p2 = P (X mod 3 = 2). The probability in question is p3 + p3 + p3 + 3p0 p1 p2 2 1 0 which after some manipulations can be written as 1 (p0 p1 + p0 p2 + p1 p2 3p0 p1 p2 ) The expressions can be maximized/minimized using standard methods, I havent found a more elegant solution than that.

1 IMM - DTU
02405 Probability 2004-5-13 BFN/bfn

Solution for exercise 3.2.3 in Pitman

Question a) Let X dene the number of sixes appearing on three rolls. We nd 52 5 3 P (X = 0) = 6 , P (X = 1) = 3 63 , P (X = 2) = 3 653 , and P (X = 3) = 613 . Using the denition of expectation page 163
3

E(X) =
x=0

x(X = x) = 0

5 6

+13

52 5 1 1 +23 3 +3 3 = 63 6 6 2

or realizing that X binomial 3, 1 example 7 page 169 we have E(X) = 3 1 = 6 6 1 . 2 Question b) Let Y denote the number of odd numbers on three rolls, then Y 1 binomial 3, 1 thus E(Y ) = 3 2 = 3 . 2 2

1 IMM - DTU
02405 Probability 2003-10-5 BFN/bfn

Solution for exercise 3.2.7 in Pitman

We dene the indicator variables Ii which are 1 of switch i are closed 0 elsewhere. We have X = I1 + I2 + + In , such that
n

E(X) = E(I1 + I2 + + In ) = E(I1 ) + E(I2 ) + + E(In ) = p1 + p2 + + pn =

pi
i=0

1 IMM - DTU
02405 Probability 2003-10-12 BFN/bfn

Solution for exercise 3.2.17 in Pitman

Question a) The event D 9 occurs if all the red balls are among the rst 9 balls drawn. The probability of this event is given by the Hypergeometric distribution p. 125 and 127. 10 3 6 3 P (D 9) = = 0.2937 13 9 Question b) 3 3 13 9 10 6 3 3 13 8 10 5

P (D = 9) = P (D 9) P (D 8) =

= 0.2284

Question c) To calculate the mean we need the probabilities of P (D = i) for i = 3, 4, . . . , 13. We get 3 3 10 i3 13 i 10 i3 13 i
10! (13i)!(i3)! 13! (13i)!i!

P (D i) =

10!i! i(i 1)(i 2) = 13!(i 3)! 13 12 11

P (D = i) = P (D i)P (D i1) =
12

E(D) =
i=3

3 3(i 1)(i 2) = i 13 12 11 13 12 11

i(i 1)(i 2) (i 1)(i 2)(i 3) 3(i 1)(i 2) = 13 12 11 13 12 11 13 12 11

12

i(i1)(i2) =
i=3

3 6, 006 = 10.5 13 12 11

1 IMM - DTU
02405 Probability 2003-10-2 BFN/bfn

Solution for exercise 3.3.4 in Pitman

The computational formula for the variance page 186 is quite useful (important). This exercise is solved by applying it twice. First we use it once to get: V ar(X1 X2 ) = E((X1 X2 )2 ) (E(X1 X2 ))2 Now by the independence of X1 and X2
2 2 2 2 E((X1 X2 )2 )(E(X1 X2 ))2 = E(X1 X2 )(E(X1 )E(X2 ))2 = E(X1 )E(X2 )(E(X1 )E(X2 ))2

using the multiplication rule for Expectation page.177 valid for independent random variables. We have also used the fact that if X1 and X2 are independent then f (X1 ) and g(X2 ) are independent too, for arbitrary functions f () and g(). We now use the computational formula for the variance once more to get V ar(X1 X2 ) = (V ar(X1 ) + (E(X1 ))2 )(V ar(X2 ) + (E(X2 ))2 ) (E(X1 )E(X2 ))2 Now inserting the symbols of the exercise we get
2 2 2 2 V ar(X1 X2 ) = 1 2 + 2 2 + 2 1 1 2

1 IMM - DTU
02405 Probability 2003-10-5 BFN/bfn

Solution for exercise 3.3.19 in Pitman

We apply the Normal approximation (the Central Limit Theorem (p.196). Let X i denote the weight of the ith passenger. The total load W is W = 30 Xi . i=1 P (W > 5000)=1 5000 30 150 55 30 = 1 (1.66) = 0.0485

1 IMM - DTU
02405 Probability 2003-10-5 BFN/bfn

Solution for exercise 3.3.23 in Pitman

We dene Sn as the time of installment of the nth battery. Similarly we dene N t to be the number of batteries replaced in the interval [0, t(. We have P (S n t) = P (Nt n), thus P (N104 26) = P (S26 104) where the time unit is weeks. We now apply the Normal approximation (Central Limit Theorem) to S26 . P (S26 104)= 104 26 4 1 104 = 0.5

1 IMM - DTU
02405 Probability 2003-10-5 BFN/bfn

Solution for exercise 3.4.2 in Pitman

First we restate D : number of balls drawn to get two of the same colour. We draw one ball which is either red or black. Having drawn a ball of some colour the number 1 of draws to get one of the same colour is geometrically distributed with probability 2 . Thus D = X + 1 where X is geometrically distributed with p = 1 . 2 Question a) P (D = i) = p(1 p)i2 , Question b) E(D) = E(X + 1) = E(X) + 1 = from page 212 or 476,482. Question c) V (D) = V (X + 1) = V (X) = from page 213 or 476,482. 1p = 2, p2 SD(D) = 2 p = 2, 3, . . . 1 +1=3 p

1 IMM - DTU
02405 Probability 2003-10-13 BFN/bfn

Solution for exercise 3.4.9 in Pitman

We dene the random variable N as the number of throws to get heads. The pay back value is N 2 , the expected win from the game can be expressed as E(N 2 10) = E(N 2 ) 10 using the rule for the expectation of a linear function of a random variable p. 175 b. We could derive E(N 2 ) from the general rule for expectation of a function of a random variable p. 175 t. However, it is more convenient to use the fact the N follows a Geometric distribution and use the Computational Formula for the Variance p. 186. E(N 2 ) = V ar(N ) + (E(N ))2 = 1p + p2 1 p
2

=2+4=6

The values for V ar(N ) and E(N ) can be found p. 476 in the distribution summary.

1 IMM - DTU
02405 Probability 2003-10-13 BFN/bfn

Solution for exercise 3.5.13 in Pitman

Question a) Using the Poisson Scatter Theorem p.230 we get (x) = x3 and (x) = Question b) 6.023 1023 = 2.688 1019 x3 22.4 103 (x) = 5.1854 109 x x

5.1854 109 x x 0.01 x 7.1914 106 2.688 1019 x3

1 IMM - DTU
02405 Probability 2003-10-5 BFN/bfn

Solution for exercise 3.5.16 in Pitman

We assume that the chocolate chips and mashmallows are randomly scattered in the dough. Question a) The number of chocoloate chips in one cubic inch is Poisson distributed with parameter 2 according to our assumptions. The number of chocolate chips in thre cubic inches is thus Poisson distributed with parameter 6. Let X denote the number of chocolate chops in a three cubic inch cookie. P (X 4) = e6 1 + 6 + 36 36 6 216 6 + + 2 6 46 = 115 e6 = 0.285

Question b) We have three Poisson variates Xi : total number of chocolate chips and marshmallows in cookie i. According to our assumptions, X1 follows a Poisson distribution with parameter 6, while X2 and X3 follow a Poisson distribution with parameter 9. The complementary event is the event that we get two or three cookies without chocoloate chips and marshmallows. P (X1 = 0, X2 = 0, X3 = 0) + P (X1 > 1, X2 = 0, X3 = 0) +P (X1 = 0, X2 > 1, X3 = 0) + P (X1 = 0, X2 = 0, X3 > 1) = e6 e9 e9 + (1 e6 )e9 e9 + e6 (1 e9 )e9 + e6 e9 (1 e9 )=0 we are almost certain that we will get at most one cookie without goodies.

1 IMM - DTU
02405 Probability 2003-10-5 BFN/bfn

Solution for exercise 3.5.18 in Pitman

Question a) The variable X1 is the sum of a thinned Poisson variable (X0 ) and a Poisson distributed random variable (the immigration). The two contributions are independent, thus X1 is Poisson distributed. The same argument is true for any n and we have proved that Xn is Poisson distributed by induction. Ee denote the parameter of the nth distribution by n . We have the following recursion: n == pn1 + with 0 = such that 1 = (1 + p) and more generally n =
i=0 n

pi =
. 1p

1 pn+1 1p

Question b) As n we get n

This value is also a xpoint of

n == pn1 +

1 IMM - DTU
02405 Probability 2003-10-15 BFN/bfn

Solution for exercise 4.1.4 in Pitman

Question a) The integral of f (x) over the range of X should be one (see e.g. page 263).
1 0 1 2

x (1 x) dx =

x
0

2 i=0 n i=0

2 i n i

(x)i dx ai bni .
2

using the binomial formula (a + b)n =

1 2

x
0

2 i=0

2 i

(x)

dx =
i=0

2 i

(x)
0

i+2

dx =
i=0

2 i

(1)

xi+3 i+3

x=1

=
x=0

1 30

such that f (x) = 30 x2 (1 x)2 Question b) We derive the mean

1 1 2

0<x<1

This is an example of the Beta distribution page 327,328,478.

xf (x)dx =
0 0

x30x2
i=0

2 i

(x)i dx = 30
i=0

2 i

(1)i

xi+4 i+4

x=1

=
x=0

1 2

which we could have stated directly due to the symmetry of f (x) around 1 , or 2 from page 478. Question c) We apply the computational formula for variances as restated page 261. V ar(X) = E(X 2 ) (E(X))2
1 2

E(X 2 ) =
0

x2 30x2
i=0

2 i

(x)i dx = 30
i=0

2 i

(1)i

xi+5 i+5

x=1

=
x=0

30 105

such that V ar(X) = which can be veried page 478. SD(X3,3 )2 =

30 1 1 = 105 4 28 1 33 = + 3 + 1) 28

(3 +

3)2 (3

1 IMM - DTU
02405 Probability 2003-10-13 BFN/bfn

Solution for exercise 4.1.5 in Pitman

Question a) Question b) We apply the formula on page 263 for a density
b

P (a X b) = We get
2

f (x)dx
a

P (1 X 2) = 1 = 2(1 x)
1 . 2

1 dx = 2(1 + |x|)2

0 1 x=2

1 dx + 2(1 x)2 =

2 0

1 dx 2(1 + x)2

x=0

1 + 2(1 + x) x=1

x=0

7 1 1 1 1 + = 2 4 2 6 12
x= x=1

1 Question c) The distribution is symmetric so P (|X| > 1) = 2P (X > 1) = 2 2(1+x) 0

Question d) No. (the integral

1 x 2(1+x)2 dx does not exist).

1 IMM - DTU
02405 Probability 2003-10-13 BFN/bfn

Solution for exercise 4.1.9 in Pitman

We rst determine S4 and V ar(S4 ). From the distribution summary page 477 we have 1 1 1 E(S4 ) = 4 2 = 2 and due to the independence of the Xi s we have V ar(S4 ) = 4 12 = 3 . (the result from the variance follows from the result page 249 for a sum of independent random variables and the remarks page 261 which states the validity for continuous distributions). We now have 3 2 = 1 (1.73) = 1 0.9582 = 0.0418 P (S4 3) = 1
1 3

1 IMM - DTU
02405 Probability 2003-11-10 BFN/bfn

Solution for exercise 4.1.13 in Pitman

Question a) We derive the density of the distribution f (x) = c(x 0.9) 0.9 < x 1.0 c(1 x) 1.0 < x < 1.1
1.1

We can nd c the standard way using 0.9 f (x)dx = 1. However, we can derive 1 the area of the triangle directly as 2 0.02 c such that c = 100. Due to the symmetry of f (x) we have P (X < 0.925) = P (1.075 < X).
0.925

P (rod scrapped) = 2P (X < 0.925) = 2

0.9

1 10(x0.9)dx = 20 x2 0.9x 2

x=0.925

= 0.0625
x=0.9

Question b) We dene the random variable Y as the length of an item which has passed the quality inspection. The probability P (0.95 < Y < 1.05) = P (0.95 < X < 1.05) 0.75 = = 0.8 P (0.925 < X < 1.075) 0.9375

The number of acceptable items A out of c are binomially distributed. We determine c such that P (A 100) 0.95 We now use the normal approximation to get 1 100 0.5 0.8 c 0.4 c 0.95

100 0.5 0.8 c 1.645 0.4 c and we nd c 134.

1 IMM - DTU
02405 Probability 2003-10-23 BFN/bfn

Solution for exercise 4.2.4 in Pitman

Question a) We dene Ti as the lifetime of component i. The probability in question is given by the Exponential Survival Function p.279. The mean is 10hours, thus = 0.1h1 . P (Ti > 20) = e0.120 = e2 = 0.1353 Question b) The problem is similar to the determination of the half life of a radioactive isotope Example 2. p.281-282. We repeat the derivation P (Ti t50% ) = 0.5 et50% = 0.5 t50% = ln 2 = 6.93

Question c) We nd the standard deviation directly from page 279 SD(Ti ) = 1 = 10

Question d) The average life time T of 100 components is 1 T = 100

100

Ti
i=1

We know from page 286 that T is Gamma distributed. However, it is more convenient to apply CLT (Central Limit Theorem) p.268 to get P (T > 11) = 1 P (T 11)=1 11 10
10 100

= 1 (1) = 0.1587

Question e) The sum of the lifetime of two components is Gamma distributed. From p.286 (Right tail probability) we get P (T1 + T2 > 22) = e0.122 (1 + 2.2) = 0.3546

1 IMM - DTU
02405 Probability 2003-10-30 BFN/bfn

Solution for exercise 4.2.10 in Pitman

Question a) We dene T1 = (T ) such that P (T1 = 0) = 1 P (T > 1) = 1 e using the survival function for an exponential random variable. Correspondingly P (K = k) = P (T > k)P (T > k+1) = ek e(k+1) = ek 1 e = e a geometric distribution with parameter p = 1 e . Question b) P (Tm = k) = P (T > pm = e m . Question c) The mean of the geometric distribution of Tm is E(Tm ) = 1 pm pm
k+1 k k+1 k )P (T > ) = e m e m = e m m m

1 e

1 e m

1 The mean is measured in m time units so we have to multiply with this fraction to get an approximate value for E(T )

E(T )= =

1 pm 1 E(Tm ) = m pm 1 for m inf ty

1 m +o m 1 e m 1 = = m 1 e m m1 1 m +o m

1 IMM - DTU
02405 Probability 2003-11-12 BFN/bfn

Solution for exercise 4.3.4 in Pitman

The relation between the hazard rate (t) and the survival function G(t) is given by (7) page 297 t G(t) = e 0 (u)du Now inserting (u) = u1 G(t) = e
t 0

u1 du

= e[u

]u=t u=0

= et

Similarly we derive f (t) from G(t) using (5) page 297 f (t) = Finally from (6) page 297 t1 et = t1 (t) = et

dG(t) = et t1 = t1 et dt

1 IMM - DTU
02405 Probability 2003-10-15 BFN/bfn

Solution for exercise 4.4.3 in Pitman

First we introduce Y = g(U ) = U 2 and note that g() is strictly increasing on ]0, 1[. We then apply the formula in the box on page 304. In our case we have fX (x) = 1 for 0 < x < 1, Inserting in the formula y = g(x) = x2 , 1 fY (y) = 2 y x= y, dy = 2x = 2 y dx

0<y<1

Alternative solution using cumulative distribution - section 4.5

FU 2 (y) = P (U 2 y) = P (U y) = y

The last equality follows from the cumulative distribution function (CDF) of a Uniformly distributed random variable (page 487). The density is derived from the CDF by dierentation (page 313) and fU 2 (y) = 1 dFU 2 (y) = ,0 < y < 1 dy 2 y

1 IMM - DTU
02405 Probability 2003-11-12 BFN/bfn

Solution for exercise 4.4.6 in Pitman

We have tan () = y and use the change of variable result page 304 to get dtan () = 1 + tan ()2 = 1 + y 2 d Now inserting into the formula page 304 we get fY (y) = 1 1 , < y < 1 + y2

The function is symmetric (fY (y) = fY (y)) since (y)2 = y 2 , but

a 0

1 1 1 dy = ln (1 + a2 ) for a 1 + y2 2

The integral yfY (y)dy has to converge absolutely for E(Y ) to exist, i.e. E(Y ) exists if and only if E(|Y |) exists (e.g. page 263 bottom).

1 IMM - DTU
02405 Probability 2003-10-16 BFN/bfn

Solution for exercise 4.5.4 in Pitman

The operations considered are shifting (addition of b) and scaling (multiplication by a). We introduce Y = aX + b. The distribution FY (y) of Y is given by FY (y) = P (Y y) = P (aX + b y) = P (aX y b) For a > 0 we get FY (y) = P For a < 0 we get FY (y) = P X yb a =1P X yb a =1F yb a X yb a =F yb a

1 IMM - DTU
02405 Probability 2003-10-29 BFN/bfn

Solution for exercise 4.5.7 in Pitman

Question a) The exercise is closely related to exercise 4.4.9 page 310, as it is the inverse problem in a special case. We apply the standard change of variable method page 304 1 dy = Y = T , T = Y 2, dt t fY (y) = 2 yey
2

a Weibull distribution. See e.g. exercise 4.3.4 page 301 and exercise 4.4.9 page 310. Question b)
0
2

2y 2 ey dy =

y 2 ey dy

We note the similarity with the variance of an unbiased (zero mean) normal variable.

y e

2 y 2

dy =

2 2

1 1 2 2 e 1 2

y2 1 2

dy =

1 1 1 2 y1 y e 2 dy 2 1 2

1 the integral is the expected value of Z 2 , where Z is normal 0, 2 distributed. 1 Thus the value of the integral is 2 Finally we get

E(Y ) = =

E(Z 2 ) =

V ar(Z) with = 3

1 1 == 2 2

= 0.51

Question c) We apply the inverse distribution function method suggested page 320323. Thus 1 U = 1 eX X = ln (1 U ) Now 1 U and U are identically distributed such that we can generate an expo1 nential X with X = ln (U ). To generate a Weibull ( = 2) distributed Y we take the square root of X, thus Y =
1 ln (1 U ).

1 IMM - DTU
02405 Probability 2003-10-17 BFN/bfn

Solution for exercise 4.5.8 in Pitman

We let Xi denote the lifetime of the ith component, and S denote the lifetime of the system. Question a) We have the maximum of two exponential random variables S = max (X 1 , X2 ). P (S t) = P (max (X1 , X2 ) t) = 1 e1 t from page 316 and example 4 page 317/318. Thus P (S > t) = 1 1 e1 t 1 e2 t = e1 t + e2 t e(1 +2 )t 1 e2 t

Question b) In this case we have S = min(X1 , X2 ) and we apply the result for the minimum of random variables page 317. The special case of two exponentials is treated in example 3 page 317 P (S t) = 1 e(1 +2 )t Question c) From the system design we deduce S = max (min (X1 , X2 ), min (X3 , X4 )) such that P (S t) = 1 e(1 +2 )t 1 e(3 +4 )t Question d) Here S = min (max (X1 , X2 ), X3 ) such that P (S t) = 1 1 1 e1 t 1 e2 t e3 t = 1e(1 +3 )t e(2 +3 )t +e(1 +2 +3 )t

1 IMM - DTU
02405 Probability 2003-11-1 BFN/bfn

Solution for exercise 4.6.3 in Pitman

Question a) P (U(1) x, U(n) y) = P (x U1 y, x U2 y, . . . x Un y) = (y x)n Question b) P (U(1) x, U(n) > y) = P (U(1) x) P (U(1) x, U(n) y) = (1 x)n (y x)n Question c) P (U(1) x, U(n) y) = P (U(n) y) P (U(1) x, U(n) y) = y n (y x)n Question d) 1 (1 x)n y n + (y x)n Question e) n k Question f ) k < x, n k 1 > y one in between xk (1 y)nk

1 IMM - DTU
02405 Probability 2003-11-12 BFN/bfn

Solution for exercise 4.6.5 in Pitman

Question a) The probability P (Xi x) = x since Xi is uniformly distributed. The number Nx of Xi s less than or equal to x follows a binomial distribution bin(n, x) since the Xi are independent. The event {X(k) x} corresponds to {Nx k}. We get n n P (X(k) x) = P (Nx k) = xi (1 x)ni i
i=k

Question b) From the boxed result at the bottom of page 327 we have that (X (k) has beta(k, n k + 1) distribution. Substituting r = k and s = n k + 1 we get
r+s1

P (X(k) x) = which is the stated result.

i=r

r+s1 i

xi (1 x)s+ri1

Question c) The beta(r, s) density is 1 1 xr1 (1 x)s1 = xr1 f (x) = B(r, s) B(r, s) Now
x x s1

i=0

s1 i

(x)i

P (X(k) x) = 1 = B(r, s)
s1 0 x

f (x)dx =
0 0

1 ur1 B(r, s)

s1

i=0

s1 i
s1

(u)i du (x)i r+i

i=0

s1 i

(u)

r+i1

xr du = B(r, s)

i=0

s1 i

as was to be proved.

1 IMM - DTU
02405 Probability 2003-11-19 BFN/bfn

Solution for exercise 5.1.4 in Pitman

Question a) This is Example 3 page 343 with dierent numbers P (|Y X| 0.25) = 1 2 1 2 3 4
2

7 16

Question b) We see that the probability can be rewritten This is example 2 page 343 with dierent values. We get 1 Question c) P (Y X|Y > 0.25) = P (Y X) P (Y X, Y 0.25) P (Y X, Y > 0.25) = P (Y > 0.25) P (Y > 0.25) =
1 2

13 14 9 = 24 25 40

1 2 3 4

1 2 4

5 8

1 IMM - DTU
02405 Probability 2003-11-1 BFN/bfn

Solution for exercise 5.1.5 in Pitman

We note that the percentile U of a randomly chosen student is uniformly(0, 1) distributed. Question a) P (U > 0.9) = 1 P (U 0.9) = 0.9 Question b) The question is Example 3 page 343 the probability of a meeting with dierent parameters. Denoting U1 and U2 respectively as the rank of the two students P (|U1 U2 | > 0.1) = 0.92 = 0.81

1 IMM - DTU
02405 Probability 2003-10-17 BFN/bfn

Solution for exercise 5.2.7 in Pitman

We denote the radius of the circle by . The are of the circle is 2 . If a chosen point is within radius r it has to be within the circle of radius r with area r 2 . We nd the probability as the fraction of these two areas FR (r) = P (R1 r) = with density (page 333) fR (r) = dFR (r) 2r = 2 dr r2 2

With R1 and R2 indpendent we have the joint density from (2) page 350 f (r1 , r2 ) = We now integrate over the set r2 < P R1 R2 2
r1 2

4r1 r2 4

(page 349) to get 4r1 r2 1 dr2 dr1 = 4 4 2

0 3 r1 dr1 =

=
0 0

r1 2

1 8

1 IMM - DTU
02405 Probability 2003-11-20 BFN/bfn

Solution for exercise 5.2.8 in Pitman

Question a) We nd the marginal density of Y by integrating over x (page 349) fY (y) = 4 c(y 2 x2 )ey dx = c y 3 ey 3 y
y

We recognize this as a gamma density (1) page 286 with = 1 and r = 4 thus 1 c= 8 Question b) With Z = g(Y ) = 4Y 3 , result page 304 we get dg(y) = 12y 2 , Y = dy
1

Z 4

1 3

, using the boxed

z 3 1 y 3 y 1 ( z ) 3 4 4 fZ (z) = e = e 6 12y 2 72

Question c) We have |X| |Y | = Y . Thus E(|X|) E(Y ) = 4.

1 IMM - DTU
02405 Probability 2003-11-22 BFN/bfn

Solution for exercise 5.2.11 in Pitman

Question a) E(X + Y ) = E(X) + E(Y ) = 1.5 from the general rule of the expectation of a sum. Question b) E(XY ) = E(X)E(Y ) = 0.5 by the independe of X and Y . Question c) E((X Y )2 ) = E(X 2 + Y 2 2XY ) = E(X 2 ) + E(Y 2 ) 2E(XY ) 1 1 4 + +1+11 = 12 4 3 from the general rule of the expectation of a sum, the computational formula for the variance, and the specic values for the uniform and exponential distributions. = (V ar(X)+(E(X))2 )+(V ar(Y )+(E(Y ))2 )2E(XY ) = Question d) E X 2 e2Y = E(X 2 )E e2Y We recall the general formula for E(g(Y )) from page 263 or 332 E(g(Y )) =
y

g(y)f (y)dy

where f (y) is the density of Y . Here Y is exponential(1) distributed with density f (y) = 1 e1y . We get E e2Y =
0

e2y 1 ey dy =

thus E X 2 e2Y is undened ().

1 IMM - DTU
02405 Probability 2003-11-1 BFN/bfn

Solution for exercise 5.2.15 in Pitman

Question a) P (a < X b, c < Y d) = P (X b, c < Y d) P (X a, c < Y d) = P (X b, Y d) P (X b, Y c) (P (X a, Y d) P (X a, Y c)) = P (X b, Y d) P (X b, Y c) P (X a, Y d) + P (X a, Y c) = F (b, d) F (b, c) F (a, d) + F (a, c) This relation can also be derived from geometric considerations. Question b) F (x, y) =
x y

f (u, v)dudv

Question c) f (x, y) = d2 F (x, y) dxdy

from the fundamental theorem of calculus. Question d) The result follows from (2) page 350 by integration.
x y x y

F (x, y) =

fX (x)fY (y)dydx =

fX (x)dx

fY (y)dy = FX (x)FY (y)

Alternatively dene the indicator I(x, y) variables such that I(x, y) = 1 if X x and Y y and 0 otherwise. Note that F (x, y) = P (I(x, y) = 1) = E(I(x, y)) and apply the last formula on page 349. Question e) See also exercise 4.6.3 c). We nd F (x, y) = P (U(1) x, U(n) y) = P (U(n) y) P (U(1) > x, U(n) y) P (U(n) y) P (x < U1 y, x < U2 y, . . . , x < Un y) = y n (y x)n We nd the density as d2 F (x, y) = n(n 1)(y x)n2 dxdy

1 IMM - DTU
02405 Probability 2004-4-15 BFN/bfn

Solution for exercise 5.3.6 in Pitman

Question a) P (N (0, 13) > 5) = 1 Question b) 1 (1 (1))2 Question c) Drawing helpful, suggests that the following should be true (1) (1) Question d) P (1 > max (X, Y ) min (X, Y ) = P (1 > |X Y |) = 1 2 1 2 5 13

1 IMM - DTU
02405 Probability 2003-11-19 BFN/bfn

Solution for exercise 5.3.12 in Pitman

Question a) Let the coordinates shot i be denoted by (Xi , Yi ). The dierence between two shots (X2 X1 , Y2 Y1 ) is two independent normally distributed random variables with mean 0 and variance 2. a simple a scaling in example 1 problem By 2 page 361 we get E(D) = 2 = . 2 Question b) We have E(D 2 ) = 4 thus V ar(D) = 4 .

1 IMM - DTU
02405 Probability 2003-11-22 BFN/bfn

Solution for exercise 5.3.15 in Pitman

Question a) This is exercise 4.4.10 b). We recall the result Introducing Y = g(Z) = Z2
1 2 1 fZ (z) = e 2 z , 2

y = g(z) = z 2 ,

z=

y,

dy = 2z = 2 y dz

Inserting in the boxed formula page 304 and use the many to one extension. fY (y) =
y 1 e 2 2y

0<y<

We recognize the gamma density with scale parameter = 1 and shape parameter 2 1 r = 2 from the distribution summary page 481. By a slight reformulation we have
y 1 2 2 e 2 fY (y) = 2

and we deduce have

1 2

Question b) The formula is valid for n = 1. Assuming the formula valid for odd n we get n n+2 = +1 Gamma 2 2 The recursive formula for the gamma-function page 191 tells us that (r + 1) = r(r) and we derive n (n 1)! n+2 = Gamma 2 2 2n1 n1 ! 2 n = 2
n1 2

i=1

1 2

Question c) Obvious by a simple change of variable. Question d) From the additivity of the gamma distribution, which we can prove directly Question e) From the interpretation as sums of squared normal variables.

2
r Question f ) The mean of a gamma (r, ) distribution is , thus n has mean
n 2 1 2

= n.

The variance of a gamma (r, ) distribution is n 2 1 = 2n. Skewness bla bla bla
4

r , 2

thus the variance of n is

1 IMM - DTU
02405 Probability 2004-4-15 BFN/bfn

Solution for exercise 5.4.3 in Pitman

For = we have the Gamma(2, ) distribution. We denote the waiting time in queue i by Xi , and the total waiting time by Z. Question a) The distribution of the total waiting time Z is found using the density convolution formula page 372 for independent variables.
t

f (t) =
0

eu e(tu) du = et
0

eu() du =

et et

Question b) E(Z) = E(X1 ) + E(X2 ) =

1 1 +

See e.g. page 480 for the means E(Xi ) for the exponential variables . Question c) Using the independence of X1 and X2 we have V ar(Z) = V ar(X1 ) + V ar(X2 ) = The last equalit follows from e.g. page 480. 1 1 + 2 2

1 IMM - DTU
02405 Probability 2003-11-20 BFN/bfn

Solution for exercise 5.4.4 in Pitman

Question a) We introduce the random variable X1 as the time to failure of the rst component and X2 as the additional time to failure of the second component. From the assumption X1 and X2 are independent and exponentially distributed with intensity 2. The sum of two independent exponentially distributed random variables is gamma(2,2) distributed. Question b) The mean of the gamma distribution is 1 2 = 22 (page 286,481). (2)2 Question c) 1 e2t0.9 (1 + 2t0.9 ) = 0.9 e2t0.9 (1 + 2t0.9 ) = 0.1
2 2

and the variance is

1 IMM - DTU
02405 Probability 2003-11-11 BFN/bfn

Solution for exercise 5.4.6 in Pitman

The argument of example 2 page 375 is easily generalized. Since X i is gamma(ri , ) distributed we can write Xi as
ri

Xi =
j=1

Wij

where Wij are independent exponential() variables. Thus

n n ri

Xi =
i=1 i=1 j=1

Wij
n i=1 ri , )

a sum of distributed.

n i=1 ri

exponential() random variables. The sum is gamma(

1 IMM - DTU
02405 Probability 2004-4-17 BFN/bfn

Solution for exercise 6.1.5 in Pitman

Question a) The probability in distribution in question is P (X 1 = x1 |X1 + X2 = n). Using the denition of conditioned probabilities P (X1 = x1 |X1 + X2 = n) = = P (X1 = x1 , X1 + X2 = n) P (X1 + X2 = n)

P (X1 = x1 )P (X2 = n x1 ) P (X1 = x1 , X2 = n x1 ) = P (X1 + X2 = n) P (X1 + X2 = n)

where we have used the independence of X1 and X2 and the last equality. Now using the Poisson probability expression and the boxed result page 226 P (X1 = x1 |X1 + X2 = n) =
nx x1 2 1 n! 1 = = x1 !(n x1 )! (1 + 2 )n 1 1 1 2 1 2 e (nx1 )! e x1 ! (1 +2 )n (1 +2 ) e n!
x nx

n x1

px1 (1 p)nx1

with p =

1 . 1 +2

Question b) Let Xi denote the number of eggs laid by insect i. The probability in question is P (X1 90) = P (X2 60). Now Xi binomial 150, 1 . With the 2 normal approximation to the binomial distribution page 99 to get P (X2 60) = 60 + 1 150 2 1 150 2
1 2

29 150

= (2.37) = 0.0089

1 IMM - DTU
02405 Probability 2003-11-19 BFN/bfn

Solution for exercise 6.1.6 in Pitman

Question a) We recall the denition of conditional probability P (A|B) = such that
m P (AB) , P (B)

P (N1 = n1 , N2 = n2 , . . . Nm = nm |

Ni = n)
i=1

P (N1 = n1 , N2 = n2 , . . . Nm = nm P ( m Ni = n) i=1

m i=1

Ni = n)

Now realising that P (N1 = n1 , N2 = n2 , . . . Nm = nm m Ni = n) = P (N1 = i=1 m n1 , N2 = n2 , . . . Nm = nm ) and using the fact that N = i=1 Ni has Poisson m distribution with parameter = i=1 i we get
m

P (N1 = n1 , N2 = n2 , . . . Nm = nm | such that with n =

m i=1

Ni = n) =
i=1

m i i i i=1 ni ! e (
m n i=1 i m i=1 ni )!

ni
m

P (N1 = n1 , N2 = n2 , . . . Nm = nm |

Ni = n) =
i=1

n! n1 !n2 ! nm !
i .

n1

n2

nm

a multinomial distribution (page 155) with probabilities p i = Question b) Using

P (N1 = n1 , N2 = n2 , . . . Nm = nm ) = P (N = n)P (N1 = n1 , N2 = n2 , . . . Nm = nm | we see that the Ni s are independent Poisson variables.

Ni = n)
i=1

1 IMM - DTU
02405 Probability 2003-11-19 BFN/bfn

Solution for exercise 6.2.18 in Pitman

By denition V ar(Y ) =
y

(yE(Y ))2 f (y) =

y

(yE(Y ))2
x

f (x, y) =
x y

(yE(Y ))2 f (x, y)

We now apply the crucial idea of adding 0 in the form of E(Y |x) E(Y |x) inside the brackets. V ar(Y ) = (y E(Y |x) + E(Y |x) E(Y ))2 f (x, y)
x y

Next we multiply with one in the form of V ar(Y ) =

x y

f (x) f (x)

(y E(Y |x) + E(Y |x) E(Y ))2 thus

f (x, y) f (x) f (x)

By denition fY (y|x) = V ar(Y ) =

x

f (x,y) f (x)

(y E(Y |x) + E(Y |x) E(Y ))2 fY (y|x) f (x)

Expanding the square sum we get V ar(Y ) =

x y

(y E(Y |x))2 + (E(Y |x) E(Y ))2 fY (y|x) f (x)

since

y (y

E(Y |x)) = 0. Now (y E(Y |x))2 fY (y|x) f (x)+ (E(Y |x) E(Y ))2 fY (y|x) f (x)

V ar(Y ) =
x y

the inner part of the rst term is V ar(Y |X = x) while the inner part of the second term is constant. Thus V ar(Y ) =
x

V ar(Y |X = x)f (x) +

(E(Y |x) E(Y ))2 f (x)

leading to the stated equation V ar(Y ) = E(V ar(Y |X)) + V ar(E(Y |X)) an important and very useful result that is also valid for continuous and mixed distributions. Mixed distributions are distributions that are neither discrete nor continuous.

1 IMM - DTU
02405 Probability 2003-11-19 BFN/bfn

Solution for exercise 6.3.5 in Pitman

We note that Y for given X = x is uniformly distributed, on 1 + x for 1 < x < 0 and on 1 x for 0 < x < 1. Thus F (y|x) = P (Y y|X = x) = y , 0 < y < 1 |x| 1 |x|
1 |x 2

1 Question a) We have P Y 2 |X = x = 1 F 1 Question b) We have P Y 2 |X = x = F 1 |x 2

Question c) Since Y for given X = x is uniformly distributed we can apply results for the uniform distribution, see e.g. the distribution summary page 477 or 487. We get 1 |x| E(Y |X = x) = 2 Question c) Similarly V ar(Y |X = x) = (1 |x|)2 12

1 IMM - DTU
02405 Probability 2003-11-19 BFN/bfn

Solution for exercise 6.3.14 in Pitman

We have immediately
n

P (X1 = x1 , X2 = x2 , . . . , Xn = xn ) =
i=1

pXi (1 p)1Xi = p

n i=1

Xi

(1 p)n

n i=1

Xi

The posterior density of p given X1 = x1 , X2 = x2 , . . . , Xn = xn is f (p|X1 = x1 , X2 = x2 , . . . , Xn = xn ) = = f (p; X1 = x1 , X2 = x2 , . . . , Xn = xn ) f (X1 = x1 , X2 = x2 , . . . , Xn = xn )

1 0

f (X1 = x1 , X2 = x2 , . . . , Xn = xn |p)f (p)

Inserting the previous result to get

f (X1 = x1 , X2 = x2 , . . . , Xn = xn |p)f (p)dp p

1 0
n i=1 n i=1

Xi

f (p|X1 = x1 , X2 = x2 , . . . , Xn = xn ) =

Xi (1

(1 p)n

n i=1 n i=1

Xi

f (p)

p)n

Xi f (p)dp n i=1

which only dependes on the Xi s through their sum. Introducing Sn = rewrite f (p|X1 = x1 , X2 = x2 , . . . , Xn = xn ) =
1 0

Xi we

pSn (1 p)nSn f (p)

pSn (1 p)nSn f (p)dp

We note that if the prior density of p f (p) is a beta(r, s) distribution, then the posterior distribution is a beta(r + Sn , s + n Sn ) distribution.

1 IMM - DTU
02405 Probability 2003-11-12 BFN/bfn

Solution for exercise 6.4.5 in Pitman

Question a) We calculate the covariance of X and Y using the denition page 630. Cov(X, Y ) = E(XY ) E(X)E(Y ) = E(XY ) since E(X) = 0 We calculate E(XY ) = E(X 3 ) = thus X and Y are uncorrelated. Question b) We have P Y > 1 1 |X| > 4 2 =1=P Y > 1 4 1 x3 dx = 0 2 1
1

thus X and Y are not independent.

1 IMM - DTU
02405 Probability 2003-11-11 BFN/bfn

Solution for exercise 6.4.6 in Pitman

X and Y are clearly not indpendent. P (X = 0|Y = 12) = P (X1 X2 = 0|X1 +X2 = 12) = 1 = P (X1 X2 = 0) = P (X = 0) However, X and Y are uncorrelated: Cov(X, Y ) = E((X E(X))(Y E(Y ))) = E(XY ) E(X)E(Y ) = E(XY )
2 2 2 2 = E((X1 X2 )(X1 + X2 )) = E(X1 X2 ) = E(X1 ) E(X2 ) = 0

using the denition of covariance page 630

1 IMM - DTU
02405 Probability 2004-5-13 BFN/bfn

Solution for exercise 6.4.7 in Pitman

Question a) X2 0 0 1 1 X3 0 1 0 1 X2 + X 3 0 1 1 2 X2 X 3 0 -1 1 0 -1 0 0
1 6

Probability
1 3 1 6 1 3 1 6

X2 + X 3 / X 2 X 3 0 1 2

0 0
1 6 1 3

1 0 0
1 3 1 = 6. 1 3

1 3 Question b) With Z2 = X2 X3 we get E((X2 X3 )3 ) = E(Z2 ) = 6 +

Question c) X2 and X3 are independent thus uncorrelated. The new variables Z 1 = 2 2 X2 + X3 and Z2 = X2 X3 are correlated. E(Z1 Z2 ) = E(X2 ) E(X3 ) = 1 1 = 2 3 1 1 = 5 6 = E(Z1 )E(Z2 ) 6 6

1 IMM - DTU
02405 Probability 2003-11-11 BFN/bfn

Solution for exercise 6.5.4 in Pitman

Question a) We have from the boxed result page 363 that X + 2Y is normally distributed with mean = 0 + 2 0 = 0 and variance 2 = 1 + 4 1 = 5. We now evaluate P (X + 2Y 3) = P X + 2Y 3 5 5 = 3 5 = (1.34) = 0.9099

Question b) We have from the boxed result page 451 1 Y = X+ 2 1 1 Z 4

where X and Z are indpendent standard normal variables. Thus X + 2Y = 2X + 3Z This is the sum of two independent normal variables which itself is N ormal(0, 2 2 + 2 3 ) distributed. Thus P (X + 2Y 3) = 3 7 = (1.13) = 0.8708

1 IMM - DTU
02405 Probability 2004-5-13 BFN/bfn

Solution for exercise 6.5.6 in Pitman

Question a) From the boxed result page 363 we know that Z = X kY is normal(0, 1 + k 2 ) 1 distributed, thus P (X kY > 0) = 2 .
1 Question b) Arguing along the same lines we nd P (U > kV ) = 2 .

P (X > kY ) = P (X kY > 0)

Question c) P (U 2 + V 2 < 1) = P (3X 2 + Y 2 + 2 3XY + X 2 + 3Y 2 2 3XY < 1) =P X2 + Y 2 < 1 4 = 1 e 8 = 0.118

1

where we have used X 2 + Y 2 exponential(0.5) in the last equality (page 360, 364-366, 485). Question d) X =v+ 3Y normal(v, 3)

1 IMM - DTU
02405 Probability 2003-12-12 BFN/bfn

Solution for exercise 6.5.10 in Pitman

Question a) We rst note from page that since V are W are bivariate normal, then W W V V Y = X= V W are bivariate standardized normal. From page we have that we can write Y = X + where X and Z are standardized independent normal variables. Thus any linear combination of V and W will be a linear combination of X and Z. We know from chapter 5. that such a combination is a normal variable. After some tedious calculations we nd the actual linear combinations to be aV + bW = aV + bW + (aV + bW )X + b2 and cV + dW = cV + dW + (cV + dW )X + d2 1 2 Z 1 2 Z 1 2 Z

2 2 Such that (aV + bW normal(aV + bW , a2 V + b2 W + 2abV W ) and (cV + 2 2 2 2 dW normal(cV + dW , c V + d W + 2cdV W ).

Question b) We have from question a) that V1 = aV + bW = 1 + 11 X + 12 Z W1 = cV + dW = 2 + 21 X + 22 Z for some appropriate constants. We can rewrite these expressions to get V1 1 11 X + 12 Z W1 2 21 X + 22 Z = = Y1 = X1 = 2 2 2 2 2 2 2 2 11 + 12 11 + 12 21 + 22 21 + 22 such that X1 and Y1 are standard normal variables. We see that with some eort we would be able to write Y1 = 1 X 1 + 1 2 Z1 1

and we conclude from page 454 that V1 and W2 are bivariate normal variables. Question c) We nd the parameters using standard results for mean and variance 1 = E(aV + bW ) = aV + bW
2 1

2 = E(cV + dW ) = cV + bW
2 2 2 2 = c2 V + d2 W + 2cdV W

2 a 2 V

2 b 2 W

+ 2abV W

We nd the covariance from = E[(a(V V ) + b(W bW ))(c(V V ) + d(W W ))] E((aV + bW (aV + bW ))(cV + dW (cV + dW )))

etc

1 IMM - DTU
02405 Probability 2003-9-11 BFN/bfn

Solution for review exercise 1 (chapter 1) in Pitman

Dene the events B0 : 0 defective items in box B1 : 1 defective item in box B2 : 2 defective items in box I : Item picked at random defective The question can be stated formally(mathematically) as P (B2|I) = 1 0.03 6 P (I|B2)P (B2) = = P (I|B0)P (B0) + P (I|B1)P (B1) + P (I|B2)P (B2) 0 0.92 + 0.5 0.05 + 1 0.03 11

1 IMM - DTU
02405 Probability 2004-2-10 BFN/bfn

Solution for review exercise 3 (chapter 1) in Pitman

The outcomes of the experiment are HHH, HHT, HT H, HT T, T HH, T HT, T T H, T T T taking the sequence into account, assuming that these 8 outcomes are equally likely we 1 see that the probability that the coin lands the same way at all three tosses is 4 . The aw in the argument is the lack of independence. We use knowledge obtained from the experiment to choose the tosses which satisfy the requirement that the coin landed the same way at these specic tosses. It is thus less likely that the toss not chosen in the selection procedure had the same result, as one can verify by examining the outcome space.

1 IMM - DTU
02405 Probability 2004-10-16 BFN/bfn

Solution for review exercise 10 (chapter 1) in Pitman

We dene the events Ek Exactly k blood types are represented Ai i persons have blood type A Bi i persons have blood type B Ci i persons have blood type C Di i persons have blood type D

Question a) P (E2 ) = P (A2 ) + P (B2 ) + P (C2 ) + P (D2 ) = p2 + p2 + p2 + p2 = 0.3816 a b c d Question b) We have p(k) = P (Ek ). By combinatorial considerations we can show P (Ai1 Bi2 Ci3 Di4 ) = (i1 + i2 + i3 + i4 )! i1 i2 i3 i4 pa pb pc pd i1 !i2 !i3 !i4 !

with i1 + i2 + i3 + i4 = 4, in our case. We have to sum over the appropriate values of (i1 , i2 , i3 , i4 ). It is doable but much more cumbersome to use basic rules. We get p(1) = 0.0687 p(2) = 0.5973 p(3) = 0.3163 p(4) = 0.0177

p(1) = P (E1 ) = P (A4 ) + P (B4 ) + P (C4 ) + P (D4 ) = p4 + p4 + p4 + p4 = 0.0687 a b c d p(4) = P (E4 ) = P (A1 B1 C1 D1 ) = 24pa pb pc pd = 0.0177 To calculate p(3) = P (E3 ) we use the law of averaged conditional probabilities
4

p(3) = P (E3 ) =
i=0

P (E3 |Ai )P (Ai ).

We immediately have P (E3 |A4 ) = P (E3 |A3 ) = 0

2 To establish P (E3 |A2 ) we argue P (E3 |A2 ) = P (B1 C1 |A2 )+P (B1 D1 |A2 )+P (C1 D1 |A2 ) = further P (E3 |A0 ) = P (B2 C1 D1 |A0 )+P (B1 C2 D1 |A0 )+P (B1 C1 D2 |A0 ) = 4pb pc pd (pb + pc + pd (1 pa )4 pb pc + pb pd + pc pd (1 pa )2

To evaluate P (E3|A1 ) we use the law of averaged conditional probability once more (see Review Exercise 1.13)
4

P (E3|A1) =
i=1

P (E3|A1 Bi )P (Bi |A1 ) 3pc pd (pc + pd ) (1 pa pb )3

with P (E3|A1 B0 ) = P (E3|A1 B1 ) =

p2 + p2 c d (1 pa pb )2 pc + pd P (E3|A1 B2 ) = 1 pa pb P (E3|A1 B3 ) = 0

and we get P (E3|A1) = 3pc pd (pc + pd ) (1 pa pb )3 1 pa pb 1 pa

3

p2 + p2 c d (1 pa pb )2

1 IMM - DTU
02405 Probability 2003-10-2 BFN/bfn

Solution for review exercise 15 (chapter 1) in Pitman

Dene the events Bi that box i is chosen, and the event G that a gold coin is found. We have 1 P (G|B1) = 1, P (G|B2) = 0, P (G|B3) = 2 We want to nd P (B1|G). The probability is found using Bayes rule (p.49) P (B1|G) = 2 P (G|B1)P (B1) = P (G|B1)P (B1) + P (G|B2)P (B2) + P (G|B3)P (B3) 3

1 IMM - DTU
02405 Probability 2003-10-12 BFN/bfn

Solution for review exercise 13 (chapter 2) in Pitman

The probability that the manufacturer will have to replace a packet is
50

P (replace) =
i=3

50 i

0.01 0.99

50i

=1

i=0

50 i

0.01i 0.9950i

= 0.9950 1 +

0.01 0.01 49 50 1 + 0.99 0.99 2

= 0.0138

Pitman claims this probability to be 0.0144. We evaluate the second probability using the Normal approximation to the Binomial distribution. Let X denote the number of packets the manufacturer has to replace. The random variable X follows a Binomial distribution with n = 4000 and p =. We can evaluate the probability using the normal approximation. P (X > 40) = 1 P (X 40)=1 1 14.77 7.38
1 40 + 2 4000 0.0138 4000 0.0138 0.9862

= 1 (2.00) = 0.9772

Slightly dierent from Pitmans result due to the dierence above.

 sprpr3

r2 2

r3 3

pr2f !7`

j l Iyx

 (

wvv

r h t h x Dprym t h n} x h o n DIpIrDIp

rpmhrDIDpIphpIpspruDpzBu h x l n n t j } x l h o n ~ l h q h x v v u U v u y U h fr h

n h h r

v Cu

hrpmDvrDIIyIp`ICI(mpICrDI8pIpwDppsrI x h} l n j j x h o n l t n x o n o | t h q t } | { h ~}} x n h x v u t n l h q h h o y UQ ~ t v u h o n t h n l s2CppwhCpi n l h q h h o n t t h { h l x | h hsr8IDDCI8ybig 7`

~} n | h | j h n n t l x n t j x n x o pDCDrpC`DIyppzCryIpn n l h q h h o n j l x ~} n | h | j h n n t l x n h x t n v ~} n | x { h n x o n v u t n l h q h h o n h l k h j h ur8IpXIypDCrpC`DIpDpuCpDIfhDp8zympwxDsr8IpmIig

f c e ed

g e q s c bhd

bx!vs s e q s y

w s e t s q q W xvufrhpi

g W e c a Y W hfdb`XV

TR P H F US& QIGE  A @  A @ 9   DBCB88 6 4010 & ' & $ " 7532()(%#!    

 

1 IMM - DTU
02405 Probability 2003-10-12 BFN/bfn

Solution for review exercise 25 (chapter 2) in Pitman

Question a) We dene the events Ai that player A wins in i sets. We have immediately P (A3) = p3 Correspondingly, player A can win in 4 sets if he wins 2 out of the rst 3 and the 4th. P (A4) = p p q p + p q p p + q p p p = 3p3 q similary we nd P (A5) = 6p3 q 2 Question b) The event A (player A wins) is A = A1 A2 A3. The events Ai are mutually exclusive and we get P (A) = P (A1 A2 A3) = P (A1) + P (A2) + P (A3) = p3 (1 + 3q + 6q 2 ) Question c) The question can be reformulated as P (A3|A) = P (A3 A) 1 = P (A) 1 + 3q + 6q 2

using the general formula for conditional probability p.36. Question d) 3 8

Question e) Pitman suggests no, which is reasonable. However, the way to assess whether we can assume independence or not would be to analyze the distribution of the number of sets played in a large number of matches.

1 IMM - DTU
02405 Probability 2003-10-3 BFN/bfn

Solution for review exercise 33 (chapter 2) in Pitman

Question a) Throw the coin twice, repeat if you get two heads. The event with 1 probability 3 now occurs if you got two tails, otherwise the complentary event occurred. Question b) Throw the coin twice, repeat until you get one head and one tail. Then use HT or T H as the two possibilities.

1 IMM - DTU
02405 Probability 2003-10-15 BFN/bfn

Solution for review exercise 35 (chapter 2) in Pitman

Question a)
35

i=20

1000 i

1 38

37 38

1000i

1 37 Question b) The standard deviation 1000 38 38 =5.1 is acceptable for the Normal approximation. 1 1 1000 1 20 2 38 35 + 2 1000 38 = (1.814) (1.346) = 0.8764 1 37 1 37 1000 38 38 1000 38 38

1 IMM - DTU
02405 Probability 2003-10-13 BFN/bfn

Solution for review exercise 19 (chapter 3) in Pitman

Question a) P (Y X) =
x=0

P (X = x)P (Y X|X = x)

now X and Y are independent such that P (Y X) =

x=0

P (X = x)P (Y x)

There is a convenient formula for the tail probabilities of a geometric distribution, see eg. page 482. We need to adjust this result to the present case of a geometric distribution with range 0, 1, . . . (counting only failures), such that P (Y x) = (1 p)x . We now insert this result and the Poisson densities to get P (Y X) =
x=0

x e (1 p)x = e e(1p) = ep x!
((1p))x x=0 x!

where we have used the exponential series Question b)

1

= e(1p) .

ep = e 2 = 0.6065

1 IMM - DTU
02405 Probability 2003-10-13 BFN/bfn

Solution for review exercise 25 (chapter 3) in Pitman

Question a) The joint distribution of (Y1 , Y2 ) is given by Y1 /Y2 0 1 2 0
9 36 6 36 3 36

1
6 36 4 36 2 36

2
3 36 2 36 1 36

as a check we verify that the sum of all entries in the table is 1. We derive the distribution of Y1 + Y2 Y1 + Y 2 = i 0 1 2 3 4 4 1 9 P (Y1 + Y2 = i) 36 12 10 36 36 36 36 Question b) E(3Y1 +2Y2 ) = E(3Y1 )+E(2Y2 ) = 3E(Y1 )+2E(Y2 ) = 5E(Y1 ) = 5 0 1 1 1 +1 +2 2 3 6 = 10 3

The rst equality is true due to the addition rule for expectations (page 181), the second equality is true due to the result for linear functions of random variables page 175 b., the third equality is true since Y1 and Y2 has the same distribution, and the fourth equality is obtained from the denition of the mean see page 181. Question c) for X 3 0 1 for 4 X 5 f (x) = 2 for X = 6

or something similar.

1 IMM - DTU
02405 Probability 2003-11-6 BFN/bfn

Solution for review exercise 29 (chapter 3) in Pitman

Question a) We note that the probability does not depend on the ordering, i.e. the probability of a certain sequence depends on the number of 1s among the X i s not on the ordering. k1 nk1 (w + jd) j=0 (b + jd) j=0
k1 j=0 (b

+ w + jd)

Question b) To obtain the distribution of Sn the number of black balls drawn, we n dierent sequences each with the probability derived in note that there is k question a) that lead to the event Sn = k. P (Sn = k) = n k
k1 j=0 (b

+ jd)
k1 j=0 (b

nk1 (w j=0

+ jd)

+ w + jd)

Question c) n k 1 k!(n k)! = (n + 1)! n+1

Question d) Not independent since, but interchangeable Question e) We approach the question by induction. We rst show P (X1 = 1) = b b+w

We then derive P (Xn+1 = 1) assuming P (Xn = 1) = bb + w in a Polya model.

P (Xn+1 = 1) = P (Xn+1 = 1|X1 = 1)P (X1 = 1)+P (Xn+1 = 1|X1 = 0)P (X1 = 0) = P (Xn+1 = 1|X1 = To proceed we note that the probability P (Xn+1 = 1|X1 = 1) is the probability of P (Yn = 1) in an urn scheme starting with b + d blacks and w whites, thus b+d P (Xn+1 = 1|X1 = 1) = P (Yn = 1) = b+w+d . Correspondingly P (Xn+1 = 1|X1 = b 0) = b+w+d . Finally P (Xn+1 = 1) = b+d b w b b + = b+w+db+w b+w+db+w b+w

2 Question f ) P (X5 = 1|X10 = 1) = P (X10 = 1|X5 = 1)P (X5 = 1) = P (X10 = 1|X5 = 1) P (X10 = 1)

using Bayes rule, or from the exchangeability. From the exchangeability we also have b+d P (X10 = 1|X5 = 1) = P (X2 = 1|X1 = 1) = b+w+d

1 IMM - DTU
02405 Probability 2003-10-22 BFN/bfn

Solution for review exercise 34 (chapter 3) in Pitman

Question a) The function gz (x) = z x denes a function of x for any |z| < 1. For xed z we can nd the E(gz (X)) using the denition in the box on the top of page 175. We nd E(gz (X)) = E(z ) =
X x=0

z x P (X = x)

However, this is a power series in z that is absolutely convergent for |z| 1 and thus denes a C function of z for |z| < 1. Question b) The more elegant and maybe more abstract proof is GX+Y (z) = E z X+Y = E z X z Y From the independence of X and Y we get (page 177) GX+Y (z) == E z X E z Y = GX (z)GY (z) The more crude analytic proof goes as follows GX+Y (z) = E z
X+Y k=0 k=0 k

z P (X+Y = k) =

k i=0

P (X = i, Y = k i)

again from the independence of X and Y we get GX+Y (z) =

k=0 k

k i=0

P (X = i)P (Y = k i)

z k P (X = i)P (Y = ki)

i=0 k=i

The interchange of the sums are justied since all terms are positive. The rearrangement is a commonly used tool in analytic derivations in probability. It is quite instructive to draw a small diagram to verify the limits of the sums. We now make further rearrangements GX+Y (z) =
i=0

i=0 k=i

z k P (X = i)P (Y = k i)
i=0

z i P (X = i)

k=i

z ki P (Y = k i) =

z i P (X = i)

m=0

z m P (Y = m)

by a change of variable (m = k i). Now GX+Y (z) =

z i P (X = i)

z m P (Y = m) =

i=0

z i P (X = i)GY (z) = GX (z)GY (z)

i=0

m=0

2 Question c) By rearranging Sn = (X1 + + Xn1 ) + Xn we deduce

n

GSn (z) =
i=1

GXi (z)

We rst nd the generating function of a Bernoulli distributed random variable(binomial with n = 1)

1

E(z ) =
x=0

z x P (X = x) = z 0 (1 p) + z 1 p = 1 p(1 z)

Now using the general result for Xi with binomial distribution b(ni , p) we get E(z Xi ) = (E(z X ))ni = (1 p(1 z))ni Generalizing this result we nd E(z Sn ) = (1 p(1 z))
n i=1

ni

i.e. that the sum of independent binomially distributed random variables is itself binomially distributed provided equality of the p i s. Question d) The generating function of the Poisson distribution is given in exercise 3.5.19. Such that
n

GSn (z) =
i=1

ei (1z) = e

n i=1

i (1z)

The result proofs that the sum of independent Poisson random variables is itself Poisson. Question e) GX (z) = Question f ) G Sn = zp 1 z(1 p) zp 1 z(1 p) G Sn = zp 1 z(1 p)
n i=1 ri

1 IMM - DTU
02405 Probability 2003-11-1 BFN/bfn

Solution for review exercise 7 (chapter 4) in Pitman

Question a) We require ex dx = 1. 0

f (x)dx =

e|x| dx = 1. We have =

since

Question b) We immediately get E(X) = 0 since f (x) is symmetric around zero. The second moment E(X 2 ) is identical to the second moment of the standard exponential, which we can nd from the computational formula for the variance. We additionally have V ar(X) = E(X 2 ) since E(X) = 0. 1 V ar(X) = E(X ) = 2 +
2

2 2

Question c) P (|X| > y) = 2P (X > y) = 2

y

t e dt = 2

et dt = ey

the standard exponential survival function. Question d) From the result in c) we are lead to P (X x) = 0.5
1 x e 2 1 + 2 ex

x<0 0<x

1 IMM - DTU
02405 Probability 2003-10-16 BFN/bfn

Solution for review exercise 13 (chapter 4) in Pitman

We introduce the random variables Nloc (t) and Ndis (t) as the number of local respectively long distance calls arriving within time t (where t is given in minutes). Question a) P (Nloc (1) = 5, Ndis (1) = 3) = P (Nloc (1) = 5)P (Ndis (1) = 3) due to the independence of the Poisson processes. The variables N loc (t) and Ndis (t) has Poisson distributions (page 289) such that P (Nloc (1) = 5, Ndis (1) = 3) =
5 3 (loc 1)5 1 (dis 1)3 1 dis loc e loc e dis = e loc dis 5! 3! 5!3!

Question b) The sum of two indpendent Poisson random variables is Poisson distributed (boxed result page 226), leading to P (Nloc (3) + Ndis (3) = 50) = ((loc + dis )3)50 ( + )3 e loc dis 50!

Question c) We now introduce the random variables Siloc and Sidis as the time of the ith local and long distance call respectively. These random variables are Gamma distributed according to the box on the top of page 286 or to 4. page 289 The probability in question can be expressed as The waiting time to the rst long distance in terms of calls are geometrically distributed P (X > 10) = (1 pdis )10 = loc loc + dis
10

1 IMM - DTU
02405 Probability 2003-11-2 BFN/bfn

Solution for review exercise 21 (chapter 4) in Pitman

Question a) We rst note using exercise 4.3.4 page 301 and exercise 4.4.9 page 310 1 hat R1 and R2 are both Weibull = 2, = 2 distributed. The survival function 1 2 is thus (from E4.3.4) G(x) = e 2 x . We now apply the result for the minimum of independent random variables page 317 to get P (Y y) = P (min (R1 , R2 ) y) = 1P (R1 > y, R2 > y) = 1P (R1 > y)(R2 > y) = 1 e 2 y e 2 y = 1 ey
1 2 1 2 2

a new Weibull distribution with = 2 and = 1. If we did not recognize the distribution as a Weibull we would derive the survival function of the R i s by P (Ri > x) =
x

ue 2 u du = e 2 x

We nd the density using (5) page 297 or directly using E4.3.4 (i) fY (y) = 2yey
2

Question b) This is a special case of E4.4.9 a). We can re-derive this result using the dg(y) = 2y. change of variable formula page 304. With Z = g(Y ) = Y 2 we get dy Inserting we get 2 1 = ez fZ (z) = 2yey 2y an exponential(1) distribution. Question c) We have E(Z) = 1 (see e.g. the mean of an exponential variable page 279 or the distribution summary page 477 or page 480).

1 IMM - DTU
02405 Probability 2003-10-16 BFN/bfn

Solution for review exercise 23 (chapter 4) in Pitman

We introduce Y = M 3 such that Y has the exponential distribution with mean 2. Question a) E(M ) = E(Y + 3) = E(Y ) + 3 = 5 V ar(M ) = V ar(Y + 3) = V ar(Y ) = 4

where we have used standard rules for mean and variance see eg. page 249, and the result page 279 for the variance of the exponential distribution. Question b) We get the density fM (m) of the random variable M is 1 1 fM (m) = e 2 (m3) 2 m > 3.

from the stated assumptions. We can apply the box page 304 to get
1 2 (log (x)3) e2 e fM (m) 2 2 fX (x) = = = , dx x x x dm
1 3

x > e3

where X = g(M ) = eM . Alternatively FX (x) = P (X x) = P (log (X) log (x) = P (log (X) 3 log (x) 3) = P (Y log (x) 3) = 1 e taking derivative we get
e dFX (x) 2 == , fX (x) = dx x x
3 2 (log (x)3) 2

e2 =1 x

x > e3

x > e3

Question c) We do the calculations in terms of the random variables Y i = Mi 3, Mi = log (Xi ). Here Xi denotes the magnitude of the ith earthquake. From Example 3 page 317 we know that the minimum Z of the Yi s, Z = min (Y1 , Y2 ) is exponentially distributed with mean 1. P (M > 4) = P (Z > 1) = e1

1 IMM - DTU
02405 Probability 2003-11-1 BFN/bfn

Solution for review exercise 25 (chapter 4) in Pitman

1 Question a) We rst note that the range of Y is 0 < Y 2 .

P (Y y) = P

1 2

(Y y|U
1 2

1 +P 2

1 <U 2

1 (Y y| < U 2

= 2P (U y)

The density is 2 for 0 < y <

0 elsewhere.

Question b) The standard uniform density f (y) = 1 for 0 < y < 1, 0 elsewhere. Question c) E(Y ) =
1 2

0 1 = , V ar(Y ) = 2 4

1 2

0 12

1 48

1 IMM - DTU
02405 Probability 2003-11-12 BFN/bfn

Solution for review exercise 26 (chapter 4) in Pitman

Question a) E(Wt ) = E XetY = E(X)E etY by the independence of X and Y . We nd E etY from the denition of the mean. 3 2 2et t tY e2 1 = E e ety 2dy = t 1 Inserting this result and E(X) = 2 we get 2et t E(Wt ) = 2 e2 1 t Alternatively we could derive the joint density of X and Y to f (x, y) = 2(2x)3 e2x , 0 < x, 0 < y < 1

where we have used that X has Gamma (4,2) density, and apply the formula for E(g(X, Y )) page 349. Question b) Since X and Y are independent we nd E(Wt2 ) E(Wt2 ) = E(X 2 )E etY
2

where E(X 2 ) = V ar(X) + (E(X))2 = 5, see eg. page 481. Next we derive E etY
2

e2t t e 1 t

and apply the computational formula for the variance page 261 SD(Wt ) = 5 e2t t 2et t e2 1 (e 1) 2 t t
2

1 IMM - DTU
02405 Probability 2003-11-1 BFN/bfn

Solution for review exercise 1 (chapter 5) in Pitman

First apply the denition of conditional probability page 36 P 1 Y |Y X 2 2 P Y 1 Y X2 2 = P (Y X 2 )

The joint density of X and Y is the product of the marginal densities since X and Y are independent (page 349). We calculate the denominator using the formula for the probability of a set B page 349
1 1 x2 1

P (Y X 2 ) = and the numerator P Y

1 1 dydx =

(1 x2 )dx = 1

1 2 = 3 3

1 Y X2 2

= P (Y X 2 ) P

Y <

1 Y X2 2

Now for the last term P 1 Y < Y X2 2 = Finally we get P 1 Y |Y X 2 2 = =

0
1 2 1 2

x2

1 dydx =

1 2

1 ( x2 )dx 2

1 11 1 1 1 = 2 2 32 2 3 2
2 3

1 3 2

2 3

=1

2 4

1 IMM - DTU
02405 Probability 2003-10-17 BFN/bfn

Solution for review exercise 20 (chapter 5) in Pitman

Question a) This is example 3 page 317. A rederivation gives us P (Tmin t) = 1 P (Tmin > t) = 1 P (T1 > t, T2 > t) with T1 and T2 independent we get P (Tmin t) = 1 P (T1 > t)P (T2 > t) now inserting the exponential survival function page 279 we get P (Tmin t) = 1 1 1 e1 t 1 1 e2 t = 1 e(1 +2 )t

the cumulative distribution function of an exponentially distributed random variable with parameter 1 + 2 . Question b) This question is Example 2 page 352. A slightly dierent handling of the integrals gives us P (T1 < T2 ) =
0 0 t1 0

1 e1 t1 2 e2 t2 dt2 dt1 fT1 (t1 )P (T2 > t1 )dt1

1 e1 t1 e2 t1 dt1 =

which is an application of the rule of averaged conditional probability (page 41) for a continuous density. The general result is stated page 417 as the Integral Conditioning Formula. We get P (T1 < T2 ) =
0

1 e1 t1 e2 t1 dt1 =

1 1 + 2

Question c) Consider P (Tmin > t|Xmin = 2) = P (T1 > t|T2 > T1 ) = P (T1 > t, T2 > T1 ) P (T1 > t, T2 > T1 ) = P (T2 > T1 ) P (Xmin = 2)

We evaluate the probability in the denominator by integrating the joint density over a proper region (page 349), similarly to example 2 page 352 P (T1 > t, T2 > T1 ) =
t t1

1 e1 t1 2 e2 t2 dt2 dt1

2 =
t

1 e1 t1 e2 t1 dt1 =

1 e(1 +2 )t 1 + 2

By inserting back we nally get P (Tmin > t|Xmin = 2) = e(1 +2 )t = P (Tmin > t) such that Tmin and Xmin are independent. Question d) We can dene Xmin = i whenever Tmin = Ti . Then P (Xmin = i) = i , and Tmin and Xmin are independent. 1 ++n

1 IMM - DTU
02405 Probability 2004-5-13 BFN/bfn

Solution for review exercise 8 (chapter 6) in Pitman

Question a) Using the multiplication rule see e.g. page 425 top we get 1 x f (x, y) = fY (y)fX (x|Y = y) = 2e2y e y y The marginal density fX (x) of X is given by fX (x) =
0

1 x 2e2y e y dy y

a non-standard density. Question b) Using average conditional expectation page 425 bottom we get E(X) = E(E(X|Y )) = E(Y ) = noting that the roles of X and Y are interchanged. Question c) Similarly E(XY ) = E(E(XY |Y )) = E(Y E(X|Y )) = E(Y 2 ) = V ar(Y ) + (E(Y ))2 = 1 2 1 2

We have E(X 2 ) = E(E(X 2 |Y )) = E(2Y 2 ) = 1. Thus V ar(X) = SD(X)2 = 1 1 3 1 4 1 1 = 4 and SD(Y ) = 2 . Finally Corr(X, Y ) = 2 3 1 = 33 4
2 2