6.041/6.

431 Spring 2008 Quiz 2

Wednesday, April 16, 7:30 - 9:30 PM.
SOLUTIONS
Question 0 1 2 Part all a b c d e f a b c d e Score Out of 3 36 4 5 5 8 5 6 4 6 6 6 6 100

Name: Recitation Instructor:

TA:
6.041/6.431:

Total

Write your solutions in this quiz packet, only solutions in the quiz packet will be graded. Question one, multiple choice questions, will receive no partial credit. Partial credit for question two and three will be awarded. You are allowed 2 two-sided 8.5 by 11 formula sheet plus a calculator. You have 120 minutes to complete the quiz. Be neat! You will not get credit if we cant read it. We will send out an email with more information on how to obtain your quiz before drop date. Good Luck!

Massachusetts Institute of Technology

Department of Electrical Engineering & Computer Science 6.041/6.431: Probabilistic Systems Analysis (Spring 2008)

Question 1
Multiple choice questions. CLEARLY circle the best answer for each question below. Each question is worth 4 points each, with no partial credit given. a. Let X1 , X2 , and X3 be independent random variables with the continuous uniform distribution over [0, 1]. Then P(X1 < X2 < X3 ) = (i) 1/6 (ii) 1/3 (iii) 1/2 (iv) 1/4 Solution: To understand the principle, rst consider a simpler problem with X1 and X2 as given above. Note that P(X1 < X2 ) + P(X2 < X1 ) + P(X1 = X2 ) = 1 since the corresponding events are disjoint and exaust all the possibilities. But P(X1 < X2 ) = P(X2 < X1 ) by symmetry. Furthermore, P(X1 = X2 ) = 0 since the random variables are continuous. Therefore, P(X1 < X2 ) = 1/2. Analogously, omitting the events with zero probability but making sure to exhaust all other possibilities, we have that P(X1 < X2 < X3 ) + P(X1 < X3 < X2 ) + P(X2 < X1 < X3 ) + P(X2 < X3 < X1 ) + P(X3 < X1 < X2 ) + P(X3 < X2 < X1 ) = 1. And, by symmetry, P(X1 < X2 < X3 ) = P(X1 < X3 < X2 ) = P(X2 < X1 < X3 ) = P(X2 < X3 < X1 ) = P(X3 < X1 < X2 ) = P(X3 < X2 < X1 ). Thus, P(X1 < X2 < X3 ) = 1/6. b. Let X and Y be two continuous random variables. Then (i) E[XY ] = E[X ]E[Y ] (ii) E[X 2 + Y 2 ] = E[X 2 ] + E[Y 2 ] (iii) fX +Y (x + y ) = fX (x)fY (y ) (iv) var(X + Y ) = var(X ) + var(Y ) Solution: Since X 2 and Y 2 are random variables, the result follows by the linearity of expecta tion. c. Suppose X is uniformly distributed over [0, 4] and Y is uniformly distributed over [0, 1]. Assume X and Y are independent. Let Z = X + Y . Then (i) fZ (4.5) = 0 (ii) fZ (4.5) = 1/8 (iii) fZ (4.5) = 1/4 (iv) fZ (4.5) = 1/2 Solution: Since X and Y are independent, the result follows by convolution: 4 1 1 fZ (4.5) = fX ()fY (4.5 ) d = d = . 8 3.5 4

Page 3 of 10

Massachusetts Institute of Technology

Department of Electrical Engineering & Computer Science 6.041/6.431: Probabilistic Systems Analysis (Spring 2008)
d. For the random variables dened in part (c), P(max(X, Y ) > 3) is equal to (i) 0 (ii) 9/4 (iii) 3/4 (iv) 1/4 Solution: Note that P(max(X, Y ) > 3) = 1 P(max(X, Y ) 3) = 1 P({X 3} {Y 3}). But, X and Y are independent, so P({X 3} {Y 3}) = P(X 3)P(Y 3). Finally, computing the probabilities, we have P(X 3) = 3/4 and P(X 3) = 1. Thus, P(max(X, Y ) > 3) = 1 3/4 = 1/4. e. Recall the hat problem from lecture: N people put their hats in a closet at the start of a party, where each hat is uniquely identied. At the end of the party each person randomly selects a hat from the closet. Suppose N is a Poisson random variable with parameter . If X is the number of people who pick their own hats, then E[X] is equal to (i) (ii) 1/2 (iii) 1/ (iv) 1 Solution: Let X = X1 + . . . + XN where each Xi is the indicator function such that Xi = 1 if the ith person picks their own hat and Xi = 0 otherwise. By the linearity of the expectation, E[X | N = n] = E[X1 | N = n] + . . . + E[XN | N = n]. But E[Xi | N = n] = 1/n for all i = 1, . . . , n. Thus, E[X | N = n] = nE[Xi | N = n] = 1. Finally, E[X ] = E[E[X | N = n]] = 1. f. Suppose X and Y are Poisson random variables with parameters 1 and 2 respectively, where X and Y are independent. Dene W = X + Y , then (i) W is Poisson with parameter min(1 , 2 ) (ii) W is Poisson with parameter 1 + 2 (iii) W may not be Poisson but has mean equal to min(1 , 2 ) (iv) W may not be Poisson but has mean equal to 1 + 2 Solution: The quickest way to obtain the answer is through transforms: MX (s) = e1 (e 1) and s s s MY (s) = e2 (e 1) . Since X and Y are independent, we have that MW (s) = e1 (e 1) e2 (e 1) = s e(1 +2 )(e 1) , which equals the transform of a Poisson random variable with mean 1 + 2 .
s

Page 4 of 10

Massachusetts Institute of Technology

Department of Electrical Engineering & Computer Science 6.041/6.431: Probabilistic Systems Analysis (Spring 2008)
g. Let X be a random variable whose transform is given by MX (s) = (0.4 + 0.6es )50 . Then (i) P(X = 0) = P(X = 50) (ii) P(X = 51) > 0 (iii) P(X = 0) = (0.4)50 (iv) P(X = 50) = 0.6 Solution: Note that MX (s) is the transform of a binomial random variable X with n = 50 trials and the probability of success p = 0.6. Thus, P(X = 0) = 0.450 . h. Let Xi , i = 1, 2, . . . be independent random variables all distributed according to the PDF 1 100 fX (x) = x/8 for 0 x 4. Let S = 100 i=1 Xi . Then P(S > 3) is approximately equal to (i) 1 (5) (ii) (5)
5 (iii) 1 2 5 (iv) 2

1 100 Solution: Let S = 100 i=1 Yi where Yi is the random variable given by Yi = X1 /100. Since Yi are iid, the distribution of S is approximately normal with mean E[S ] and variance var(S ). 3 E(S ) Thus, P(S > 3) = 1 P(S 3) 1 . Now, var(S )

4 x x3 =8 E[Xi ] = x dx = 8 24 0 3 0 4 2 2 4 8 x4 8 2 2 2x 8 var(Xi ) = E[Xi ] (E[Xi ]) = x dx = = 8 3 32 0 3 9 0

4

Therefore,
E[S ] = var(S ) = and
1 1
E[Xi ] + . . . + E[X100 ] = 8/3. 100 100 1 1 8 1 var(Xi ) + . . . + var(Xi ) = . 1002 1002 9 100

3 8 / 3 5 P(S > 3) 1 =1 .
8 1 2 9 100

Page 5 of 10

Massachusetts Institute of Technology

Department of Electrical Engineering & Computer Science 6.041/6.431: Probabilistic Systems Analysis (Spring 2008)
i. Let Xi , i = 1, 2, . . . be independent random variables all distributed according to the PDF fX (x) = 1, 0 x 1. Dene Yn = X1 X2 X3 . . . Xn , for some integer n. Then var(Yn ) is equal to n 12
1 1
(ii) n n
3 4 1
(iii) 12n
1
(iv) 12 (i) Solution: Since X1 , . . . , Xn are independent, we have that E[Yn ] = E[X1 ] . . . E[Xn ]. Sim 2 ] = E[X 2 ] . . . E[X 2 ]. Since E[X ] = 1/2 and E[X 2 ] = 1/3 for i = 1, . . . , n, it ilarly, E[Yn i n 1 i 1 1 2 2 follows that var(Sn ) = E[Yn ] (E[Yn ]) = n n . 3 4

Question 2
Each Mac book has a lifetime that is exponentially distributed with parameter . The lifetime of Mac books are independent of each other. Suppose you have two Mac books, which you begin using at the same time. Dene T1 as the time of the rst laptop failure and T2 as the time of the second laptop failure. a. Compute fT1 (t1 ). Solution Let M1 be the life time of mac book 1 and M2 the lifetime of mac book 2, where M1 and M2 are iid exponential random variables with CDF FM (m) = 1 em . T1 , the time of the rst mac book failure, is the minimum of M1 and M2 . To derive the distribution of T1 , we rst nd the CDF FT1 (t), and then dierentiate to nd the PDF fT1 (t).

FT1 (t) = P (min(M1 , M2 ) < t) = 1 P (min(M1 , M2 ) t) = 1 P (M1 t)P (M2 t) = 1 (1 FM (t))2 = 1 e2t Dierentiating FT1 (t) with respect to t yields: fT1 (t) = 2e2t t0 t0

Page 6 of 10

Massachusetts Institute of Technology

Department of Electrical Engineering & Computer Science 6.041/6.431: Probabilistic Systems Analysis (Spring 2008)
b. Let X = T2 T1 . Compute fX |T1 (x|t1 ). Solution Conditioned on the time of the rst mac book failure, the time until the other mac book fails is an exponential random variable by the memoryless property. The memoryless property tells us that regardless of the elapsed life time of the mac book, the time until failure has the same exponential CDF. Consequently, fX |T1 (x) = ex x 0.

c. Is X independent of T1 ? Give a mathematical justication for your answer. Solution Since we have shown in 2(c) that fX |T1 (x | t) does not depend on t, X and T1 are independent. d. Compute fT2 (t2 ) and E[T2 ]. Solution The time of the second laptop failure T2 is equal to T1 + X. Since X and T1 were shown to be independent in 2(b), we convolve the densities found in 2(a) and 2(b) to determine fT2 (t). fT2 (t) = =
0 t

fT1 ( )fX (t )d

2()2 e2 e(t ) d t t = 2e e d
0 0

= 2e

(1 et )

t0
1 2

Also, by the linearity of expectation, we have that E[T2 ] = E[T1 ] + E[X ] =

3 2 .

An equivalent method for solving this problem is to note that T2 is the maximum of M1 and M2 , and deriving the distribution of T2 in our standard CDF to PDF method: FT2 (t) = P(max(M1 , M2 ) < t) = P(M2 t)P (M2 t) = FM (t)2 = 1 2et + e2t Dierentiating FT2 (t) with respect to t yields: fT2 (t) = 2et 2e2t which is equivalent to our solution by convolution above.
Finally, from the above density we obtain that E[T2 ] = solution.
2

t0

t0
1 2 3 2 ,

which matches our earlier

Page 7 of 10

Massachusetts Institute of Technology

Department of Electrical Engineering & Computer Science 6.041/6.431: Probabilistic Systems Analysis (Spring 2008)
e. Now suppose you have 100 Mac books, and let Y be the time of the rst laptop failure. Find the best answer for P(Y < 0.01) Solution Y is equal to the minimum of 100 independent exponential random variables. Following the derivation in (a), we determine by analogy: fY (y ) = 100e100y t0

Integrating over y from 0 to .01, we nd P(Y < .01) = 1 e .

Your friend, Charlie, loves Mac books so much he buys S new Mac books every day! On any given day S is equally likely to be 4 or 8, and all days are independent from each other. Let S100 be the number of Mac books Charlie buys over the next 100 days. f. (6 pts) Find the best approximation for P(S100 608). Express your nal answer in terms of (), the CDF of the standard normal. Solution Using the De Moivre - Laplace Approximation to the Binomial, and noting that the step size between values that S can take on is 4, 608 + 2 100 6 608) = 100 4 10 = 20 = (.5)

P (S100

Question 3
Saif is a well intentioned though slightly indecisive fellow. Every morning he ips a coin to decide where to go. If the coin is heads he drives to the mall, if it comes up tails he volunteers at the local shelter. Saifs coin is not necessarily fair, rather it possesses a probability of heads equal to q . We do not know q , but we do know it is well-modeled by a random variable Q where the density of Q is 2q for 0 q 1 fQ (q ) = 0 otherwise Assume conditioned on Q each coin ip is independent. Note parts a, b, c, and {d, e} may be answered independent of each other.

Page 8 of 10

Massachusetts Institute of Technology

Department of Electrical Engineering & Computer Science 6.041/6.431: Probabilistic Systems Analysis (Spring 2008)
a. (4 pts) Whats the probability that Saif goes to the local shelter if he ips the coin once? Solution Let Xi be the outcome of a coin toss on the ith trial, where Xi = 1 if the coin lands heads, and Xi = 0 if the coin lands tails. By the total probability theorem: P(X1 = 0) = =
0 1

0 1

P (X1 = 0 | Q = q )f (q )dq (1 q )2q dq

1 3

In an attempt to promote virtuous behavior, Saifs father oers to pay him $4 every day he volunteers at the local shelter. Dene X as Saifs payout if he ips the coin every morning for the next 30 days. b. Find var(X ) Solution Let Yi be a Bernoulli random variable describing the outcome of a coin tossed on morning i. Then, Yi = 1 corresponds to the event that on morning i, Saif goes to the local shelter; Yi = 0 corresponds to the event that on morning i, Saif goes to the mall. Assuming that the coin lands heads with probability q , i.e. that Q = q , we have that P (Yi = 1) = q , and P (Yi = 0) = 1 q for i = 1, . . . , 30. Saifs payout for next 30 days is described by random variable X = 4(Y1 + Y2 + + Y30 ). var(X ) = 16 var(Y1 + Y2 + + Y30 ) = 16 var(E[Y1 + Y2 + + Y30 | Q]) + E[var(Y1 + Y2 + + Y30 | Q)] Now note that, conditioned on Q = q , Y1 , . . . , Y30 are independent. Thus, var(Y1 + Y2 + + Y30 | Q) = var(Y1 | Q) + . . . + var(Y30 | Q). So, var(X ) = 16 var(30Q) + 16 E[var(Y1 | Q) + . . . + var(Y30 | Q)] = 16 302 var(Q) + 16 30 E[Q(1 Q)] = 16 302 (E[Q2 ] (E[Q])2 ) + 16 30(E[Q] E[Q2 ]) = 16 302 (1/2 4/9) + 16 30(2/3 1/2) = 880 1 1 since E[Q] = 0 2q 2 dq = 2/3 and E[Q2 ] = 0 2q 3 dq = 1/2.

Page 9 of 10

Massachusetts Institute of Technology

Department of Electrical Engineering & Computer Science 6.041/6.431: Probabilistic Systems Analysis (Spring 2008)
Let event B be that Saif goes to the local shelter at least once in k days. c. Find the conditional density of Q given B , fQ|B (q ) Solution By Bayes Rule: P (B | Q = q )fQ (q ) P (B | Q = q )fQ (q )dq (1 q k )2q (1 q k )2q dq 0q1

fQ|B (q ) = = =

1
0

2q (1 q k ) 1 2/(k + 2)

While shopping at the mall, Saif gets a call from his sister Mais. They agree to meet at the Coco Cabana Court yard at exactly 1:30PM. Unfortunately Mais arrives Z minutes late, where Z is a continuous uniform random variable from zero to 10 minutes. Saif is furious that Mais has kept him waiting, and demands Mais pay him R dollars, where R = exp(Z + 2). e. Find Saifs expected payout, E[R]. Solution E[R] =
0
10

ez +2 f (z )dz

= = f. Find the density of Saifs payout, fR (r). Solution

e2 10 z e dz 10 0 e12 e2 10

FR (r) = P(eZ +2 r) = P(Z + 2 ln(r)) ln(r)2 1 = dz 10 0 ln(r) 2 = e2 r e12 10 fR (r) = 1 10r e2 r e12

Page 10 of 10

MIT OpenCourseWare http://ocw.mit.edu

6.041 / 6.431 Probabilistic Systems Analysis and Applied Probability

Fall 2010

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.