Fall 2009 Final Solution

Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science 6.041/6.431: Probabilistic Systems Analysis (Fall 2009)
Final Solutions:
December 15, 2009
Problem 2. (20 points) (a) (5 points) Were given that the joint PDF is constant in the shaded region, and since the PDF must integrate to 1, we know that the constant must equal 1 over the area of the region. Thus, c= 1 = 2. 1/2
(b) (5 points) The marginal PDFs of X and Y are found by integrating the joint PDF over all possible y s and xs, respectively. To nd the marginal PDF of X , we take a particular value x and integrate over all possible y values in that vertical slice at X = x. Since the joint PDF is constant, this integral simplies to just multiplying the joint PDF by the width of the slice. Because the width of the slice is always 1/2 for any x [0, 1], we have that the marginal PDF of X is uniform over that interval: 1, 0 x 1, fX (x) = 0, otherwise. Since the joint PDF is symmetric, the marginal PDF of Y is also uniform: 1, 0 y 1, fY (y ) = 0, otherwise. (c) (5 points) To nd the conditional expectation and variance, rst we need to determine what the condi tional distribution is given Y = 1/4. At Y = 1/4, we take a horizontal slice of a uniform joint PDF, which gives us a uniform distribution over the interval x [1/4, 3/4]. Thus, we have 1 E[X | Y = 1/4] = , 2 var(X | Y = 1/4) = (1/2)2 1 = . 12 48
(d) (5 points) At Y = 3/4, we have a horizontal slice of the joint PDF, which is nonzero when x [0, 1/4] [3/4, 1]. Since the joint PDF is uniform, the slice will also be uniform, but only in the range of x where the joint PDF is nonzero (i.e. where (x, y ) lies in the shaded region). Thus, the conditional PDF of X is 2, x [0, 1/4] [3/4, 1], fX |Y (x | 3/4) = 0, otherwise.
Page 1 of 7

Problem 3. (25 points) (a) (5 points)
The recurrent states are {3,4}.
(b) (5 points) The 2-step transition probability from State 2 to State 4 can be found by enumerating all the possible sequences. They are {2 1 4} and {2 4 4}. Thus, P(X2 = 4 | X0 = 2) = (c) (5 points)
Generally,
r11 (n + 1) =
4 j =1
1 1 1 7 + 1= . 3 6 3 18
p1j rj 1 (n).
Since states 3 and 4 are absorbing states, this expression simplies to 1 1 r11 (n + 1) = r11 (n) + r21 (n). 4 4 Alternatively, r11 (n + 1) =
4 k=1
r1k (n)pk1 1 1 + r12 (n) . 4 3
= r11 (n)
(d) (5 points) The steady-state probabilities do not exist since there is more than one recurrent class. The long-term state probabilities would depend on the initial state. (e) (5 points) To nd the probability of being absorbed by state 4, we set up the absorption probabilities. Note that a4 = 1 and a3 = 0. 1 a1 = a1 + 4 1 = a1 + 4 1 a2 = a1 + 3 1 = a1 + 3 Solving these equations yields a1 = 3 8. 1 a2 + 4 1 a2 + 4 1 1 a3 + a4 3 6 1 6
1 1 a3 + a4 3 3 1 3
Page 2 of 7

Problem 4. (30 points) (a) (5 points) Given the problem statement, we can treat Al, Bonnie, and Clydes running as 3 independent Poisson processes, where the arrivals correspond to lap completions and the arrival rates indicate the number of laps completed per hour. Since the three processes are independent, we can merge them to create a new process that captures the lap completions of all three runners. This merged process will have arrival rate M = A + B + C = 68. The total number of completed laps, L, over the rst hour is then described by a Poisson PMF with M = 68 and = 1: 68 e68 , = 0, 1, 2, . . . , ! pL () = 0, otherwise. (b) (5 points) Let L be the total number of completed laps over the rst hour, and let Ci be the number of cups of water consumed at the end of the ith lap. Then, the total number of cups of water consumed is L C= Ci ,
i=1
which is a sum of a random number of i.i.d. random variables. Thus, we can use the law of iterated expectations to nd 1 2 5 340 E[C ] = E[E[C | L]] = E[LCi ] = E[L]E[Ci ] = (M ) 1 + 2 = 68 = . 3 3 3 3 (c) (5 points) Let X be the number of laps (out of 72) after which Al drank 2 cups of water. Then, in order for him to drink at least 130 cups, we must have 1 (72 X ) + 2 X 130, which implies that we need
X 58.
Now, let Xi be i.i.d. Bernoulli random variables that equal 1 if Al drank 2 cups of water following his ith lap and 0 if he drank 1 cup. Then X = X1 + X2 + + X72 . X is evidently a binomial random variable with n = 72 and p = 2/3, and the probability we are looking for is 72 k 72k 72 2 1 P(X 58) = . k 3 3
k=58
This expression is dicult to calculate, but since were dealing with the sum of a relatively
large number of i.i.d. random variables, we can invoke the Central Limit Theorem to approx
imate this probability using a normal distribution. In particular, we can approximate X as
Page 3 of 7

a normal random variable with mean np = 72 2/3 = 48 and variance np(1 p) = 16 and approximate the desired probability as 58 48 P(X 58) = 1 P(X < 58) 1 = 1 (2.5) 0.0062. 16 (d) (5 points) The event that Al is the rst to nish a lap is the same as the event that the rst arrival in the merged process came from Als process. This probability is A 21 = . A + B + C 68 (e) (5 points) This is an instance of the random incidence paradox, so the duration of Als current lap consists of the sum of the duration from the time of your arrival until Als next lap completion and the duration from the time of your arrival back to the time of Als previous lap completion. This is the sum of 2 independent exponential random variables with parameter A = 21 (i.e. a second- order Erlang random variable): 2 21t 21 te , t 0, fT (t) = 0, otherwise. (f) (5 points) As in the previous part, the duration of Als second lap consists of the time remaining from t = 1/4 until he completes his second lap and the time elapsed since he began his second lap until t = 1/4. Let X be the time elapsed and Y be the time remaining. We can still model the time remaining Y as an exponential random variable. However, we can no longer do the same for the time elapsed X because we know X can be no larger than 1/4, whereas the exponential random variable can be arbitrarily large. To nd the PDF of X , lets rst consider its CDF. P(X x) = P(The 1 arrival occurred less than x hours ago from time 1/4) P(1 arrival in the interval [1/4 x, 1/4] and no arrivals in the interval [0, 1/4 x]) = P(1 arrival in the interval [0, 1/4]) P(1 arrival in the interval [1/4 x, 1/4])P(no arrivals in the interval [0, 1/4 x]) = P(1 arrival in the interval [0, 1/4]) P(1, x)P(0, 1/4 x) = P(1, 1/4)
e21x (21x)e21(1/4x)
e21/4 (21/4) 0, x < 0,
4x, x [0, 1/4],
= 1, x > 1/4.
=
Thus, we nd that the X is uniform over the interval [0, 1/4], with PDF 4, x [0, 1/4], fX (x) = 0, otherwise. Page 4 of 7

The total time that Al spends on his second lap is T = X + Y . Since X and Y correspond to disjoint time intervals in the Poisson process, they are independent, and therefore we can use convolution to nd the PDF of T : fT (t) = fX (x)fY (t x) dx 4 21e21(tx) dx 0 21t 21 min(1/4,t) 4e e 1 , t 0, = 0, otherwise. = Problem 5. (25 points) (a) (5 points)
Using the law of iterated expectations and the law of total variance,
E[N ] = E[E[N | X ]] = E[X ] 2 = , var(N ) = E[var(N | X )] + var(E[N | X ]) = E[X ] + var(X ) 2 2 = + 2 , where var(N | X ) = E[N | X ] = X. (b) (5 points) pN (n) =
x
min(1/4,t)
fX (x)pN |X (n | x)dx

2 n+1 (1+)x x e dx x=0 n! 2 (n + 1)! = n! (1 + )n+2 2 (n+1) n = 0, 1, 2 . . . = (1+)n+2 0 otherwise. = (c) (5 points) lin (N ), the linear least-squares estimator of X based on an observation of The equation for X N , is cov(X, N ) X (N E(N )). lin (N ) = E[X ] + var(N )
Page 5 of 7

The only unknown quantity is cov(X, N ) = E[XN ] E[X ]E[N ] = E[XN ] (E[X ])2 . Using the law of iterated expectations again, E[XN ] = E[E[XN | X ]] = E[X E[N | X ]] = E[X 2 ] = var(X ) + (E[X ])2 6 = 2. Thus, cov(X, N ) = 6/2 4/2 = 2/2 . Combining this result with those from (a),
2 lin (N ) = 2 + 2 X 2 2 + 2 2+N = . 1+
(d) (5 points) MAP (N ), the MAP estimator of X based on an observation of N is The expression for X MAP (N ) = arg max fX |N (x | n) X
x
fX (x)pN |X (n | x) x pN (n) = arg max fX (x)pN |X (n | x) = arg max

x
2 n+1 (1+)x x e x n! = arg max xn+1 e(1+)x , = arg max

x
where the third equality holds since pN (n) has no dependency on x and the last equality holds by removing all quantities that have no dependency on x. The max can be found by dierentiation and the result is: MAP (N ) = 1 + N . X 1+ This is the only local extremum in the range x [0, ). Moreover, fX |N (x | n) equals 0 at x = 0 and goes to 0 as x and fX |N (x | n) > 0 otherwise. We can therefore conclude MAP (N ) is indeed a maximum. that X (e) (5 points) To minimize the probability of error, we choose the hypothesis that has the larger posterior
Page 6 of 7

probability. We will choose the hypothesis that X = 2 if P(X = 2 | N = 3) > P(X = 3 | N = 3)
P(X = 3)P(N = 3 | X = 3)
P(X = 2)P(N = 3 | X = 2) > P(N = 3) P(N = 3)
P(X = 2)P(N = 3 | X = 2) > P(X = 3)P(N = 3 | X = 3)
23 33 e3
33 23 e2 >
35 35 3! 3! 2 3 e >e . The inequality holds so we choose the hypothesis that X = 2 to minimize the probability of error.
Page 7 of 7
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.

Fall 2009 Final Solution

Uploaded by

Copyright:

Available Formats

Fall 2009 Final Solution

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fall 2009 Final Solution

Uploaded by

Copyright:

Available Formats

Massachusetts Institute of Technology

Massachusetts Institute of Technology

r1k (n)pk1 1 1 + r12 (n) . 4 3

Massachusetts Institute of Technology

Massachusetts Institute of Technology

Massachusetts Institute of Technology

Massachusetts Institute of Technology

fX (x)pN |X (n | x) x pN (n) = arg max fX (x)pN |X (n | x) = arg max

2 n+1 (1+)x x e x n! = arg max xn+1 e(1+)x , = arg max

Massachusetts Institute of Technology

MIT OpenCourseWare http://ocw.mit.edu

6.041 / 6.431 Probabilistic Systems Analysis and Applied Probability

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.