6.041/6.

431 Spring 2009 Final Exam

Thursday, May 21, 1:30 - 4:30 PM.

Question Part Score Out of

0 2
1 all 18
2 all 24
3 a 4
b 4
c 4
4 a 6
Name: b 6
c 6
5 a 6
Recitation Instructor: b 6
6 a 4
b 4
c 4
d 5
e 5
7 a 6
b 6
Total 120

• Write your solutions in this quiz packet, only solutions in the quiz packet will be graded.
• You are allowed three two-sided 8.5 by 11 formula sheet plus a calculator.
• You have 180 minutes to complete the quiz.
• Be neat! You will not get credit if we can’t read it.
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

Problem 1: True or False (2pts. each, 18 pts. total)

No partial credit will be given for individual questions in this part of the quiz.

a. Let {Xn } be a sequence of i.i.d random variables taking values in the interval [0, 0.5]. Consider
the following statements:

(A) If E[Xn2 ] converges to 0 as n → ∞ then Xn converges to 0 in probability.

(B) If all Xn have E[Xn ] = 0.2 and var (Xn ) converges to 0 as n → ∞ then Xn converges to 0.2
in probability.
(C) The sequence of random variables Zn , defined by Zn = X1 · X2 · · · Xn , converges to 0 in
probability as n → ∞.

Which of these statements are always true? Write True or False in each of the boxes below.

A: True B: True C: True

Solution:

(A) True. The fact that limn→∞ E[Xn2 ] = 0 implies limn→∞ E[Xn ] = 0 and limn→∞ var(Xn ) = 0.
Hence, one has

P (|Xn − 0| ≥ ǫ) ≤ P (|Xn − E[Xn ]| ≥ ǫ/2) + P (|E[Xn ] − 0| ≥ ǫ/2)

var(Xn )
≤ + P (|E[Xn ] − 0| ≥ ǫ/2) → 0,
(ǫ/2)2

where we have applied Chebyshev inequality.

(B) True. Applying Chebyshev inequality gives

var(Xn )
P (|Xn − E[Xn ]| ≥ ǫ) ≤ → 0.
ǫ2
Hence Xn converges to E[Xn ] = 0.2 in probability.
(C) True. For all ǫ > 0, since Zn ≤ (1/2)n ⇒ P (|Zn − 0| ≥ ǫ) = 0 for n > − log ǫ/ log 2.

b. Let Xi (i = 1, 2, . . . ) be i.i.d. random variables with mean 0 and variance 2; Yi (i = 1, 2, . . . ) be

i.i.d. random variables with mean 2. Assume that all variables Xi , Yj are independent. Consider
the following statements:
X1 +···+Xn
(A) n converges to 0 in probability as n → ∞.
2
X1 +···+Xn 2
(B) n converges to 2 in probability as n → ∞.
X1 Y1 +···+Xn Yn
(C) n converges to 0 in probability as n → ∞.

Page 3 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

Which of these statements are always true? Write True or False in each of the boxes below.

A: True B: True C: True

Solution:

(A) True. Note that E[ X1 +···+X

n
n
] = 0 and var( X1 +···+X
n
n
)= n·2
n2
= 2
n. One can see X1 +···+Xn
n
converges to 0 in probability.
(B) True. Let Zi = Xi2 and E[Zi ] = 2. Note Zi are i.i.d. since Xi are i.i.d., and hence one has
that Z1 +···+Z
n
n
converges to E[Zi ] = 2 in probability by the WLLN.
(C) True. Let Wi = Xi Yi and E[Wi ] = E[Xi ]E[Yi ] = 0. Note Wi are i.i.d. since Xi and Yi are
respectively i.i.d., and hence one has that W1 +···+W
n
n
converges to E[Wi ] = 0 in probability
by the WLLN.

c. We have i.i.d. random variables X1 . . . Xn with an unknown distribution, and with µ = E[Xi ].
We define Mn = (X1 + . . . + Xn )/n. Consider the following statements:

(A) Mn is a maximum-likelihood estimator for µ, irrespective of the distribution of the Xi ’s.

(B) Mn is a consistent estimator for µ, irrespective of the distribution of the Xi ’s.
(C) Mn is an asymptotically unbiased estimator for µ, irrespective of the distribution of the
Xi ’s.

Which of these statements are always true? Write True or False in each of the boxes below.

A: False B: True C: True

Solution:

(A) False. Consider Xi follow a uniform distribution U [µ − 12 , µ + 12 ]. The ML estimator for µ

is any value between max(X1 , · · · , Xn ) − 12 and min(X1 , · · · , Xn ) + 12 , instead of Mn .
(B) True. By the WLLN, Mn converges to µ in probability and hence it is a consistent estimator.
(C) True. Since E[Mn ] = E[Xi ] = µ, Mn is unbiased estimator for µ and hence asymptotically
unbiased.

Page 4 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

Problem 2: Multiple Choice (4 pts. each, 24 pts. total)

Clearly circle the appropriate choice. No partial credit will be given for individual questions in this
part of the quiz.

a. Earthquakes in Sumatra occur according to a Poisson process of rate λ = 2/year. Conditioned

on the event that exactly two earthquakes take place in a year, what is the probability that both
earthquakes occur in the first three months of the year? (for simplicity, assume all months have
30 days, and each year has 12 months, i.e., 360 days).

(i) 1/12
(ii) 1/16
(iii) 64/225
(iv) 4e−4
(v) There is not enough information to determine the required probability.
(vi) None of the above.

Solution: Consider the interval of a year be [0, 1].

P 2 in [0, 14 ) , 0 in [ 14 , 1]
 

1
P 2 in [0, ) | 2 in [0, 1] =
4 P(2 in [0, 1]))
(λ·1/4)2 −λ·1/4 (λ·3/4)0 −λ·3/4
2! e · 0! e
= λ2 −λ
2! e
1
=
16

(alternative explanation) Given that exactly two earthquakes happened in 12 months, each earth­
quake is equally likely to happen in any month of the 12, the probability that it happens in the
first 3 months is 3/12 = 1/4. The probability that both happen in the first 3 months is (1/4)2 .

b. Consider a continuous-time Markov chain with three states i ∈ {1, 2, 3}, with dwelling time in
each visit to state i being an exponential random variable with parameter νi = i, and transition
probabilities pij defined by the graph

What is the long-term expected fraction of time spent in state 2?

(i) 1/2
(ii) 1/4
(iii) 2/5

Page 5 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

(iv) 3/7
(v) None of the above.

Solution: First, we calculate the qij = νi pij , i.e., q12 = q21 = q23 = 1 and q32 = 3. The balance
and normalization equations of this birth-death markov chain can be expressed as, π1 = π2 , π2 =
3π3 and π1 + π2 + π3 = 1, yielding π2 = 3/7.

Page 6 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

c. Consider the following Markov chain:

Starting in state 3, what is the steady-state probability of being in state 1?

(i) 1/3
(ii) 1/4
(iii) 1
(iv) 0
(v) None of the above.

Solution: State 1 is transient.

d. Random variables X and Y are such that the pair (X, Y ) is uniformly distributed over the
trapezoid A with corners (0, 0), (1, 2), (3, 2), and (4, 0) shown in Fig. 1:
X

Y
1 3 4

Figure 1: fX,Y (x, y) is constant over the shaded area, zero otherwise.

i.e. (
c, (x, y) ∈ A
fX,Y (x, y) =
0, else .

We observe Y and use it to estimate X. Let X̂ be the least mean squared error estimator of X
given Y . What is the value of var(X̂ − X|Y = 1)?

(i) 1/6
(ii) 3/2
(iii) 1/3
(iv) The information is not sufficient to compute this value.

Page 7 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

(v) None of the above.

Solution: fX|Y =1 (x) is uniform on [0, 2] therefore X̂ = E[X |Y = 1] = 1 and var(X̂ − X |Y =

1) = var(X|Y = 1) = (2 − 0)2 /12 = 1/3.

e. X1 . . . Xn are i.i.d. normal random variables with mean value µ and variance v. Both µ and v
are unknown. We define Mn = (X1 + . . . + Xn )/n and
n
1 X
Vn = (Xi − Mn )2
n−1
i=1

We also define Φ(x) to be the CDF for the standard normal distribution, and Ψn−1 (x) to be the
CDF for the t-distribution with n − 1 degrees of freedom. Which of the following choices gives
an exact 99% confidence interval for µ for all n > 1?
q q
(i) [Mn − δ n , Mn + δ Vnn ] where δ is chosen to give Φ(δ) = 0.99.
Vn

q q
(ii) [Mn − δ Vnn , Mn + δ Vnn ] where δ is chosen to give Φ(δ) = 0.995.
q q
(iii) [Mn − δ Vnn , Mn + δ Vnn ] where δ is chosen to give Ψn−1 (δ) = 0.99.
q q
(iv) [Mn − δ Vnn , Mn + δ Vnn ] where δ is chosen to give Ψn−1 (δ) = 0.995.

(v) None of the above.

Solution: See Lecture 23, slides 10-12.

f. We have i.i.d. random variables X1 , X2 which have an exponential distribution with unknown
parameter θ. Under hypothesis H0 , θ = 1. Under hypothesis H1 , θ = 2. Under a likelihood-ratio
test, the rejection region takes which of the following forms?

(i) R = {(x1 , x2 ) : x1 + x2 > ξ} for some value ξ.

(ii) R = {(x1 , x2 ) : x1 + x2 < ξ} for some value ξ.
(iii) R = {(x1 , x2 ) : ex1 + ex2 > ξ} for some value ξ.
(iv) R = {(x1 , x2 ) : ex1 + ex2 < ξ} for some value ξ.
(v) None of the above.

Solution: We defined R = {x = (x1 , x2 )|L(x) > c} where

fX (x; H1 ) θ1 e−θ1 x1 θ1 e−θ1 x2 θ12 (θ0 −θ1 )(x1 +x2 )

L(x) = = = e = 4e−(x1 +x2 ) .
fX (x; H0 ) θ0 e−θ0 x1 θ0 e−θ0 x2 θ02

So R = {(x1 , x2 )|x1 + x2 < − log (c/4)}

Page 8 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

Problem 3 (12 pts. total)

Aliens of two races (blue and green) are arriving on Earth independently according to Poisson process
distributions with parameters λb and λg respectively. The Alien Arrival Registration Service Authority
(AARSA) will begin registering alien arrivals soon.
Let T1 denote the time AARSA will function until it registers its first alien. Let G be the event
that the first alien to be registered is a green one. Let T2 be the time AARSA will function until at
least one alien of both races is registered.

(a) (4 points.) Express µ1 = E[T1 ] in terms of λg and λb . Show your work.

1
Answer: µ1 = E[T1 ] = λg +λb

Solution: We consider the process of arrivals of both types of Aliens. This is a merged Poisson
process with arrival rate λg +λb . T1 is the time until the first arrival, and therefore is exponentially
1
distributed with parameter λg + λb . Therefore µ1 = E[T1 ] = λg + λb .

One can also go about this using derived distributions, since T1 = min(T1g , T1b ) where T1g and T1b
are the first arrival times of green and blue Aliens respectively (i.e., T1g and T1b are exponentially
distributed with parameters λg and λb , respectively. )

(b) (4 points.) Express p = P(G) in terms of λg and λb . Show your work.

λg
Answer: P(G) = λg +λb

Solution: We consider the same merged Poisson process as before, with arrival rate λg +λb . Any
λ
particular arrival of the merged process has probability λg +gλb of corresponding to a green Alien
λb λg
and probability λg +λb of corresponding to a blue Alien. The question asks for P(G) = λg +λb .

Page 9 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

(c) (4 points.) Express µ2 = E[T2 ] in terms of λg and λb .

Show your work.

   
1 λg 1 λb 1
Answer: + +
λg + λb λg + λb λb λg + λb λg
Solution: The time T2 until at least one green and one red Aliens have arrived can be expressed as
T2 = max(T1g , T1b ), where T1g and T1b are the first arrival times of green and blue Aliens respectively
(i.e., T1g and T1b are exponentially distributed with parameters λg and λb , respectively.)
1
The expected time till the 1st Alien arrives was calculated in (a), µ1 = E[T1 ] = λg +λb
. To
compute the remaining time we simply condition on the 1st Alien being green(e.g. event G) or
blue(event Gc ), and use the memoryless property of Poisson, i.e.,

E[T2 ] = E[T1 ] + P(G)E[Time until first Blue arrives|G] + P(Gc )E[Time until first Green arrives|Gc ]
= E[T1 ] + P(G)E[T2b ] + (1 − P(G))E[T2g ]
   
1 λg 1 λb 1
= + +
λg + λb λg + λb λb λg + λb λg

Page 10 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

Problem 4 (18 pts. total)

Researcher Jill is interested in studying employment in technology firms in Dilicon Valley. She denotes
by Xi the number of employees in technology firm i and assumes that Xi are independent and identically
distributed with mean p. To estimate p, Jill randomly interviews n technology firms and observes the
number of employees in these firms.

(a) (6 points.) Jill uses

X1 + · · · + Xn
Mn =
n
as an estimator for p. Find the limit of P(Mn ≤ x) as n → ∞ for x < p. Find the limit of
P(Mn ≤ x) as n → ∞ for x > p. Show your work.

Solution: Since Xi is i.i.d., Mn converges to p in probability, i.e., limn→∞ P(|Mn − p| > ǫ) = 0,

implying limn→∞ P(Mn < p − ǫ) = 0 and limn→∞ P(Mn > p + ǫ) = 0, for all ǫ > 0. Hence

0, x < p ,
lim P(Mn ≤ x) =
n→∞ 1, x > p .

(b) (6 points.) Find the smallest n, the number of technology firms Jill must sample, for which the
Chebyshev inequality yields a guarantee

P(|Mn − p| ≥ 0.5) ≤ 0.05.

Assume that var (Xi ) = v for some constant v. State your solution as a function of v. Show
your work.

Solution: Since Mn converges to p in probability and var(Mn ) = nn2 · var(Xi ) = v/n, Chebyshev
inequality gives
var(Mn ) v
P (|Mn − p| ≥ 0.5) ≤ = = 0.05
0.52 n · 0.52
⇒ n = 80v.

Page 11 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

(c) (6 points.) Assume now that the researcher samples n = 5000 firms. Find an approximate value
for the probability
P(|M5000 − p| ≥ 0.5)
using the Central Limit Theorem. Assume again that var (Xi ) = v for some constant v. Give
your answer in terms of v, and the standard normal CDF Φ. Show your work.

Solution: By CLT, we can approximate by a standard normal distribution

Pn
i=1√Xi − np
nv

when n is large, and hence,

 Pn √   √ 
i=1√Xi− np 0.5 n 0.5 n
P (|M5000 − p| ≥ 0.5) = P ≥ √ = 2 − 2Φ √ ,
nv v v

where n = 5000.

Page 12 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

Problem 5 (12 pts. total)

The RandomView window factory produces window panes. After manufacturing, 1000 panes were
loaded onto a truck. The weight Wi of the i-th pane (in pounds) on the truck is modeled as a random
variable, with the assumption that the Wi ’s are independent and identically distributed.

(a) (6 points.) Assume that the measured weight of the load on the truck was 2340 pounds, and
that var (Wi ) ≤ 4. Find an approximate 95 percent confidence interval for µ = E[Wi ], using the
Central Limit Theorem (you may use the standard normal table which was handed out with this
quiz). Show your work.

 
Answer: 2.216, 2.464

W1 +···+Wn
Solution: The sample mean estimator Θ̂n = n in this case is
2340
Θ̂1000 = = 2.34
1000

Using the CDF Φ(z) of the standard normal available in the normal tables, we have Φ(1.96) =
0.975, so we obtain
|Θ̂1000 − µ|
P( p ≤ 1.96) ≈ 0.95.
var (Wi ) /1000
Because the variance is less than 4, we have
p p
P(|Θ̂1000 − µ| ≤ 1.96 var (Wi ) /1000) ≤ P(|Θ̂1000 − µ| ≤ 1.96 4/1000),

and letting the right-hand side of the above equation ≈ 0.95 gives a 95% confidence, i.e.,
r r
 4 4   
Θ̂1000 − 1.96 , Θ̂1000 + 1.96 = Θ̂1000 − 0.124, Θ̂1000 + 0.124
1000 1000

Page 13 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

(b) (6 points.) Now assume instead that the random variables Wi are i.i.d, with an exponential
distribution with parameter θ > 0, i.e., a distribution with PDF

fW (w; θ) = θe−θw

What is the maximum likelihood estimate of θ, given that the truckload has weight 2340 pounds?
Show your work.

1000
Answer: Θ̂mle
1000 = 2340 = 0.4274

Solution: The likelihood function is

n
Y n
Y
fW (w; θ) = fWi (wi ; θ) = θe−θwi ,
i=1 i=1

And the log-likelihood function is

n
X
log fW (w; θ) = n log θ − θ wi ,
i=1
Pn
The derivative with respect to θ is nθ − i=1 wi ,, and by setting it to zero, we see that the
maximum of log fW (w; θ) over θ ≥ 0 is attained at θ̂n = P nn wi . The resulting estimator is
i=1

n
Θ̂mle
n = Pn .
i=1 Wi

In our case,
1000
Θ̂mle
1000 = = 0.4274
2340

Page 14 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

Problem 6 (21 pts. total)

In Alice’s Wonderland, there are six different seasons: Fall (F), Winter (W), Spring (Sp), Summer (Su),
Bitter Cold (B), and Golden Sunshine (G). The seasons do not follow any particular order, instead, at
the beginning of each day the Head Wizard assigns the season for the day, according to the following
Markov chain model:

Thus, for example, if it is Fall one day then there is 1/6 probability that it will be Winter the next
day (note that it is possible to have the same season again the next day).

(a) (4 points.) For each state in the above chain, identify whether it is recurrent or transient. Show
your work.

Solution: F and W are transient states; Sp, Su, B, and G are recurrent states.

(b) (4 points.) If it is Fall on Monday, what is the probability that it will be Summer on Thursday
of the same week? Show your work.

Solution: There is only one path from F to Su in three days.

P(S4 = Su|S1 = F) = P(S2 = W|S1 = F) · P(S3 = Sp|S2 = W) · P(S4 = Su|S3 = Sp)

1 1 1
= · ·1=
6 5 30

Page 15 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

(c) (4 points.) If it is Spring today, will the chain converge to steady-state probabilities? If so,
compute the steady-state probability for each state. If not, explain why these probabilities do
not exist. Show your work.

Solution: The Markov chain will stay in the recurrent class {Sp, Su, B}, and

πSp · 1 = πSu · 1

 π · 1 = π · 12




 B 4 Su 4
πF = 0

 πW = 0

π =0

 G



πF + πW + πG + πSp + πSu + πB = 1

⇒ πF = 0, πW = 0, πG = 0, πSp = 1/5, πSu = 2/5, πB = 2/5.

(d) (5 points.) If it is Fall today, what is the probability that Bitter Cold will never arrive in the
future? Show your work.

Solution: Let aF and aW be the probabilities that Bitter Cold will never arrive starting from
Fall and Winter, respectively. This is equivalent to the Markov chain ends up in G.

aF = 23 · aF + 16 · aW + 16 · 1


aW = 15 · aF + 12 · aW + 101
·1

⇒ aF = 3/4.

Page 16 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

(e) (5 points.) If it is Fall today, what is the expected number of days till either Summer or Golden
Sunshine arrives for the first time? Show your work.

Solution: Let µF and µW be expected number of days till either Summer or Golden Sunshine
arrives for the first time, respectively.

µF = 1 + 23 · µF + 16 · µW + 16 · 0


µW = 1 + 15 · µF + 12 · µW + 15 · 1

⇒ µF = 5.25.

Page 17 of 18
 Massachusetts Institute of Technology
Department of Electrical Engineering & Computer Science
6.041/6.431: Probabilistic Systems Analysis
(Spring 2009)

Problem 7 (12 pts. total)

A newscast covering the final baseball game between Sed Rox and Y Nakee becomes noisy at the
crucial moment when the viewers are informed whether Y Nakee won the game.
Let a be the parameter describing the actual outcome: a = 1 if Y Nakee won, a = −1 otherwise.
There were n viewers listening to the telecast. Let Yi be the information received by viewer i (1 ≤ i ≤
n). Under the noisy telecast, Yi = a with probability p, and Yi = −a with probability 1 − p. Assume
that the random variables Yi are independent of each other.
The viewers as a group come up with a joint estimator
( Pn
1 if i=1 Yi ≥ 0,
Zn =
−1 otherwise.

(a) (6 points.) Find limn→∞ P(Zn = a) assuming that p > 0.5 and a = 1. Show your work.

Solution: Note that

n
!  Pn 
i=1 Yi
X
lim P(Zn = 1) = lim P Yi ≥ 0 = lim P ≥0 .
n→∞ n→∞ n→∞ n
i=1

Since Yi are i.i.d. with mean E[Yi ] = 2p − 1 and finite variance var(Yi ) = 1 − (2p − 1)2 , one has,
by Chebyshev inequality, for all ǫ > 0
 Pn 
i=1 Yi
lim P − (2p − 1) ≥ ǫ = 0.
n→∞ n
 Pn 
i=1 Yi
Take ǫ = p− 12 , and the above equation implies limn→∞ P n ≤ (2p − 1)/2 = 0. Therefore,
limn→∞ P(Zn = 1) = 1.

(b) (6 points.) Find limn→∞ P(Zn = a), assuming that p = 0.5 and a = 1. Show your work.

Solution: Note that

n
!  Pn 
X
i=1 Yi
lim P(Zn = 1) = lim P Yi ≥ 0 = lim P √ ≥0 .
n→∞ n→∞ n→∞ n
i=1
Pn
=1 Yi
Since Yi are i.i.d. with E[Yi ] = 0 and var(Yi ) = 1, we can approximate i√
n
as a standard
normal random variable when n goes to infinity. Thus, limn→∞ P(Zn = 1) = 1/2.

Page 18 of 18
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.