Sol Manual

Probability and Stochastic Processes
A Friendly Introduction for Electrical and Computer Engineers

SECOND EDITION
Problem Solutions
September 28, 2005 Draft
Roy D. Yates, David J. Goodman, David Famolari
September 28, 2005
• This solution manual remains under construction. The current count is that 678 (out of 687)
problems have solutions. The unsolved problems are
12.1.7, 12.1.8, 12.5.8, 12.5.9, 12.11.5 – 12.11.9.
If you volunteer a solution for one of those problems, we’ll be happy to include it . . . and, of
course, “your wildest dreams will come true.”
• Of course, the correctness of every single solution reamins unconfirmed. If you find errors or
have suggestions or comments, please send email: ryates@winlab.rutgers.edu.
• If you need to make solution sets for your class, you might like the Solution Set Constructor
at the instructors site www.winlab.rutgers.edu/probsolns. If you need access, send email:
ryates@winlab.rutgers.edu.
• Matlab functions written as solutions to homework problems can be found in the archive
matsoln.zip (available to instructors) or in the directory matsoln. Other Matlab functions
used in the text or in these homework solutions can be found in the archive matcode.zip
or directory matcode. The .m files in matcode are available for download from the Wiley
website. Two other documents of interest are also available for download:
– A manual probmatlab.pdf describing the matcode .m functions is also available.

– The quiz solutions manual quizsol.pdf.
• A web-based solution set constructor for the second edition is available to instructors at
http://www.winlab.rutgers.edu/probsolns
• The next update of this solution manual is likely to occur in January, 2006.
1
Problem Solutions – Chapter 1
Problem 1.1.1 Solution

Based on the Venn diagram
M O
the answers are fairly straightforward:

(a) Since T ∩ M = φ, T and M are not mutually exclusive.
(b) Every pizza is either Regular (R), or Tuscan (T ). Hence R ∪ T = S so that R and T are
collectively exhaustive. Thus its also (trivially) true that R ∪ T ∪ M = S. That is, R, T and
M are also collectively exhaustive.
(c) From the Venn diagram, T and O are mutually exclusive. In words, this means that Tuscan
pizzas never have onions or pizzas with onions are never Tuscan. As an aside, “Tuscan” is
a fake pizza designation; one shouldn’t conclude that people from Tuscany actually dislike
onions.
(d) From the Venn diagram, M ∩ T and O are mutually exclusive. Thus Gerlanda’s doesn’t make
Tuscan pizza with mushrooms and onions.
(e) Yes. In terms of the Venn diagram, these pizzas are in the set (T ∪ M ∪ O)c .

Based on the Venn diagram,
M O
the complete Gerlandas pizza menu is

• Regular without toppings
• Regular with mushrooms
• Regular with onions
• Regular with mushrooms and onions
• Tuscan without toppings
• Tuscan with mushrooms
(a) An outcome specifies whether the fax is high (h), medium (m), or low (l) speed, and whether
the fax has two (t) pages or four (f ) pages. The sample space is
S = {ht, hf, mt, mf, lt, lf } . (1)
2
(b) The event that the fax is medium speed is A1 = {mt, mf }.
(c) The event that a fax has two pages is A2 = {ht, mt, lt}.
(d) The event that a fax is either high speed or low speed is A3 = {ht, hf, lt, lf }.
(e) Since A1 ∩ A2 = {mt} and is not empty, A1 , A2 , and A3 are not mutually exclusive.
(f) Since
A1 ∪ A2 ∪ A3 = {ht, hf, mt, mf, lt, lf } = S, (2)
the collection A1 , A2 , A3 is collectively exhaustive.
(a) The sample space of the experiment is
S = {aaa, aaf, af a, f aa, f f a, f af, af f, f f f} . (1)
(b) The event that the circuit from Z fails is
ZF = {aaf, af f, f af, f f f } . (2)
The event that the circuit from X is acceptable is
XA = {aaa, aaf, af a, af f } . (3)
(c) Since ZF ∩ XA = {aaf, af f } = φ, ZF and XA are not mutually exclusive.
(d) Since ZF ∪ XA = {aaa, aaf, af a, af f, f af, f f f} = S, ZF and XA are not collectively exhaus-
tive.
(e) The event that more than one circuit is acceptable is
C = {aaa, aaf, af a, f aa} . (4)
The event that at least two circuits fail is
D = {f f a, f af, af f, f f f } . (5)
(f) Inspection shows that C ∩ D = φ so C and D are mutually exclusive.
(g) Since C ∪ D = S, C and D are collectively exhaustive.

The sample space is
S = {A♣, . . . , K♣, A♦, . . . , K♦, A♥, . . . , K♥, A♠, . . . , K♠} . (1)
The event H is the set

H = {A♥, . . . , K♥} . (2)
3
The sample space is
⎧ ⎫
⎨ 1/1 . . . 1/31, 2/1 . . . 2/29, 3/1 . . . 3/31, 4/1 . . . 4/30, ⎬
S= 5/1 . . . 5/31, 6/1 . . . 6/30, 7/1 . . . 7/31, 8/1 . . . 8/31, . (1)
⎩ ⎭
9/1 . . . 9/31, 10/1 . . . 10/31, 11/1 . . . 11/30, 12/1 . . . 12/31
The event H defined by the event of a July birthday is described by following 31 sample points.
H = {7/1, 7/2, . . . , 7/31} . (2)

Of course, there are many answers to this problem. Here are four event spaces.
1. We can divide students into engineers or non-engineers. Let A1 equal the set of engineering
students and A2 the non-engineers. The pair {A1 , A2 } is an event space.
2. We can also separate students by GPA. Let Bi denote the subset of students with GPAs G
satisfying i − 1 ≤ G < i. At Rutgers, {B1 , B2 , . . . , B5 } is an event space. Note that B5 is
the set of all students with perfect 4.0 GPAs. Of course, other schools use different scales for
GPA.
3. We can also divide the students by age. Let Ci denote the subset of students of age i in years.
At most universities, {C10 , C11 , . . . , C100 } would be an event space. Since a university may
have prodigies either under 10 or over 100, we note that {C0 , C1 , . . .} is always an event space
4. Lastly, we can categorize students by attendance. Let D0 denote the number of students who
have missed zero lectures and let D1 denote all other students. Although it is likely that D0
is an empty set, {D0 , D1 } is a well defined event space.

Let R1 and R2 denote the measured resistances. The pair (R1 , R2 ) is an outcome of the experiment.
Some event spaces include
1. If we need to check that neither resistance is too high, an event space is
A1 = {R1 < 100, R2 < 100} , A2 = {either R1 ≥ 100 or R2 ≥ 100} . (1)
2. If we need to check whether the first resistance exceeds the second resistance, an event space
is
B1 = {R1 > R2 } B2 = {R1 ≤ R2 } . (2)
3. If we need to check whether each resistance doesn’t fall below a minimum value (in this case
50 ohms for R1 and 100 ohms for R2 ), an event space is
C1 = {R1 < 50, R2 < 100} , C2 = {R1 < 50, R2 ≥ 100} , (3)
C3 = {R1 ≥ 50, R2 < 100} , C4 = {R1 ≥ 50, R2 ≥ 100} . (4)
4. If we want to check whether the resistors in parallel are within an acceptable range of 90 to
110 ohms, an event space is

D1 = (1/R1 + 1/R2 )−1 < 90 , (5)
−1

D2 = 90 ≤ (1/R1 + 1/R2 ) ≤ 110 , (6)
−1

D2 = 110 < (1/R1 + 1/R2 ) . (7)
4
The sample space of the experiment is
S = {LF, BF, LW, BW } . (1)
From the problem statement, we know that P [LF ] = 0.5, P [BF ] = 0.2 and P [BW ] = 0.2. This
implies P [LW ] = 1 − 0.5 − 0.2 − 0.2 = 0.1. The questions can be answered using Theorem 1.5.
(a) The probability that a program is slow is
P [W ] = P [LW ] + P [BW ] = 0.1 + 0.2 = 0.3. (2)
(b) The probability that a program is big is
P [B] = P [BF ] + P [BW ] = 0.2 + 0.2 = 0.4. (3)
(c) The probability that a program is slow or big is
P [W ∪ B] = P [W ] + P [B] − P [BW ] = 0.3 + 0.4 − 0.2 = 0.5. (4)

A sample outcome indicates whether the cell phone is handheld (H) or mobile (M ) and whether
the speed is fast (F ) or slow (W ). The sample space is
S = {HF, HW, M F, M W } . (1)
The problem statement tells us that P [HF ] = 0.2, P [M W ] = 0.1 and P [F ] = 0.5. We can use
these facts to find the probabilities of the other outcomes. In particular,
P [F ] = P [HF ] + P [M F ] . (2)
This implies
P [M F ] = P [F ] − P [HF ] = 0.5 − 0.2 = 0.3. (3)
Also, since the probabilities must sum to 1,
P [HW ] = 1 − P [HF ] − P [M F ] − P [M W ] = 1 − 0.2 − 0.3 − 0.1 = 0.4. (4)
Now that we have found the probabilities of the outcomes, finding any other probability is easy.
(a) The probability a cell phone is slow is
P [W ] = P [HW ] + P [M W ] = 0.4 + 0.1 = 0.5. (5)
(b) The probability that a cell hpone is mobile and fast is P [M F ] = 0.3.
(c) The probability that a cell phone is handheld is
P [H] = P [HF ] + P [HW ] = 0.2 + 0.4 = 0.6. (6)
5
A reasonable probability model that is consistent with the notion of a shuffled deck is that each
card in the deck is equally likely to be the first card. Let Hi denote the event that the first card
drawn is the ith heart where the first heart is the ace, the second heart is the deuce and so on. In
that case, P [Hi ] = 1/52 for 1 ≤ i ≤ 13. The event H that the first card is a heart can be written
as the disjoint union
H = H1 ∪ H2 ∪ · · · ∪ H13 . (1)
Using Theorem 1.1, we have

13
P [H] = P [Hi ] = 13/52. (2)
i=1
This is the answer you would expect since 13 out of 52 cards are hearts. The point to keep in
mind is that this is not just the common sense answer but is the result of a probability model for
a shuffled deck and the axioms of probability.

Let si denote the outcome that the down face has i dots. The sample space is S = {s1 , . . . , s6 }.
The probability of each sample outcome is P [si ] = 1/6. From Theorem 1.1, the probability of the
event E that the roll is even is
P [E] = P [s2 ] + P [s4 ] + P [s6 ] = 3/6. (1)

Let si equal the outcome of the student’s quiz. The sample space is then composed of all the
possible grades that she can receive.
S = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10} . (1)
Since each of the 11 possible outcomes is equally likely, the probability of receiving a grade of i, for
each i = 0, 1, . . . , 10 is P [si ] = 1/11. The probability that the student gets an A is the probability
that she gets a score of 9 or higher. That is
P [Grade of A] = P [9] + P [10] = 1/11 + 1/11 = 2/11. (2)
The probability of failing requires the student to get a grade less than 4.
P [Failing] = P [3] + P [2] + P [1] + P [0] = 1/11 + 1/11 + 1/11 + 1/11 = 4/11. (3)

From the table we look to add all the disjoint events that contain H0 to express the probability
that a caller makes no hand-offs as
P [H0 ] = P [LH0 ] + P [BH0 ] = 0.1 + 0.4 = 0.5. (1)
In a similar fashion we can express the probability that a call is brief by
P [B] = P [BH0 ] + P [BH1 ] + P [BH2 ] = 0.4 + 0.1 + 0.1 = 0.6. (2)
The probability that a call is long or makes at least two hand-offs is
P [L ∪ H2 ] = P [LH0 ] + P [LH1 ] + P [LH2 ] + P [BH2 ] (3)
= 0.1 + 0.1 + 0.2 + 0.1 = 0.5. (4)
6
(a) From the given probability distribution of billed minutes, M , the probability that a call is
billed for more than 3 minutes is
P [L] = 1 − P [3 or fewer billed minutes] (1)

= 1 − P [B1 ] − P [B2 ] − P [B3 ] (2)
= 1 − α − α(1 − α) − α(1 − α) 2
(3)
= (1 − α)3 = 0.57. (4)
(b) The probability that a call will billed for 9 minutes or less is

9
P [9 minutes or less] = α(1 − α)i−1 = 1 − (0.57)3 . (5)
i=1

The first generation consists of two plants each with genotype yg or gy. They are crossed to produce
the following second generation genotypes, S = {yy, yg, gy, gg}. Each genotype is just as likely as
any other so the probability of each genotype is consequently 1/4. A pea plant has yellow seeds if
it possesses at least one dominant y gene. The set of pea plants with yellow seeds is
Y = {yy, yg, gy} . (1)
So the probability of a pea plant with yellow seeds is
P [Y ] = P [yy] + P [yg] + P [gy] = 3/4. (2)

Each statement is a consequence of part 4 of Theorem 1.4.
(a) Since A ⊂ A ∪ B, P [A] ≤ P [A ∪ B].
(b) Since B ⊂ A ∪ B, P [B] ≤ P [A ∪ B].
(c) Since A ∩ B ⊂ A, P [A ∩ B] ≤ P [A].
(d) Since A ∩ B ⊂ B, P [A ∩ B] ≤ P [B].

Specifically, we will use Theorem 1.7(c) which states that for any events A and B,
P [A ∪ B] = P [A] + P [B] − P [A ∩ B] . (1)
To prove the union bound by induction, we first prove the theorem for the case of n = 2 events. In
this case, by Theorem 1.7(c),
P [A1 ∪ A2 ] = P [A1 ] + P [A2 ] − P [A1 ∩ A2 ] . (2)
7
By the first axiom of probability, P [A1 ∩ A2 ] ≥ 0. Thus,
P [A1 ∪ A2 ] ≤ P [A1 ] + P [A2 ] . (3)
which proves the union bound for the case n = 2. Now we make our induction hypothesis that the
union-bound holds for any collection of n − 1 subsets. In this case, given subsets A1 , . . . , An , we
define
A = A1 ∪ A2 ∪ · · · ∪ An−1 , B = An . (4)
By our induction hypothesis,
P [A] = P [A1 ∪ A2 ∪ · · · ∪ An−1 ] ≤ P [A1 ] + · · · + P [An−1 ] . (5)
This permits us to write
P [A1 ∪ · · · ∪ An ] = P [A ∪ B] (6)
≤ P [A] + P [B] (by the union bound for n = 2) (7)
= P [A1 ∪ · · · ∪ An−1 ] + P [An ] (8)
≤ P [A1 ] + · · · P [An−1 ] + P [An ] (9)
which completes the inductive proof.
(a) For convenience, let pi = P [F Hi ] and qi = P [V Hi ]. Using this shorthand, the six unknowns
p0 , p1 , p2 , q0 , q1 , q2 fill the table as
H0 H1 H2
F p0 p1 p2 . (1)
V q0 q1 q2
However, we are given a number of facts:
p0 + q0 = 1/3, p1 + q1 = 1/3, (2)

p2 + q2 = 1/3, p0 + p1 + p2 = 5/12. (3)
Other facts, such as q0 + q1 + q2 = 7/12, can be derived from these facts. Thus, we have
four equations and six unknowns, choosing p0 and p1 will specify the other unknowns. Un-
fortunately, arbitrary choices for either p0 or p1 will lead to negative values for the other
probabilities. In terms of p0 and p1 , the other unknowns are
q0 = 1/3 − p0 , p2 = 5/12 − (p0 + p1 ), (4)

q1 = 1/3 − p1 , q2 = p0 + p1 − 1/12. (5)
Because the probabilities must be nonnegative, we see that
0 ≤ p0 ≤ 1/3, (6)
0 ≤ p1 ≤ 1/3, (7)
1/12 ≤ p0 + p1 ≤ 5/12. (8)
8
Although there are an infinite number of solutions, three possible solutions are:
p0 = 1/3, p1 = 1/12, p2 = 0, (9)

q0 = 0, q1 = 1/4, q2 = 1/3. (10)
and
p0 = 1/4, p1 = 1/12, p2 = 1/12, (11)

q0 = 1/12, q1 = 3/12, q2 = 3/12. (12)
and
p0 = 0, p1 = 1/12, p2 = 1/3, (13)

q0 = 1/3, q1 = 3/12, q2 = 0. (14)
(b) In terms of the pi , qi notation, the new facts are p0 = 1/4 and q1 = 1/6. These extra facts
uniquely specify the probabilities. In this case,
p0 = 1/4, p1 = 1/6, p2 = 0, (15)

q0 = 1/12, q1 = 1/6, q2 = 1/3. (16)

It is tempting to use the following proof:
Since S and φ are mutually exclusive, and since S = S ∪ φ,
1 = P [S ∪ φ] = P [S] + P [φ] . (1)
Since P [S] = 1, we must have P [φ] = 0.
The above “proof” used the property that for mutually exclusive sets A1 and A2 ,
P [A1 ∪ A2 ] = P [A1 ] + P [A2 ] . (2)
The problem is that this property is a consequence of the three axioms, and thus must be proven.
For a proof that uses just the three axioms, let A1 be an arbitrary set and for n = 2, 3, . . ., let
An = φ. Since A1 = ∪∞i=1 Ai , we can use Axiom 3 to write
∞

P [A1 ] = P [∪∞
i=1 Ai ] = P [A1 ] + P [A2 ] + P [Ai ] . (3)
i=3
By subtracting P [A1 ] from both sides, the fact that A2 = φ permits us to write
∞

P [φ] + P [Ai ] = 0. (4)
n=3

By Axiom 1, P [Ai ] ≥ 0 for all i. Thus, ∞ n=3 P [Ai ] ≥ 0. This implies P [φ] ≤ 0. Since Axiom 1
requires P [φ] ≥ 0, we must have P [φ] = 0.
9
Following the hint, we define the set of events {Ai |i = 1, 2, . . .} such that i = 1, . . . , m, Ai = Bi and
for i > m, Ai = φ. By construction, ∪m ∞
i=1 Bi = ∪i=1 Ai . Axiom 3 then implies
∞

∞
P [∪m
i=1 Bi ] = P [∪i=1 Ai ] = P [Ai ] . (1)
i=1

m
m
For i > m, P [Ai ] = P [φ] = 0, yielding the claim P [∪m
i=1 Bi ] = i=1 P [Ai ] = i=1 P [Bi ].
Note that the fact that P [φ] = 0 follows from Axioms 1 and 2. This problem is more challenging
if you just use Axiom 3. We start by observing

m−1 ∞

P [∪m
i=1 Bi ] = P [Bi ] + P [Ai ] . (2)
i=1 i=m
Now, we use Axiom 3 again on the countably infinite sequence Am , Am+1 , . . . to write
∞

P [Ai ] = P [Am ∪ Am+1 ∪ · · ·] = P [Bm ] . (3)
i=m

m
Thus, we have used just Axiom 3 to prove Theorem 1.4: P [∪m
i=1 Bi ] = i=1 P [Bi ].

Each claim in Theorem 1.7 requires a proof from which we can check which axioms are used.
However, the problem is somewhat hard because there may still be a simpler proof that uses fewer
axioms. Still, the proof of each part will need Theorem 1.4 which we now prove.
For the mutually exclusive events B1 , . . . , Bm , let Ai = Bi for i = 1, . . . , m and let Ai = φ for
i > m. In that case, by Axiom 3,
P [B1 ∪ B2 ∪ · · · ∪ Bm ] = P [A1 ∪ A2 ∪ · · ·] (1)

m−1 ∞

= P [Ai ] + P [Ai ] (2)
i=1 i=m

m−1 ∞
= P [Bi ] + P [Ai ] . (3)
i=1 i=m
Now, we use Axiom 3 again on Am , Am+1 , . . . to write

∞

P [Ai ] = P [Am ∪ Am+1 ∪ · · ·] = P [Bm ] . (4)
i=m
Thus, we have used just Axiom 3 to prove Theorem 1.4:

m
P [B1 ∪ B2 ∪ · · · ∪ Bm ] = P [Bi ] . (5)
i=1
(a) To show P [φ] = 0, let B1 = S and let B2 = φ. Thus by Theorem 1.4,
P [S] = P [B1 ∪ B2 ] = P [B1 ] + P [B2 ] = P [S] + P [φ] . (6)
Thus, P [φ] = 0. Note that this proof uses only Theorem 1.4 which uses only Axiom 3.
10
(b) Using Theorem 1.4 with B1 = A and B2 = Ac , we have
P [S] = P [A ∪ Ac ] = P [A] + P [Ac ] . (7)
Since, Axiom 2 says P [S] = 1, P [Ac ] = 1 − P [A]. This proof uses Axioms 2 and 3.
(c) By Theorem 1.2, we can write both A and B as unions of disjoint events:
A = (AB) ∪ (AB c ) B = (AB) ∪ (Ac B). (8)
Now we apply Theorem 1.4 to write
P [A] = P [AB] + P [AB c ] , P [B] = P [AB] + P [Ac B] . (9)
We can rewrite these facts as
P [AB c ] = P [A] − P [AB], P [Ac B] = P [B] − P [AB]. (10)
Note that so far we have used only Axiom 3. Finally, we observe that A ∪ B can be written
as the union of mutually exclusive events
A ∪ B = (AB) ∪ (AB c ) ∪ (Ac B). (11)
Once again, using Theorem 1.4, we have
P [A ∪ B] = P [AB] + P [AB c ] + P [Ac B] (12)
Substituting the results of Equation (10) into Equation (12) yields
P [A ∪ B] = P [AB] + P [A] − P [AB] + P [B] − P [AB] , (13)
which completes the proof. Note that this claim required only Axiom 3.
(d) Observe that since A ⊂ B, we can write B as the disjoint union B = A ∪ (Ac B). By
Theorem 1.4 (which uses Axiom 3),
P [B] = P [A] + P [Ac B] . (14)
By Axiom 1, P [Ac B] ≥ 0, hich implies P [A] ≤ P [B]. This proof uses Axioms 1 and 3.

Each question requests a conditional probability.
(a) Note that the probability a call is brief is
P [B] = P [H0 B] + P [H1 B] + P [H2 B] = 0.6. (1)
The probability a brief call will have no handoffs is
P [H0 B] 0.4 2
P [H0 |B] = = = . (2)
P [B] 0.6 3
(b) The probability of one handoff is P [H1 ] = P [H1 B] + P [H1 L] = 0.2. The probability that a
call with one handoff will be long is
P [H1 L] 0.1 1
P [L|H1 ] = = = . (3)
P [H1 ] 0.2 2
(c) The probability a call is long is P [L] = 1 − P [B] = 0.4. The probability that a long call will
have one or more handoffs is
P [H1 L ∪ H2 L] P [H1 L] + P [H2 L] 0.1 + 0.2 3
P [H1 ∪ H2 |L] = = = = . (4)
P [L] P [L] 0.4 4
11
Let si denote the outcome that the roll is i. So, for 1 ≤ i ≤ 6, Ri = {si }. Similarly, Gj =
{sj+1 , . . . , s6 }.
(a) Since G1 = {s2 , s3 , s4 , s5 , s6 } and all outcomes have probability 1/6, P [G1 ] = 5/6. The event
R3 G1 = {s3 } and P [R3 G1 ] = 1/6 so that
P [R3 G1 ] 1
P [R3 |G1 ] = = . (1)
P [G1 ] 5
(b) The conditional probability that 6 is rolled given that the roll is greater than 3 is
P [R6 G3 ] P [s6 ] 1/6
P [R6 |G3 ] = = = . (2)
P [G3 ] P [s4 , s5 , s6 ] 3/6
(c) The event E that the roll is even is E = {s2 , s4 , s6 } and has probability 3/6. The joint
probability of G3 and E is
P [G3 E] = P [s4 , s6 ] = 1/3. (3)
The conditional probabilities of G3 given E is
P [G3 E] 1/3 2
P [G3 |E] = = = . (4)
P [E] 1/2 3
(d) The conditional probability that the roll is even given that it’s greater than 3 is
P [EG3 ] 1/3 2
P [E|G3 ] = = = . (5)
P [G3 ] 1/2 3

Since the 2 of clubs is an even numbered card, C2 ⊂ E so that P [C2 E] = P [C2 ] = 1/3. Since
P [E] = 2/3,
P [C2 E] 1/3
P [C2 |E] = = = 1/2. (1)
P [E] 2/3
The probability that an even numbered card is picked given that the 2 is picked is
P [C2 E] 1/3
P [E|C2 ] = = = 1. (2)
P [C2 ] 1/3

Define D as the event that a pea plant has two dominant y genes. To find the conditional probability
of D given the event Y , corresponding to a plant having yellow seeds, we look to evaluate
P [DY ]
P [D|Y ] = . (1)
P [Y ]
Note that P [DY ] is just the probability of the genotype yy. From Problem 1.4.3, we found that
with respect to the color of the peas, the genotypes yy, yg, gy, and gg were all equally likely. This
implies
P [DY ] = P [yy] = 1/4 P [Y ] = P [yy, gy, yg] = 3/4. (2)
Thus, the conditional probability can be expressed as
P [DY ] 1/4
P [D|Y ] = = = 1/3. (3)
P [Y ] 3/4
12
The sample outcomes can be written ijk where the first card drawn is i, the second is j and the
third is k. The sample space is
S = {234, 243, 324, 342, 423, 432} . (1)
and each of the six outcomes has probability 1/6. The events E1 , E2 , E3 , O1 , O2 , O3 are
E1 = {234, 243, 423, 432} , O1 = {324, 342} , (2)
E2 = {243, 324, 342, 423} , O2 = {234, 432} , (3)
E3 = {234, 324, 342, 432} , O3 = {243, 423} . (4)
(a) The conditional probability the second card is even given that the first card is even is
P [E2 E1 ] P [243, 423] 2/6
P [E2 |E1 ] = = = = 1/2. (5)
P [E1 ] P [234, 243, 423, 432] 4/6
(b) The conditional probability the first card is even given that the second card is even is
P [E1 E2 ] P [243, 423] 2/6
P [E1 |E2 ] = = = = 1/2. (6)
P [E2 ] P [243, 324, 342, 423] 4/6
(c) The probability the first two cards are even given the third card is even is
P [E1 E2 E3 ]
P [E1 E2 |E3 ] = = 0. (7)
P [E3 ]
(d) The conditional probabilities the second card is even given that the first card is odd is
P [O1 E2 ] P [O1 ]
P [E2 |O1 ] = = = 1. (8)
P [O1 ] P [O1 ]
(e) The conditional probability the second card is odd given that the first card is odd is
P [O1 O2 ]
P [O2 |O1 ] = = 0. (9)
P [O1 ]

The problem statement yields the obvious facts that P [L] = 0.16 and P [H] = 0.10. The words
“10% of the ticks that had either Lyme disease or HGE carried both diseases” can be written as
P [LH|L ∪ H] = 0.10. (1)
(a) Since LH ⊂ L ∪ H,
P [LH ∩ (L ∪ H)] P [LH]
P [LH|L ∪ H] = = = 0.10. (2)
P [L ∪ H] P [L ∪ H]
Thus,
P [LH] = 0.10P [L ∪ H] = 0.10 (P [L] + P [H] − P [LH]) . (3)
Since P [L] = 0.16 and P [H] = 0.10,
0.10 (0.16 + 0.10)
P [LH] = = 0.0236. (4)
1.1
13
(b) The conditional probability that a tick has HGE given that it has Lyme disease is
P [LH] 0.0236
P [H|L] = = = 0.1475. (5)
P [L] 0.16

This problems asks whether A and B can be independent events yet satisfy A = B? By definition,
events A and B are independent if and only if P [AB] = P [A]P [B]. We can see that if A = B, that
is they are the same set, then
P [AB] = P [AA] = P [A] = P [B] . (1)
Thus, for A and B to be the same set and also independent,
P [A] = P [AB] = P [A] P [B] = (P [A])2 . (2)
There are two ways that this requirement can be satisfied:
• P [A] = 1 implying A = B = S.
• P [A] = 0 implying A = B = φ.
In the Venn diagram, assume the sample space has area 1 correspond-
A ing to probability 1. As drawn, both A and B have area 1/4 so that
P [A] = P [B] = 1/4. Moreover, the intersection AB has area 1/16
and covers 1/4 of A and 1/4 of B. That is, A and B are independent
since
B P [AB] = P [A] P [B] . (1)
(a) Since A and B are disjoint, P [A ∩ B] = 0. Since P [A ∩ B] = 0,
P [A ∪ B] = P [A] + P [B] − P [A ∩ B] = 3/8. (1)
A Venn diagram should convince you that A ⊂ B c so that A ∩ B c = A. This implies
P [A ∩ B c ] = P [A] = 1/4. (2)
It also follows that P [A ∪ B c ] = P [B c ] = 1 − 1/8 = 7/8.
(b) Events A and B are dependent since P [AB] = P [A]P [B].
14
(c) Since C and D are independent,
P [C ∩ D] = P [C] P [D] = 15/64. (3)
The next few items are a little trickier. From Venn diagrams, we see
P [C ∩ Dc ] = P [C] − P [C ∩ D] = 5/8 − 15/64 = 25/64. (4)
It follows that
P [C ∪ Dc ] = P [C] + P [Dc ] − P [C ∩ Dc ] (5)
= 5/8 + (1 − 3/8) − 25/64 = 55/64. (6)
Using DeMorgan’s law, we have
P [C c ∩ Dc ] = P [(C ∪ D)c ] = 1 − P [C ∪ D] = 15/64. (7)
(d) Since P [C c Dc ] = P [C c ]P [Dc ], C c and Dc are independent.
(a) Since A ∩ B = ∅, P [A ∩ B] = 0. To find P [B], we can write

P [A ∪ B] = P [A] + P [B] − P [A ∩ B] (1)
5/8 = 3/8 + P [B] − 0. (2)
Thus, P [B] = 1/4. Since A is a subset of B c , P [A ∩ B c ] = P [A] = 3/8. Furthermore, since
A is a subset of B c , P [A ∪ B c ] = P [B c ] = 3/4.
(b) The events A and B are dependent because
P [AB] = 0 = 3/32 = P [A] P [B] . (3)
(c) Since C and D are independent P [CD] = P [C]P [D]. So

P [CD] 1/3
P [D] = = = 2/3. (4)
P [C] 1/2
In addition, P [C ∩ Dc ] = P [C] − P [C ∩ D] = 1/2 − 1/3 = 1/6. To find P [C c ∩ Dc ], we first
observe that
P [C ∪ D] = P [C] + P [D] − P [C ∩ D] = 1/2 + 2/3 − 1/3 = 5/6. (5)
By De Morgan’s Law, C c ∩ Dc = (C ∪ D)c . This implies
P [C c ∩ Dc ] = P [(C ∪ D)c ] = 1 − P [C ∪ D] = 1/6. (6)
Note that a second way to find P [C c ∩ Dc ] is to use the fact that if C and D are independent,
then C c and Dc are independent. Thus
P [C c ∩ Dc ] = P [C c ] P [Dc ] = (1 − P [C])(1 − P [D]) = 1/6. (7)
Finally, since C and D are independent events, P [C|D] = P [C] = 1/2.
(d) Note that we found P [C ∪ D] = 5/6. We can also use the earlier results to show
P [C ∪ Dc ] = P [C] + P [D] − P [C ∩ Dc ] = 1/2 + (1 − 2/3) − 1/6 = 2/3. (8)
(e) By Definition 1.7, events C and Dc are independent because

P [C ∩ Dc ] = 1/6 = (1/2)(1/3) = P [C] P [Dc ] . (9)
15
For a sample space S = {1, 2, 3, 4} with equiprobable outcomes, consider the events
A1 = {1, 2} A2 = {2, 3} A3 = {3, 1} . (1)
Each event Ai has probability 1/2. Moreover, each pair of events is independent since
P [A1 A2 ] = P [A2 A3 ] = P [A3 A1 ] = 1/4. (2)
However, the three events A1 , A2 , A3 are not independent since
P [A1 A2 A3 ] = 0 = P [A1 ] P [A2 ] P [A3 ] . (3)

There are 16 distinct equally likely outcomes for the second generation of pea plants based on a
first generation of {rwyg, rwgy, wryg, wrgy}. They are listed below
rryy rryg rrgy rrgg

rwyy rwyg rwgy rwgg
(1)
wryy wryg wrgy wrgg
wwyy wwyg wwgy wwgg
A plant has yellow seeds, that is event Y occurs, if a plant has at least one dominant y gene. Except
for the four outcomes with a pair of recessive g genes, the remaining 12 outcomes have yellow seeds.
From the above, we see that
P [Y ] = 12/16 = 3/4 (2)
and
P [R] = 12/16 = 3/4. (3)
To find the conditional probabilities P [R|Y ] and P [Y |R], we first must find P [RY ]. Note that
RY , the event that a plant has rounded yellow seeds, is the set of outcomes
RY = {rryy, rryg, rrgy, rwyy, rwyg, rwgy, wryy, wryg, wrgy} . (4)
Since P [RY ] = 9/16,

P [RY ] 9/16
P [Y |R ] = = = 3/4 (5)
P [R] 3/4
and
P [RY ] 9/16
P [R |Y ] = = = 3/4. (6)
P [Y ] 3/4
Thus P [R|Y ] = P [R] and P [Y |R] = P [Y ] and R and Y are independent events. There are four
visibly different pea plants, corresponding to whether the peas are round (R) or not (Rc ), or yellow
(Y ) or not (Y c ). These four visible events have probabilities
P [RY ] = 9/16 P [RY c ] = 3/16, (7)

c c c
P [R Y ] = 3/16 P [R Y ] = 1/16. (8)
16
(a) For any events A and B, we can write the law of total probability in the form of
P [A] = P [AB] + P [AB c ] . (1)
Since A and B are independent, P [AB] = P [A]P [B]. This implies
P [AB c ] = P [A] − P [A] P [B] = P [A] (1 − P [B]) = P [A] P [B c ] . (2)
Thus A and B c are independent.
(b) Proving that Ac and B are independent is not really necessary. Since A and B are arbitrary
labels, it is really the same claim as in part (a). That is, simply reversing the labels of A and
B proves the claim. Alternatively, one can construct exactly the same proof as in part (a)
with the labels A and B reversed.
(c) To prove that Ac and B c are independent, we apply the result of part (a) to the sets A and
B c . Since we know from part (a) that A and B c are independent, part (b) says that Ac and
B c are independent.
A AC
In the Venn diagram at right, assume the sample space has area 1 cor-
responding to probability 1. As drawn, A, B, and C each have area 1/2
AB ABC C
and thus probability 1/2. Moreover, the three way intersection ABC has
probability 1/8. Thus A, B, and C are mutually independent since
B BC P [ABC] = P [A] P [B] P [C] . (1)
A AB B
In the Venn diagram at right, assume the sample space has area 1 cor-
responding to probability 1. As drawn, A, B, and C each have area
AC C BC
1/3 and thus probability 1/3. The three way intersection ABC has zero
probability, implying A, B, and C are not mutually independent since
P [ABC] = 0 = P [A] P [B] P [C] . (1)
However, AB, BC, and AC each has area 1/9. As a result, each pair of events is independent
since
P [AB] = P [A] P [B] , P [BC] = P [B] P [C] , P [AC] = P [A] P [C] . (2)
17
A sequential sample space for this experiment is
1/4 H2 •H1 H2 1/16

1/4 H1 T2 •H1 T2 3/16
3/4

XXX
XX
3/4 X T1 XX
1/4
H2 •T1 H2 3/16
XXX
X 3/4 T2 •T1 T2 9/16
(a) From the tree, we observe
P [H2 ] = P [H1 H2 ] + P [T1 H2 ] = 1/4. (1)
This implies
P [H1 H2 ] 1/16
P [H1 |H2 ] = = = 1/4. (2)
P [H2 ] 1/4
(b) The probability that the first flip is heads and the second flip is tails is P [H1 T2 ] = 3/16.

The tree with adjusted probabilities is
G2 •G1 G2

3/4 3/8

1/2 G1 XXX
XX
1/4 X R2 •G1 R2 1/8

H
HH
1/4 G2 •R1 G2
H
1/8
1/2HH R1
XXX
XX
3/4 X R2 •R1 R2 3/8
From the tree, the probability the second light is green is
P [G2 ] = P [G1 G2 ] + P [R1 G2 ] = 3/8 + 1/8 = 1/2. (1)
The conditional probability that the first light was green given the second light was green is
P [G1 G2 ] P [G2 |G1 ] P [G1 ]

P [G1 |G2 ] = = = 3/4. (2)
P [G2 ] P [G2 ]
Finally, from the tree diagram, we can directly read that P [G2 |G1 ] = 3/4.

Let Gi and Bi denote events indicating whether free throw i was good (Gi ) or bad (Bi ). The tree
for the free throw experiment is
18
G2 •G1 G2

3/4 3/8

1/2 G1 XXX
XX
1/4 X B2 •G1 B2 1/8

HH
HH 1/4 G2 •B1 G2 1/8
1/2HH B1
XXX
XX
3/4 X B2 •B1 B2 3/8
The game goes into overtime if exactly one free throw is made. This event has probability
P [O] = P [G1 B2 ] + P [B1 G2 ] = 1/8 + 1/8 = 1/4. (1)

The tree for this experiment is
1/4 H •AH 1/8

1/2 A T •AT

3/8
3/4

XX
XXX
1/2 X B XX
3/4
•BH 3/8
XXX H
1/4 X T •BT 1/8
The probability that you guess correctly is

P [C] = P [AT ] + P [BH] = 3/8 + 3/8 = 3/4. (1)

The P [− |H ] is the probability that a person who has HIV tests negative for the disease. This is
referred to as a false-negative result. The case where a person who does not have HIV but tests
positive for the disease, is called a false-positive result and has probability P [+|H c ]. Since the test
is correct 99% of the time,
P [−|H] = P [+|H c ] = 0.01. (1)
Now the probability that a person who has tested positive for HIV actually has the disease is
P [+, H] P [+, H]
P [H|+] = = . (2)
P [+] P [+, H] + P [+, H c ]
We can use Bayes’ formula to evaluate these joint probabilities.
P [+|H] P [H]
P [H|+] = (3)
P [+|H] P [H] + P [+|H c ] P [H c ]
(0.99)(0.0002)
= (4)
(0.99)(0.0002) + (0.01)(0.9998)
= 0.0194. (5)
Thus, even though the test is correct 99% of the time, the probability that a random person who
tests positive actually has HIV is less than 0.02. The reason this probability is so low is that the a
priori probability that a person has HIV is very small.
19
Let Ai and Di indicate whether the ith photodetector is acceptable or defective.
4/5 A2 •A1 A2 12/25

3/5 A1 D2 •A1 D2 3/25
1/5

XXX
XX
2/5 X D1 XX
2/5
A2 •D1 A2 4/25
XXX
X 3/5 D2 •D1 D2 6/25
(a) We wish to find the probability P [E1 ] that exactly one photodetector is acceptable. From
the tree, we have
P [E1 ] = P [A1 D2 ] + P [D1 A2 ] = 3/25 + 4/25 = 7/25. (1)
(b) The probability that both photodetectors are defective is P [D1 D2 ] = 6/25.

3/4 H2 •A1 H1 H2 3/32

1/4 H1 T2 •A1 H1 T2 1/32
1/4
1/2 A1
T1 XX
3/4
H2 •A1 T1 H2 9/32
XX

3/4
XX •A1 T1 T2

1/4 T2 3/32

HH
HH 1/4 H2 •B1 H1 H2

3/32
1/2HH B1 X 3/4 •B1 H1 T2
XXX H1
3/4
T2 9/32
XX 1/4
T1 XX H 2 •B1 T1 H2 1/32
1/4
XXX
X 3/4 T2 •B1 T1 T2 3/32
The event H1 H2 that heads occurs on both flips has probability
P [H1 H2 ] = P [A1 H1 H2 ] + P [B1 H1 H2 ] = 6/32. (1)
The probability of H1 is
P [H1 ] = P [A1 H1 H2 ] + P [A1 H1 T2 ] + P [B1 H1 H2 ] + P [B1 H1 T2 ] = 1/2. (2)
Similarly,
P [H2 ] = P [A1 H1 H2 ] + P [A1 T1 H2 ] + P [B1 H1 H2 ] + P [B1 T1 H2 ] = 1/2. (3)
Thus P [H1 H2 ] = P [H1 ]P [H2 ], implying H1 and H2 are not independent. This result should not
be surprising since if the first flip is heads, it is likely that coin B was picked first. In this case, the
second flip is less likely to be heads since it becomes more likely that the second coin flipped was
coin A.
20
(a) The primary difficulty in this problem is translating the words into the correct tree diagram.
The tree for this problem is shown below.
1/2 H3 •T1 H2 H3 1/8

H1 •H1 1/2
1/2
H4 •T1 H2 T3 H4 1/16
1/2

1/2 H2
1/2
T3
1/2
T4 •T1 H2 T3 T4 1/16

XXX
X
1/2 X T1 1/2
1/2 1/2
T2 H3 X X H •T T H H 1/16
Z XXX 4 1 2 3 4
Z 1/2 T 4 •T T H
1 2 3 4T 1/16
Z
Z
1/2 Z
T3 •T1 T2 T3 1/8
(b) From the tree,

P [H3 ] = P [T1 H2 H3 ] + P [T1 T2 H3 H4 ] + P [T1 T2 H3 H4 ] (1)
= 1/8 + 1/16 + 1/16 = 1/4. (2)
Similarly,
P [T3 ] = P [T1 H2 T3 H4 ] + P [T1 H2 T3 T4 ] + P [T1 T2 T3 ] (3)
= 1/8 + 1/16 + 1/16 = 1/4. (4)
(c) The event that Dagwood must diet is

D = (T1 H2 T3 T4 ) ∪ (T1 T2 H3 T4 ) ∪ (T1 T2 T3 ). (5)
The probability that Dagwood must diet is
P [D] = P [T1 H2 T3 T4 ] + P [T1 T2 H3 T4 ] + P [T1 T2 T3 ] (6)
= 1/16 + 1/16 + 1/8 = 1/4. (7)
The conditional probability of heads on flip 1 given that Dagwood must diet is
P [H1 D]
P [H1 |D] = = 0. (8)
P [D]
Remember, if there was heads on flip 1, then Dagwood always postpones his diet.
(d) From part (b), we found that P [H3 ] = 1/4. To check independence, we calculate
P [H2 ] = P [T1 H2 H3 ] + P [T1 H2 T3 ] + P [T1 H2 T4 T4 ] = 1/4 (9)
P [H2 H3 ] = P [T1 H2 H3 ] = 1/8. (10)
Now we find that
P [H2 H3 ] = 1/8 = P [H2 ] P [H3 ] . (11)
Hence, H2 and H3 are dependent events. In fact, P [H3 |H2 ] = 1/2 while P [H3 ] = 1/4. The
reason for the dependence is that given H2 occurred, then we know there will be a third flip
which may result in H3 . That is, knowledge of H2 tells us that the experiment didn’t end
after the first flip.
21
(a) We wish to know what the probability that we find no good photodiodes in n pairs of diodes.
Testing each pair of diodes is an independent trial such that with probability p, both diodes
of a pair are bad. From Problem 1.7.6, we can easily calculate p.
p = P [both diodes are defective] = P [D1 D2 ] = 6/25. (1)
The probability of Zn , the probability of zero acceptable diodes out of n pairs of diodes is pn
because on each test of a pair of diodes, both must be defective.

n n
n 6
P [Zn ] = p=p = (2)
25
i=1
(b) Another way to phrase this question is to ask how many pairs must we test until P [Zn ] ≤ 0.01.
Since P [Zn ] = (6/25)n , we require
n
6 ln 0.01
≤ 0.01 ⇒ n ≥ = 3.23. (3)
25 ln 6/25
Since n must be an integer, n = 4 pairs must be tested.

The experiment ends as soon as a fish is caught. The tree resembles
p C1 p C2 p C3

C1c C2c C3c ...
1−p 1−p 1−p
From the tree, P [C1 ] = p and P [C2 ] = (1 − p)p. Finally, a fish is caught on the nth cast if no fish
were caught on the previous n − 1 casts. Thus,
P [Cn ] = (1 − p)n−1 p. (1)

There are 25 = 32 different binary codes with 5 bits. The number of codes with exactly 3 zeros
equals
5 the number of ways of choosing the bits in which those zeros occur. Therefore there are
3 = 10 codes with exactly 3 zeros.

Since each letter can take on any one of the 4 possible letters in the alphabet, the number of 3
letter words that can be formed is 43 = 64. If we allow each letter to appear only once then we
have 4 choices for the first letter and 3 choices for the second and two choices for the third letter.
Therefore, there are a total of 4 · 3 · 2 = 24 possible codes.
22
(a) The experiment of picking two cards and recording them in the order in which they were
selected can be modeled by two sub-experiments. The first is to pick the first card and
record it, the second sub-experiment is to pick the second card without replacing the first
and recording it. For the first sub-experiment we can have any one of the possible 52 cards
for a total of 52 possibilities. The second experiment consists of all the cards minus the one
that was picked first(because we are sampling without replacement) for a total of 51 possible
outcomes. So the total number of outcomes is the product of the number of outcomes for
each sub-experiment.
52 · 51 = 2652 outcomes. (1)
(b) To have the same card but different suit we can make the following sub-experiments. First
we need to pick one of the 52 cards. Then we need to pick one of the 3 remaining cards that
are of the same type but different suit out of the remaining 51 cards. So the total number
outcomes is
52 · 3 = 156 outcomes. (2)
(c) The probability that the two cards are of the same type but different suit is the number of
outcomes that are of the same type but different suit divided by the total number of outcomes
involved in picking two cards at random from a deck of 52 cards.
156 1
P [same type, different suit] = = . (3)
2652 17
(d) Now we are not concerned with the ordering of the cards. So before, the outcomes (K♥, 8♦)
and (8♦, K♥) were distinct. Now, those two outcomes are not distinct and are only considered
to be the single outcome that a King of hearts and 8 of diamonds were selected. So every
pair of outcomes before collapses to a single outcome when we disregard ordering. So we can
redo parts (a) and (b) above by halving the corresponding values found in parts (a) and (b).
The probability however, does not change because both the numerator and the denominator
have been reduced by an equal factor of 2, which does not change their ratio.

We can break down the experiment of choosing a starting lineup into a sequence of subexperiments:

1. Choose 1 of the 10 pitchers. There are N1 = 10 1 = 10 ways to do this.

2. Choose 1 of the 15 field players to be the designated hitter (DH). There are N2 = 15 1 = 15
ways to do this.
3. Of the remaining
14 field players, choose 8 for the remaining field positions. There are
N3 = 148 to do this.
4. For the 9 batters (consisting of the 8 field players and the designated hitter), choose a batting
lineup. There are N4 = 9! ways to do this.
23
So the total number of different starting lineups when the DH is selected among the field players is

14
N = N1 N2 N3 N4 = (10)(15) 9! = 163,459,296,000. (1)
8
Note that this overestimates the number of combinations the manager must really consider because
most field players can play only one or two positions. Although these constraints on the manager
reduce the number of possible lineups, it typically makes the manager’s job more difficult. As
for the counting, we note that our count did not need to specify the positions played by the field
players. Although this is an important consideration for the manager, it is not part of our counting
of different lineups. In fact, the 8 nonpitching field players are allowed to switch positions at any
time in the field. For example, the shortstop and second baseman could trade positions in the
middle of an inning. Although the DH can go play the field, there are some coomplicated rules
about this. Here is an an excerpt from Major league Baseball Rule 6.10:
The Designated Hitter may be used defensively, continuing to bat in the same posi-
tion in the batting order, but the pitcher must then bat in the place of the substituted
defensive player, unless more than one substitution is made, and the manager then must
designate their spots in the batting order.
If you’re curious, you can find the complete rule on the web.

When the DH can be chosen among all the players, including the pitchers, there are two cases:
• The DH is a field player. In this case, the number of possible lineups, NF , is given in
Problem 1.8.4. In this case, the designated hitter must be chosen from the 15 field players.
We repeat the solution of Problem 1.8.4 here: We can break down the experiment of choosing
a starting lineup into a sequence of subexperiments:

1. Choose 1 of the 10 pitchers. There are N1 = 10 1 = 10 ways to do this.

2. Choose 1 of the 15 field players to be the designated hitter (DH). There are N2 = 15
1 =
15 ways to do this.
3. Of the remaining
14 field players, choose 8 for the remaining field positions. There are
14
N3 = 8 to do this.
4. For the 9 batters (consisting of the 8 field players and the designated hitter), choose a
batting lineup. There are N4 = 9! ways to do this.
So the total number of different starting lineups when the DH is selected among the field
players is
14
N = N1 N2 N3 N4 = (10)(15) 9! = 163,459,296,000. (1)
8
• The DH is a pitcher. In this case, there are 10 choices for the pitcher,
15 10 choices for the
DH among the pitchers (including the pitcher batting for himself), 8 choices for the field
players, and 9! ways of ordering the batters into a lineup. The number of possible lineups is

15
N = (10)(10) 9! = 233, 513, 280, 000. (2)
8
The total number of ways of choosing a lineup is N + N = 396,972,576,000.
24
(a) We can find the number of valid starting lineups by noticing that the swingman presents
three situations: (1) the swingman plays guard, (2) the swingman plays forward, and (3) the
swingman doesn’t play. The first situation is when the swingman can be chosen to play the
guard position, and the second where the swingman can only be chosen to play the forward
position. Let Ni denote the number of lineups corresponding to case i. Then we can write
the total number of lineups as N1 + N2 + N3 . In the first situation, we have to choose 1 out
of 3 centers, 2 out of 4 forwards, and 1 out of 4 guards so that

3 4 4
N1 = = 72. (1)
1 2 1
In the second case, we need to choose 1 out of 3 centers, 1 out of 4 forwards and 2 out of 4
guards, yielding
3 4 4
N2 = = 72. (2)
1 1 2
Finally, with the swingman on the bench, we choose 1 out of 3 centers, 2 out of 4 forward,
and 2 out of four guards. This implies

3 4 4
N3 = = 108, (3)
1 2 2
and the total number of lineups is N1 + N2 + N3 = 252.

What our design must specify is the number of boxes on the ticket, and the number of specially
marked boxes. Suppose each ticket has n boxes and 5 + k specially marked boxes. Note that when
k > 0, a winning ticket will still have k unscratched boxes with the special mark. A ticket is a
winner if each time a box is scratched off, the box has the special mark. Assuming the boxes are
scratched off randomly, the first box scratched off has the mark with probability (5 + k)/n since
there are 5 + k marked boxes out of n boxes. Moreover, if the first scratched box has the mark,
then there are 4 + k marked boxes out of n − 1 remaining boxes. Continuing this argument, the
probability that a ticket is a winner is
5+k4+k 3+k 2+k 1+k (k + 5)!(n − 5)!

p= = . (1)
n n−1n−2n−3n−4 k!n!
By careful choice of n and k, we can choose p close to 0.01. For example,
n 9 11 14 17
k 0 1 2 3 (2)
p 0.0079 0.012 0.0105 0.0090
A gamecard with N = 14 boxes and 5 + k = 7 shaded boxes would be quite reasonable.
25
(a) Since the probability of a zero is 0.8, we can express the probability of the code word 00111
as 2 occurrences of a 0 and three occurrences of a 1. Therefore
P [00111] = (0.8)2 (0.2)3 = 0.00512. (1)
(b) The probability that a code word has exactly three 1’s is

5
P [three 1’s] = (0.8)2 (0.2)3 = 0.0512. (2)
3

Given that the probability that the Celtics win a single championship in any given year is 0.32, we
can find the probability that they win 8 straight NBA championships.
P [8 straight championships] = (0.32)8 = 0.00011. (1)
The probability that they win 10 titles in 11 years is

11
P [10 titles in 11 years] = (.32)10 (.68) = 0.00084. (2)
10
The probability of each of these events is less than 1 in 1000! Given that these events took place
in the relatively short fifty year history of the NBA, it should seem that these probabilities should
be much higher. What the model overlooks is that the sequence of 10 titles in 11 years started
when Bill Russell joined the Celtics. In the years with Russell (and a strong supporting cast) the
probability of a championship was much higher.

We know that the probability of a green and red light is 7/16, and that of a yellow light is 1/8.
Since there are always 5 lights, G, Y , and R obey the multinomial probability law:
2 2
5! 7 1 7
P [G = 2, Y = 1, R = 2] = . (1)
2!1!2! 16 8 16
The probability that the number of green lights equals the number of red lights
P [G = R] = P [G = 1, R = 1, Y = 3] + P [G = 2, R = 2, Y = 1] + P [G = 0, R = 0, Y = 5] (2)
3 2 2 5
5! 7 7 1 5! 7 7 1 5! 1
= + + (3)
1!1!3! 16 16 8 2!1!2! 16 16 8 0!0!5! 8
≈ 0.1449. (4)

For the team with the homecourt advantage, let Wi and Li denote whether game i was a win or a
loss. Because games 1 and 3 are home games and game 2 is an away game, the tree is
26
1−p W2 •W1 W2 p(1−p)

W3 •W1 L2 W3 p
3
p

p W1 L2 L3 •W1 L2 L3 p2 (1−p)

p 1−p

XXX
XX
1−p X L1
1−p p
HH W2 XX W3 •L1 W2 W3 p(1−p)2
XXX
HH 1−p X L3 •L1 W2 L3 (1−p)3
pHH
L2 •L1 L2 p(1−p)
The probability that the team with the home court advantage wins is
P [H] = P [W1 W2 ] + P [W1 L2 W3 ] + P [L1 W2 W3 ] (1)
= p(1 − p) + p + p(1 − p) .
3 2
(2)
Note that P [H] ≤ p for 1/2 ≤ p ≤ 1. Since the team with the home court advantage would win
a 1 game playoff with probability p, the home court team is less likely to win a three game series
than a 1 game playoff!
(a) There are 3 group 1 kickers and 6 group 2 kickers. Using Gi to denote that a group i kicker
was chosen, we have
P [G1 ] = 1/3 P [G2 ] = 2/3. (1)
In addition, the problem statement tells us that
P [K|G1 ] = 1/2 P [K|G2 ] = 1/3. (2)
Combining these facts using the Law of Total Probability yields
P [K] = P [K|G1 ] P [G1 ] + P [K|G2 ] P [G2 ] (3)
= (1/2)(1/3) + (1/3)(2/3) = 7/18. (4)
(b) To solve this part, we need to identify the groups from which the first and second kicker were
chosen. Let ci indicate whether a kicker was chosen from group i and let Cij indicate that
the first kicker was chosen from group i and the second kicker from group j. The experiment
to choose the kickers is described by the sample tree:
2/8 c1 •C11 1/12

3/9 c1 c2 •C12 1/4
6/8
XXXX
X
6/9 X c2 XX
3/8 c1 •C21 1/4
XXX
5/8 X c2 •C22 5/12
Since a kicker from group 1 makes a kick with probability 1/2 while a kicker from group 2
makes a kick with probability 1/3,
P [K1 K2 |C11 ] = (1/2)2 P [K1 K2 |C12 ] = (1/2)(1/3) (5)
P [K1 K2 |C21 ] = (1/3)(1/2) P [K1 K2 |C22 ] = (1/3)2 (6)
27
By the law of total probability,
P [K1 K2 ] = P [K1 K2 |C11 ] P [C11 ] + P [K1 K2 |C12 ] P [C12 ] (7)

+ P [K1 K2 |C21 ] P [C21 ] + P [K1 K2 |C22 ] P [C22 ] (8)
1 1 11 11 1 5
= + + + = 15/96. (9)
4 12 6 4 6 4 9 12
It should be apparent that P [K1 ] = P [K] from part (a). Symmetry should also make it
clear that P [K1 ] = P [K2 ] since for any ordering of two kickers, the reverse ordering is equally
likely. If this is not clear, we derive this result by calculating P [K2 |Cij ] and using the law of
total probability to calculate P [K2 ].
P [K2 |C11 ] = 1/2, P [K2 |C12 ] = 1/3, (10)

P [K2 |C21 ] = 1/2, P [K2 |C22 ] = 1/3. (11)
By the law of total probability,
P [K2 ] = P [K2 |C11 ] P [C11 ] + P [K2 |C12 ] P [C12 ]

+ P [K2 |C21 ] P [C21 ] + P [K2 |C22 ] P [C22 ] (12)
1 1 11 11 1 5 7
= + + + = . (13)
2 12 3 4 2 4 3 12 18
We observe that K1 and K2 are not independent since
2
15 7
P [K1 K2 ] = = = P [K1 ] P [K2 ] . (14)
96 18
Note that 15/96 and (7/18)2 are close but not exactly the same. The reason K1 and K2 are
dependent is that if the first kicker is successful, then it is more likely that kicker is from
group 1. This makes it more likely that the second kicker is from group 2 and is thus more
likely to miss.
(c) Once a kicker is chosen, each of the 10 field goals is an independent trial. If the kicker is
from group 1, then the success probability is 1/2. If the kicker is from group 2, the success
probability is 1/3. Out of 10 kicks, there are 5 misses iff there are 5 successful kicks. Given
the type of kicker chosen, the probability of 5 misses is

10 10
P [M |G1 ] = 5 5
(1/2) (1/2) , P [M |G2 ] = (1/3)5 (2/3)5 . (15)
5 5
We use the Law of Total Probability to find
P [M ] = P [M |G1 ] P [G1 ] + P [M |G2 ] P [G2 ] (16)

10
= (1/3)(1/2)10 + (2/3)(1/3)5 (2/3)5 . (17)
5

From the problem statement, we can conclude that the device components are configured in the
following way.
28
W1 W2 W3 W5
W4 W6
To find the probability that the device works, we replace series devices 1, 2, and 3, and parallel
devices 5 and 6 each with a single device labeled with the probability that it works. In particular,
P [W1 W2 W3 ] = (1 − q)3 , (1)

P [W5 ∪ W6 ] = 1 − P [W5c W6c ] =1−q .2
(2)
This yields a composite device of the form
(1-q)
3 2
1-q
1-q
The probability P [W ] that the two devices in parallel work is 1 minus the probability that neither
works:
P W = 1 − q(1 − (1 − q)3 ). (3)
Finally, for the device to work, both composite device in series must work. Thus, the probability
the device works is
P [W ] = [1 − q(1 − (1 − q)3 )][1 − q 2 ]. (4)

Suppose that the transmitted bit was a 1. We can view each repeated transmission as an indepen-
dent trial. We call each repeated bit the receiver decodes as 1 a success. Using Sk,5 to denote the
event of k successes in the five trials, then the probability k 1’s are decoded at the receiver is

5 k
P [Sk,5 ] = p (1 − p)5−k , k = 0, 1, . . . , 5. (1)
k
The probability a bit is decoded correctly is
P [C] = P [S5,5 ] + P [S4,5 ] = p5 + 5p4 (1 − p) = 0.91854. (2)
The probability a deletion occurs is
P [D] = P [S3,5 ] + P [S2,5 ] = 10p3 (1 − p)2 + 10p2 (1 − p)3 = 0.081. (3)
The probability of an error is
P [E] = P [S1,5 ] + P [S0,5 ] = 5p(1 − p)4 + (1 − p)5 = 0.00046. (4)
Note that if a 0 is transmitted, then 0 is sent five times and we call decoding a 0 a success.
You should convince yourself that this a symmetric situation with the same deletion and error
probabilities. Introducing deletions reduces the probability of an error by roughly a factor of 20.
However, the probability of successfull decoding is also reduced.
29
Note that each digit 0 through 9 is mapped to the 4 bit binary representation of the digit. That is,
0 corresponds to 0000, 1 to 0001, up to 9 which corresponds to 1001. Of course, the 4 bit binary
numbers corresponding to numbers 10 through 15 go unused, however this is unimportant to our
problem. the 10 digit number results in the transmission of 40 bits. For each bit, an independent
trial determines whether the bit was correct, a deletion, or an error. In Problem 1.10.2, we found
the probabilities of these events to be
P [C] = γ = 0.91854, P [D] = δ = 0.081, P [E] = = 0.00046. (1)
Since each of the 40 bit transmissions is an independent trial, the joint probability of c correct bits,
d deletions, and e erasures has the multinomial probability
40! c d e
c!d!e! γ δ c + d + e = 40; c, d, e ≥ 0,
P [C = c, D = d, E = d] = (2)
0 otherwise.

From the statement of Problem 1.10.1, the configuration of device components is
W1 W2 W3 W5
W4 W6
By symmetry, note that the reliability of the system is the same whether we replace component 1,
component 2, or component 3. Similarly, the reliability is the same whether we replace component
5 or component 6. Thus we consider the following cases:
I Replace component 1 In this case
q
P [W1 W2 W3 ] = (1 − )(1 − q)2 , P [W4 ] = 1 − q, P [W5 ∪ W6 ] = 1 − q 2 . (1)
2
This implies
q2
P [W1 W2 W3 ∪ W4 ] = 1 − (1 − P [W1 W2 W3 ])(1 − P [W4 ]) = 1 − (5 − 4q + q 2 ). (2)
2
In this case, the probability the system works is

q2
P [WI ] = P [W1 W2 W3 ∪ W4 ] P [W5 ∪ W6 ] = 1 − (5 − 4q + q ) (1 − q 2 ).
2
(3)
2
II Replace component 4 In this case,

q
P [W1 W2 W3 ] = (1 − q)3 , P [W4 ] = 1 − , P [W5 ∪ W6 ] = 1 − q 2 . (4)
2
This implies
q q
P [W1 W2 W3 ∪ W4 ] = 1 − (1 − P [W1 W2 W3 ])(1 − P [W4 ]) = 1 − + (1 − q)3 . (5)
2 2
q q
P [WII ] = P [W1 W2 W3 ∪ W4 ] P [W5 ∪ W6 ] = 1 − + (1 − q)3 (1 − q 2 ). (6)
2 2
30
III Replace component 5 In this case,
q2
P [W1 W2 W3 ] = (1 − q)3 , P [W4 ] = 1 − q, P [W5 ∪ W6 ] = 1 − . (7)
2
This implies

P [W1 W2 W3 ∪ W4 ] = 1 − (1 − P [W1 W2 W3 ])(1 − P [W4 ]) = (1 − q) 1 + q(1 − q)2 . (8)
P [WIII ] = P [W1 W2 W3 ∪ W4 ] P [W5 ∪ W6 ] (9)

q2
= (1 − q) 1 − 1 + q(1 − q)2 . (10)
2
From these expressions, its hard to tell which substitution creates the most reliable circuit. First,
we observe that P [WII ] > P [WI ] if and only if
q q q2
1− + (1 − q)3 > 1 − (5 − 4q + q 2 ). (11)
2 2 2
Some algebra will show that P [WII ] > P [WI ] if and only if q 2 < 2, which occurs for all nontrivial
(i.e., nonzero) values of q. Similar algebra will show that P [WII ] > P [WIII ] for all values of
0 ≤ q ≤ 1. Thus the best policy is to replace component 4.

We can generate the 200 × 1 vector T, denoted T in Matlab, via the command
T=50+ceil(50*rand(200,1))
Keep in mind that 50*rand(200,1) produces a 200 × 1 vector of random numbers, each in the
interval (0, 50). Applying the ceiling function converts these random numbers to rndom integers in
the set {1, 2, . . . , 50}. Finally, we add 50 to produce random numbers between 51 and 100.

Rather than just solve the problem for 50 trials, we can write a function that generates vectors C
and H for an arbitrary number of trials n. The code for this task is
function [C,H]=twocoin(n);
C=ceil(2*rand(n,1));
P=1-(C/4);
H=(rand(n,1)< P);
The first line produces the n × 1 vector C such that C(i) indicates whether coin 1 or coin 2 is chosen
for trial i. Next, we generate the vector P such that P(i)=0.75 if C(i)=1; otherwise, if C(i)=2,
then P(i)=0.5. As a result, H(i) is the simulated result of a coin flip with heads, corresponding
to H(i)=1, occurring with probability P(i).

Rather than just solve the problem for 100 trials, we can write a function that generates n packets
for an arbitrary number of trials n. The code for this task is
31
function C=bit100(n);
% n is the number of 100 bit packets sent
B=floor(2*rand(n,100));
P=0.03-0.02*B;
E=(rand(n,100)< P);
C=sum((sum(E,2)<=5));
First, B is an n × 100 matrix such that B(i,j) indicates whether bit i of packet j is zero or one.
Next, we generate the n×100 matrix P such that P(i,j)=0.03 if B(i,j)=0; otherwise, if B(i,j)=1,
then P(i,j)=0.01. As a result, E(i,j) is the simulated error indicator for bit i of packet j. That
is, E(i,j)=1 if bit i of packet j is in error; otherwise E(i,j)=0. Next we sum across the rows of
E to obtain the number of errors in each packet. Finally, we count the number of packets with 5 or
more errors.
For n = 100 packets, the packet success probability is inconclusive. Experimentation will show
that C=97, C=98, C=99 and C=100 correct packets are typica values that might be observed. By
increasing n, more consistent results are obtained. For example, repeated trials with n = 100, 000
packets typically produces around C = 98, 400 correct packets. Thus 0.984 is a reasonable estimate
for the probability of a packet being transmitted correctly.

To test n 6-component devices, (such that each component works with probability q) we use the
following function:
function N=reliable6(n,q);
% n is the number of 6 component devices
%N is the number of working devices
W=rand(n,6)>q;
D=(W(:,1)&W(:,2)&W(:,3))|W(:,4);
D=D&(W(:,5)|W(:,6));
N=sum(D);
The n×6 matrix W is a logical matrix such that W(i,j)=1 if component j of device i works properly.
Because W is a logical matrix, we can use the Matlab logical operators | and & to implement the
logic requirements for a working device. By applying these logical operators to the n × 1 columns
of W, we simulate the test of n circuits. Note that D(i)=1 if device i works. Otherwise, D(i)=0.
Lastly, we count the number N of working devices. The following code snippet produces ten sample
runs, where each sample run tests n=100 devices for q = 0.2.
>> for n=1:10, w(n)=reliable6(100,0.2); end
>> w
w =
82 87 87 92 91 85 85 83 90 89
>>
As we see, the number of working devices is typically around 85 out of 100. Solving Problem 1.10.1,
will show that the probability the device works is actually 0.8663.

The code
32
function n=countequal(x,y)
%Usage: n=countequal(x,y)
%n(j)= # elements of x = y(j)
[MX,MY]=ndgrid(x,y);
%each column of MX = x
%each row of MY = y
n=(sum((MX==MY),1))’;
for countequal is quite short (just two lines excluding comments) but needs some explanation.
The key is in the operation
[MX,MY]=ndgrid(x,y).
The Matlab built-in function ndgrid facilitates plotting a function g(x, y) as a surface over the
x, y plane. The x, y plane is represented by a grid of all pairs of points x(i), y(j). When x has n
elements, and y has m elements, ndgrid(x,y) creates a grid (an n × m array) of all possible pairs
[x(i) y(j)]. This grid is represented by two separate n × m matrices: MX and MY which indicate
the x and y values at each grid point. Mathematically,
MX(i,j) = x(i), MY(i,j)=y(j).
Next, C=(MX==MY) is an n × m array such that C(i,j)=1 if x(i)=y(j); otherwise C(i,j)=0. That
is, the jth column of C indicates indicates which elements of x equal y(j). Lastly, we sum along
each column j to count number of elements of x equal to y(j). That is, we sum along column j to
count the number of occurrences (in x) of y(j).

For arbitrary number of trials n and failure probability q, the following functions evaluates replacing
each of the six components by an ultrareliable device.
function N=ultrareliable6(n,q);
% n is the number of 6 component devices
%N is the number of working devices
for r=1:6,
W=rand(n,6)>q;
R=rand(n,1)>(q/2);
W(:,r)=R;
D=(W(:,1)&W(:,2)&W(:,3))|W(:,4);
D=D&(W(:,5)|W(:,6));
N(r)=sum(D);
end
This above code is based on the code for the solution of Problem 1.11.4. The n × 6 matrix W is a
logical matrix such that W(i,j)=1 if component j of device i works properly. Because W is a logical
matrix, we can use the Matlab logical operators | and & to implement the logic requirements for
a working device. By applying these logical opeators to the n × 1 columns of W, we simulate the
test of n circuits. Note that D(i)=1 if device i works. Otherwise, D(i)=0. Note that in the code,
we first generate the matrix W such that each component has failure probability q. To simulate the
replacement of the jth device by the ultrareliable version by replacing the jth column of W by the
column vector R in which a device has failure probability q/2. Lastly, for each column replacement,
we count the number N of working devices. A sample run for n = 100 trials and q = 0.2 yielded
these results:
33
>> ultrareliable6(100,0.2)
ans =
93 89 91 92 90 93
From the above, we see, for example, that replacing the third component with an ultrareliable
component resulted in 91 working devices. The results are fairly inconclusive in that replacing
devices 1, 2, or 3 should yield the same probability of device failure. If we experiment with
n = 10, 000 runs, the results are more definitive:
>> ultrareliable6(10000,0.2)
ans =
8738 8762 8806 9135 8800 8796
>> >> ultrareliable6(10000,0.2)
ans =
8771 8795 8806 9178 8886 8875
>>
In both cases, it is clear that replacing component 4 maximizes the device reliability. The somewhat
complicated solution of Problem 1.10.4 will confirm this observation.
34
(a) We wish to find the value of c that makes the PMF sum up to one.

c(1/2)n n = 0, 1, 2
PN (n) = (1)
0 otherwise

2
Therefore, n=0 PN (n) = c + c/2 + c/4 = 1, implying c = 4/7.
(b) The probability that N ≤ 1 is
P [N ≤ 1] = P [N = 0] + P [N = 1] = 4/7 + 2/7 = 6/7 (2)

From Example 2.5, we can write the PMF of X and the PMF of R as
⎧
⎪
⎪ 1/8 x = 0 ⎧
⎪
⎪
⎨ 3/8 x = 1 ⎨ 1/4 r = 0
PX (x) = 3/8 x = 2 PR (r) = 3/4 r = 2 (1)
⎪
⎪ ⎩
⎪
⎪ 1/8 x = 3 0 otherwise
⎩
0 otherwise
From the PMFs PX (x) and PR (r), we can calculate the requested probabilities
(a) P [X = 0] = PX (0) = 1/8.
(b) P [X < 3] = PX (0) + PX (1) + PX (2) = 7/8.
(c) P [R > 1] = PR (2) = 3/4.
(a) We must choose c to make the PMF of V sum to one.

4
PV (v) = c(12 + 22 + 32 + 42 ) = 30c = 1 (1)
v=1
Hence c = 1/30.
(b) Let U = {u2 |u = 1, 2, . . .} so that
1 42 17
P [V ∈ U ] = PV (1) + PV (4) = + = (2)
30 30 30
(c) The probability that V is even is
22 42 2
P [V is even] = PV (2) + PV (4) = + = (3)
30 30 3
(d) The probability that V > 2 is
32 42 5
P [V > 2] = PV (3) + PV (4) = + = (4)
30 30 6
35
(a) We choose c so that the PMF sums to one.

c c c 7c
PX (x) = + + = =1 (1)
x
2 4 8 8
Thus c = 8/7.
(b)
8 2
P [X = 4] = PX (4) = = (2)
7·4 7
(c)
8 4
P [X < 4] = PX (2) = = (3)
7·2 7
(d)
8 8 3
P [3 ≤ X ≤ 9] = PX (4) + PX (8) = + = (4)
7·4 7·8 7

Using B (for Bad) to denote a miss and G (for Good) to denote a successful free throw, the sample
tree for the number of points scored in the 1 and 1 is
1−p B •Y =0 1−p B •Y =1

p G p G •Y =2
From the tree, the PMF of Y is

⎧
⎪
⎪ 1−p y=0
⎨
p(1 − p) y=1
PY (y) = (1)
⎪ p2
⎪ y=2
⎩
0 otherwise

The probability that a caller fails to get through in three tries is (1 − p)3 . To be sure that at least
95% of all callers get through, we need (1 − p)3 ≤ 0.05. This implies p = 0.6316.

In Problem 2.2.6, each caller is willing to make 3 attempts to get through. An attempt is a failure
if all n operators are busy, which occurs with probability q = (0.8)n . Assuming call attempts are
independent, a caller will suffer three failed attempts with probability q 3 = (0.8)3n . The problem
statement requires that (0.8)3n ≤ 0.05. This implies n ≥ 4.48 and so we need 5 operators.
36
From the problem statement, a single is twice as likely as a double, which is twice as likely as a
triple, which is twice as likely as a home-run. If p is the probability of a home run, then
PB (4) = p PB (3) = 2p PB (2) = 4p PB (1) = 8p (1)
Since a hit of any kind occurs with probability of .300, p + 2p + 4p + 8p = 0.300 which implies
p = 0.02. Hence, the PMF of B is
⎧
⎪
⎪ 0.70 b=0
⎪
⎪
⎪
⎪ 0.16 b=1
⎨
0.08 b=2
PB (b) = (2)
⎪
⎪ 0.04 b=3
⎪
⎪
⎪
⎪ 0.02 b=4
⎩
0 otherwise
(a) In the setup of a mobile call, the phone will send the “SETUP” message up to six times.
Each time the setup message is sent, we have a Bernoulli trial with success probability p. Of
course, the phone stops trying as soon as there is a success. Using r to denote a successful
response, and n a non-response, the sample tree is
p r •K=1 p r •K=2 p r •K=3 p r •K=4 p r •K=5 p r •K=6

1−p n 1−p n 1−p n 1−p n 1−p n 1−p n •K=6
(b) We can write the PMF of K, the number of “SETUP” messages sent as
⎧
⎨ (1 − p)k−1 p k = 1, 2, . . . , 5
PK (k) = (1 − p) p + (1 − p) = (1 − p) k = 6
5 6 5 (1)
⎩
0 otherwise
Note that the expression for PK (6) is different because K = 6 if either there was a success or
a failure on the sixth attempt. In fact, K = 6 whenever there were failures on the first five
attempts which is why PK (6) simplifies to (1 − p)5 .
(c) Let B denote the event that a busy signal is given after six failed setup attempts. The
probability of six consecutive failures is P [B] = (1 − p)6 .
(d) To be sure that P [B] ≤ 0.02, we need p ≥ 1 − (0.02)1/6 = 0.479.
(a) If it is indeed true that Y , the number of yellow M&M’s in a package, is uniformly distributed
between 5 and 15, then the PMF of Y , is

1/11 y = 5, 6, 7, . . . , 15
PY (y) = (1)
0 otherwise
37
(b)
P [Y < 10] = PY (5) + PY (6) + · · · + PY (9) = 5/11 (2)
(c)
P [Y > 12] = PY (13) + PY (14) + PY (15) = 3/11 (3)
(d)
P [8 ≤ Y ≤ 12] = PY (8) + PY (9) + · · · + PY (12) = 5/11 (4)
(a) Each paging attempt is an independent Bernoulli trial with success probability p. The number
of times K that the pager receives a message is the number of successes in n Bernoulli trials
and has the binomial PMF
n k
k p (1 − p)
n−k k = 0, 1, . . . , n
PK (k) = (1)
0 otherwise
(b) Let R denote the event that the paging message was received at least once. The event R has
probability
P [R] = P [B > 0] = 1 − P [B = 0] = 1 − (1 − p)n (2)
To ensure that P [R] ≥ 0.95 requires that n ≥ ln(0.05)/ ln(1 − p). For p = 0.8, we must have
n ≥ 1.86. Thus, n = 2 pages would be necessary.

Whether a hook catches a fish is an independent trial with success probability h. The the number
of fish hooked, K, has the binomial PMF
m k
k h (1 − h)
m−k k = 0, 1, . . . , m
PK (k) = (1)
0 otherwise
(a) Let X be the number of times the frisbee is thrown until the dog catches it and runs away.
Each throw of the frisbee can be viewed as a Bernoulli trial in which a success occurs if the
dog catches the frisbee an runs away. Thus, the experiment ends on the first success and X
has the geometric PMF

(1 − p)x−1 p x = 1, 2, . . .
PX (x) = (1)
0 otherwise
(b) The child will throw the frisbee more than four times iff there are failures on the first 4
trials which has probability (1 − p)4 . If p = 0.2, the probability of more than four throws is
(0.8)4 = 0.4096.
38
Each paging attempt is a Bernoulli trial with success probability p where a success occurs if the
pager receives the paging message.
(a) The paging message is sent again and again until a success occurs. Hence the number of
paging messages is N = n if there are n − 1 paging failures followed by a paging success. That
is, N has the geometric PMF

(1 − p)n−1 p n = 1, 2, . . .
PN (n) = (1)
0 otherwise
(b) The probability that no more three paging attempts are required is
∞

P [N ≤ 3] = 1 − P [N > 3] = 1 − PN (n) = 1 − (1 − p)3 (2)
n=4
This answer can be obtained without calculation since N > 3 if the first three paging attempts
fail and that event occurs with probability (1 − p)3 . Hence, we must choose p to satisfy
1 − (1 − p)3 ≥ 0.95 or (1 − p)3 ≤ 0.05. This implies
p ≥ 1 − (0.05)1/3 ≈ 0.6316 (3)

The probability of more than 500,000 bits is

500,000
P [B > 500,000] = 1 − PB (b) (1)
b=1

500,000
=1−p (1 − p)b−1 (2)
b=1

500,000
Math Fact B.4 implies that (1 − x) b=1 xb−1 = 1 − x500,000 . Substituting, x = 1 − p, we obtain:
P [B > 500,000] = 1 − (1 − (1 − p)500,000 ) (3)

−5 500,000
= (1 − 0.25 × 10 ) ≈ exp(−500,000/400,000) = 0.29. (4)

Since an average of T /5 buses arrive in an interval of T minutes, buses arrive at the bus stop at a
rate of 1/5 buses per minute.
(a) From the definition of the Poisson PMF, the PMF of B, the number of buses in T minutes,
is
(T /5)b e−T /5 /b! b = 0, 1, . . .
PB (b) = (1)
0 otherwise
(b) Choosing T = 2 minutes, the probability that three buses arrive in a two minute interval is
PB (3) = (2/5)3 e−2/5 /3! ≈ 0.0072 (2)
39
(c) By choosing T = 10 minutes, the probability of zero buses arriving in a ten minute interval is
PB (0) = e−10/5 /0! = e−2 ≈ 0.135 (3)
(d) The probability that at least one bus arrives in T minutes is
P [B ≥ 1] = 1 − P [B = 0] = 1 − e−T /5 ≥ 0.99 (4)
Rearranging yields T ≥ 5 ln 100 ≈ 23.0 minutes.
(a) If each message is transmitted 8 times and the probability of a successful transmission is p,
then the PMF of N , the number of successful transmissions has the binomial PMF
8 n
n p (1 − p)
8−n n = 0, 1, . . . , 8
PN (n) = (1)
0 otherwise
(b) The indicator random variable I equals zero if and only if N = 8. Hence,
P [I = 0] = P [N = 0] = 1 − P [I = 1] (2)
Thus, the complete expression for the PMF of I is

⎧
⎨ (1 − p)8 i=0
PI (i) = 1 − (1 − p) i = 1
8 (3)
⎩
0 otherwise

n
The requirement that x=1 PX (x) = 1 implies

1
n=1: c(1) =1 c(1) = 1 (1)
1

1 1 2
n=2: c(2) + =1 c(2) = (2)
1 2 3

1 1 1 6
n=3: c(3) + + =1 c(3) = (3)
1 2 3 11

1 1 1 1 12
n=4: c(4) + + + =1 c(4) = (4)
1 2 3 4 25

1 1 1 1 1 12
n=5: c(5) + + + + =1 c(5) = (5)
1 2 3 4 5 25

1 1 1 1 1 20
n=6: c(6) + + + + =1 c(6) = (6)
1 2 3 4 6 49
As an aside, find c(n) for large values of n is easy using the recursion
1 1 1
= + . (7)
c(n + 1) c(n) n + 1
40
(a) We can view whether each caller knows the birthdate as a Bernoulli trial. As a result, L is
the number of trials needed for 6 successes. That is, L has a Pascal PMF with parameters
p = 0.75 and k = 6 as defined by Definition 2.8. In particular,
l−1 6 l−6 l = 6, 7, . . .
PL (l) = 5 (0.75) (0.25) (1)
0 otherwise
(b) The probability of finding the winner on the tenth call is

9
PL (10) = (0.75)6 (0.25)4 ≈ 0.0876 (2)
5
(c) The probability that the station will need nine or more calls to find a winner is
P [L ≥ 9] = 1 − P [L < 9] (3)
= 1 − PL (6) − PL (7) − PL (8) (4)
= 1 − (0.75) [1 + 6(0.25) + 21(0.25) ] ≈ 0.321
6 2
(5)

The packets are delay sensitive and can only be retransmitted d times. For t < d, a packet is
transmitted t times if the first t − 1 attempts fail followed by a successful transmission on attempt
t. Further, the packet is transmitted d times if there are failures on the first d − 1 transmissions,
no matter what the outcome of attempt d. So the random variable T , the number of times that a
packet is transmitted, can be represented by the following PMF.
⎧
⎨ p(1 − p)t−1 t = 1, 2, . . . , d − 1
PT (t) = (1 − p)d−1 t = d (1)
⎩
0 otherwise
(a) Since each day is independent of any other day, P [W33 ] is just the probability that a winning
lottery ticket was bought. Similarly for P [L87 ] and P [N99 ] become just the probability that a
losing ticket was bought and that no ticket was bought on a single day, respectively. Therefore
P [W33 ] = p/2 P [L87 ] = (1 − p)/2 P [N99 ] = 1/2 (1)
(b) Suppose we say a success occurs on the kth trial if on day k we buy a ticket. Otherwise, a
failure occurs. The probability of success is simply 1/2. The random variable K is just the
number of trials until the first success and has the geometric PMF

(1/2)(1/2)k−1 = (1/2)k k = 1, 2, . . .
PK (k) = (2)
0 otherwise
41
(c) The probability that you decide to buy a ticket and it is a losing ticket is (1−p)/2, independent
of any other day. If we view buying a losing ticket as a Bernoulli success, R, the number of
losing lottery tickets bought in m days, has the binomial PMF
m
r [(1 − p)/2] [(1 + p)/2]
r m−r r = 0, 1, . . . , m
PR (r) = (3)
0 otherwise
(d) Letting D be the day on which the j-th losing ticket is bought, we can find the probability
that D = d by noting that j − 1 losing tickets must have been purchased in the d − 1 previous
days. Therefore D has the Pascal PMF

j−1 [(1 − p)/2] [(1 + p)/2]
d−1 j d−j d = j, j + 1, . . .
PD (d) = (4)
0 otherwise
(a) Let Sn denote the event that the Sixers win the series in n games. Similarly, Cn is the event
that the Celtics in in n games. The Sixers win the series in 3 games if they win three straight,
which occurs with probability
P [S3 ] = (1/2)3 = 1/8 (1)
The Sixers win the series in 4 games if they win two out of the first three games and they
win the fourth game so that

3
P [S4 ] = (1/2)3 (1/2) = 3/16 (2)
2
The Sixers win the series in five games if they win two out of the first four games and then
win game five. Hence,
4
P [S5 ] = (1/2)4 (1/2) = 3/16 (3)
2
By symmetry, P [Cn ] = P [Sn ]. Further we observe that the series last n games if either the
Sixers or the Celtics win the series in n games. Thus,
P [N = n] = P [Sn ] + P [Cn ] = 2P [Sn ] (4)
Consequently, the total number of games, N , played in a best of 5 series between the Celtics
and the Sixers can be described by the PMF
⎧
⎪
⎪ 2(1/2)3 = 1/4 n=3
⎨ 3
21(1/2)4 = 3/8 n=4
PN (n) = (5)
⎪
⎪ 2 4 (1/2)5 = 3/8 n=5
⎩ 2
0 otherwise
(b) For the total number of Celtic wins W , we note that if the Celtics get w < 3 wins, then the
Sixers won the series in 3 + w games. Also, the Celtics win 3 games if they win the series in
3,4, or 5 games. Mathematically,

P [S3+w ] w = 0, 1, 2
P [W = w] = (6)
P [C3 ] + P [C4 ] + P [C5 ] w = 3
42
Thus, the number of wins by the Celtics, W , has the PMF shown below.
⎧
⎪
⎪ P [S3 ] = 1/8 w=0
⎪
⎪
⎨ P [S4 ] = 3/16 w=1
PW (w) = P [S5 ] = 3/16 w=2 (7)
⎪
⎪
⎪ 1/8 + 3/16 + 3/16 = 1/2
⎪ w=3
⎩
0 otherwise
(c) The number of Celtic losses L equals the number of Sixers’ wins WS . This implies PL (l) =
PWS (l). Since either team is equally likely to win any game, by symmetry, PWS (w) = PW (w).
This implies PL (l) = PWS (l) = PW (l). The complete expression of for the PMF of L is
⎧
⎪
⎪ 1/8 l = 0
⎪
⎪
⎨ 3/16 l = 1
PL (l) = PW (l) = 3/16 l = 2 (8)
⎪
⎪
⎪ 1/2 l = 3
⎪
⎩
0 otherwise

Since a and b are positive, let K be a binomial random variable for n trials and success probability
p = a/(a + b). First, we observe that the sum of over all possible values of the PMF of K is

n n
n
PK (k) = pk (1 − p)n−k (1)
k
k=0 k=0
n k n−k
n a b
= (2)
k a+b a+b
k=0

n n k n−k
a b
= k=0 k n (3)
(a + b)

n
Since k=0 PK (k) = 1, we see that

n n

n n n
(a + b) = (a + b) PK (k) = ak bn−k (4)
k
k=0 k=0

Using the CDF given in the problem statement we find that
(a) P [Y < 1] = 0
(b) P [Y ≤ 1] = 1/4
(c) P [Y > 2] = 1 − P [Y ≤ 2] = 1 − 1/2 = 1/2
(d) P [Y ≥ 2] = 1 − P [Y < 2] = 1 − 1/4 = 3/4
(e) P [Y = 1] = 1/4
43
(f) P [Y = 3] = 1/2
(g) From the staircase CDF of Problem 2.4.1, we see that Y is a discrete random variable. The
jumps in the CDF occur at at the values that Y can take on. The height of each jump equals
the probability of that value. The PMF of Y is
⎧
⎪
⎪ 1/4 y = 1
⎨
1/4 y = 2
PY (y) = (1)
⎪
⎪ 1/2 y = 3
⎩
0 otherwise
(a) The given CDF is shown in the diagram below.

1 ⎧
0.8 ⎪
⎪ 0 x < −1
⎨
FX(x)
0.6
0.4 0.2 −1 ≤ x < 0
FX (x) = (1)
0.2
⎪ 0.7
⎪ 0≤x<1
0 ⎩
−2 −1 0 1 2 1 x≥1
x
(b) The corresponding PMF of X is

⎧
⎪
⎪ 0.2 x = −1
⎨
0.5 x=0
PX (x) = (2)
⎪
⎪ 0.3 x=1
⎩
0 otherwise
(a) Similar to the previous problem, the graph of the CDF is shown below.
1 ⎧
0.8 ⎪
⎪ 0 x < −3
⎨
FX(x)
0.6
0.4 0.4 −3 ≤ x < 5
FX (x) = (1)
0.2
⎪
⎪ 0.8 5 ≤ x < 7
0 ⎩
−3 0 5 7 1 x≥7
x
(b) The corresponding PMF of X is

⎧
⎪
⎪ 0.4 x = −3
⎨
0.4 x=5
PX (x) = (2)
⎪
⎪ 0.2 x=7
⎩
0 otherwise
44
Let q = 1 − p, so the PMF of the geometric (p) random variable K is
k−1
pq k = 1, 2, . . . ,
PK (k) = (1)
0 otherwise.
For any integer k ≥ 1, the CDF obeys

k
k
FK (k) = PK (j) = pq j−1 = 1 − q k . (2)
j=1 j=1
Since K is integer valued, FK (k) = FK (k) for all integer and non-integer values of k. (If this
point is not clear, you should review Example 2.24.) Thus, the complete expression for the CDF
of K is
0 k < 1,
FK (k) = k (3)
1 − (1 − p) k ≥ 1.

Since mushrooms occur with probability 2/3, the number of pizzas sold before the first mushroom
pizza is N = n < 100 if the first n pizzas do not have mushrooms followed by mushrooms on pizza
n + 1. Also, it is possible that N = 100 if all 100 pizzas are sold without mushrooms. the resulting
PMF is ⎧
⎨ (1/3)n (2/3) n = 0, 1, . . . , 99
PN (n) = (1/3)100 n = 100 (1)
⎩
0 otherwise
For integers n < 100, the CDF of N obeys

n
n
FN (n) = PN (i) = (1/3)i (2/3) = 1 − (1/3)n+1 (2)
i=0 i=0
A complete expression for FN (n) must give a valid answer for every value of n, including non-integer
values. We can write the CDF using the floor function x which denote the largest integer less
than or equal to X. The complete expression for the CDF is
⎧
⎨ 0 x<0
FN (x) = 1 − (1/3)x+1 0 ≤ x < 100 (3)
⎩
1 x ≥ 100

From Problem 2.2.8, the PMF of B is
⎧
⎪
⎪ 0.70 b = 0
⎪
⎪
⎪
⎪ 0.16 b = 1
⎨
0.08 b = 2
PB (b) = (1)
⎪
⎪ 0.04 b = 3
⎪
⎪
⎪
⎪ 0.02 b = 4
⎩
0 otherwise
45
The corresponding CDF is
⎧
⎪
⎪ 0 b<0
1 ⎪
⎪ 0≤b<1
0.75 ⎪
⎪ 0.70
⎨
FB(b)
0.5 0.86 1≤b<2

FB (b) = (2)
0.25 ⎪
⎪ 0.94 2≤b<3
0 ⎪
⎪
−1 0 1 2 3 4 5 ⎪
⎪ 0.98 3≤b<4
⎩
b 1.0 b≥4

In Problem 2.2.5, we found the PMF of Y . This PMF, and its corresponding CDF are
⎧ ⎧
⎪
⎪ 1−p y=0 ⎪
⎪ 0 y<0
⎨ ⎨
p(1 − p) y = 1 1−p 0≤y <1
PY (y) = FY (y) = (1)
⎪
⎪ p2 y=2 ⎪
⎪ 1 − p2 1 ≤ y < 2
⎩ ⎩
0 otherwise 1 y≥2
For the three values of p, the CDF resembles
1 1 1
0.75 0.75 0.75
FY(y)
FY(y)
FY(y)
0.5 0.5 0.5
0.25 0.25 0.25
0 0 0 (2)
−1 0 1 2 3 −1 0 1 2 3 −1 0 1 2 3
y y y
p = 1/4 p = 1/2 p = 3/4

From Problem 2.2.9, the PMF of the number of call attempts is
⎧
⎨ (1 − p)k−1 p k = 1, 2, . . . , 5
PN (n) = (1 − p)5 p + (1 − p)6 = (1 − p)5 k = 6 (1)
⎩
0 otherwise
For p = 1/2, the PMF can be simplified to

⎧
⎨ (1/2)n n = 1, 2, . . . , 5
PN (n) = (1/2)5 n = 6 (2)
⎩
0 otherwise
The corresponding CDF of N is

⎧
⎪
⎪ 0 n<1
⎪
⎪ 1≤n<2
1 ⎪
⎪ 1/2
0.75 ⎪
⎪
⎨ 3/4 2≤n<3
FN(n)
0.5
0.25 FN (n) = 7/8 3≤n<4 (3)
⎪
⎪
0 ⎪ 15/16
⎪ 4≤n<5
0 1 2 3 4 5 6 7 ⎪
⎪
⎪
⎪ 31/32 5≤n<6
n ⎩
1 n≥6
46
For this problem, we just need to pay careful attention to the definitions of mode and median.
(a) The mode must satisfy PX (xmod ) ≥ PX (x) for all x. In the case of the uniform PMF, any
integer x between 1 and 100 is a mode of the random variable X. Hence, the set of all modes
is
Xmod = {1, 2, . . . , 100} (1)
(b) The median must satisfy P [X < xmed ] = P [X > xmed ]. Since
P [X ≤ 50] = P [X ≥ 51] = 1/2 (2)
we observe that xmed = 50.5 is a median since it satisfies
P [X < xmed ] = P [X > xmed ] = 1/2 (3)
In fact, for any x satisfying 50 < x < 51, P [X < x ] = P [X > x ] = 1/2. Thus,
Xmed = {x|50 < x < 51} (4)

Voice calls and data calls each cost 20 cents and 30 cents respectively. Furthermore the respective
probabilities of each type of call are 0.6 and 0.4.
(a) Since each call is either a voice or data call, the cost of one call can only take the two values
associated with the cost of each type of call. Therefore the PMF of X is
⎧
⎨ 0.6 x = 20
PX (x) = 0.4 x = 30 (1)
⎩
0 otherwise
(b) The expected cost, E[C], is simply the sum of the cost of each type of call multiplied by the
probability of such a call occurring.
E [C] = 20(0.6) + 30(0.4) = 24 cents (2)

From the solution to Problem 2.4.1, the PMF of Y is
⎧
⎪
⎪ 1/4 y=1
⎨
1/4 y=2
PY (y) = (1)
⎪
⎪ 1/2 y=3
⎩
0 otherwise
The expected value of Y is

E [Y ] = yPY (y) = 1(1/4) + 2(1/4) + 3(1/2) = 9/4 (2)
y
47
From the solution to Problem 2.4.2, the PMF of X is
⎧
⎪
⎪ 0.2 x = −1
⎨
0.5 x=0
PX (x) = (1)
⎪
⎪ 0.3 x=1
⎩
0 otherwise
The expected value of X is

E [X] = xPX (x) = −1(0.2) + 0(0.5) + 1(0.3) = 0.1 (2)
x

⎧
⎪
⎪ 0.4 x = −3
⎨
0.4 x=5
PX (x) = (1)
⎪
⎪ 0.2 x=7
⎩
0 otherwise

E [X] = xPX (x) = −3(0.4) + 5(0.4) + 7(0.2) = 2.2 (2)
x

From Definition 2.7, random variable X has PMF
4 4 x = 0, 1, 2, 3, 4
PX (x) = x (1/2) (1)
0 otherwise

4
4 1 4 1 4 1 4 1 4 1
E [X] = xPX (x) = 0 4
+1 4
+2 4
+3 4
+4 (2)
0 2 1 2 2 2 3 2 4 24
x=0
= [4 + 12 + 12 + 4]/24 = 2 (3)

From Definition 2.7, random variable X has PMF
5 5 x = 0, 1, 2, 3, 4, 5
PX (x) = x (1/2) (1)
0 otherwise

5
E [X] = xPX (x) (2)
x=0

5 1 5 1 5 1 5 1 5 1 5 1
=0 5
+1 5
+2 5
+3 5
+4 5
+5 (3)
0 2 1 2 2 2 3 2 4 2 5 25
= [5 + 20 + 30 + 20 + 5]/25 = 2.5 (4)
48
The following experiments are based on a common model of packet transmissions in data networks.
In these networks, each data packet contains a cylic redundancy check (CRC) code that permits
the receiver to determine whether the packet was decoded correctly. In the following, we assume
that a packet is corrupted with probability = 0.001, independent of whether any other packet is
corrupted.
(a) Let X = 1 if a data packet is decoded correctly; otherwise X = 0. Random variable X is a
Bernoulli random variable with PMF
⎧
⎨ 0.001 x=0
PX (x) = 0.999 x=1 (1)
⎩
0 otherwise
The parameter = 0.001 is the probability a packet is corrupted. The expected value of X is
E [X] = 1 − = 0.999 (2)
(b) Let Y denote the number of packets received in error out of 100 packets transmitted. Y has
the binomial PMF
100 y 100−y y = 0, 1, . . . , 100
PY (y) = y (0.001) (0.999) (3)
0 otherwise
E [Y ] = 100 = 0.1 (4)
(c) Let L equal the number of packets that must be received to decode 5 packets in error. L has
the Pascal PMF l−1 5 l−5 l = 5, 6, . . .
PL (l) = 4 (0.001) (0.999) (5)
0 otherwise
The expected value of L is
5 5
E [L] = = = 5000 (6)
0.001
(d) If packet arrivals obey a Poisson model with an average arrival rate of 1000 packets per
second, then the number N of packets that arrive in 5 seconds has the Poisson PMF

5000n e−5000 /n! n = 0, 1, . . .
PN (n) = (7)
0 otherwise
The expected value of N is E[N ] = 5000.

In this ”double-or-nothing” type game, there are only two possible payoffs. The first is zero dollars,
which happens when we lose 6 straight bets, and the second payoff is 64 dollars which happens
unless we lose 6 straight bets. So the PMF of Y is
⎧
⎨ (1/2)6 = 1/64 y=0
PY (y) = 1 − (1/2) = 63/64 y = 64
6 (1)
⎩
0 otherwise
The expected amount you take home is
E [Y ] = 0(1/64) + 64(63/64) = 63 (2)
So, on the average, we can expect to break even, which is not a very exciting proposition.
49
By the definition of the expected value,
n
n x
E [Xn ] = x p (1 − p)n−x (1)
x
x=1

n
(n − 1)!
= np px−1 (1 − p)n−1−(x−1) (2)
(x − 1)!(n − 1 − (x − 1))!
x=1
With the substitution x = x − 1, we have

n − 1
n−1
n−1
n−x
E [Xn ] = np p x
(1 − p) = np PXn−1 (x) = np (3)
x
x =0 x =0

1
The above sum is 1 because it is the sum of a binomial random variable for n − 1 trials over all
possible values.

We write the sum as a double sum in the following way:
∞
∞
∞
P [X > i] = PX (j) (1)
i=0 i=0 j=i+1
At this point, the key step is to reverse the order of summation. You may need to make a sketch
of the feasible values for i and j to see how this reversal occurs. In this case,
∞
∞
j−1 ∞

P [X > i] = PX (j) = jPX (j) = E [X] (2)
i=0 j=1 i=0 j=1

⎧
⎪
⎪ 1/4 y=1
⎨
1/4 y=2
PY (y) = (1)
⎪
⎪ 1/2 y=3
⎩
0 otherwise
(a) Since Y has range SY = {1, 2, 3}, the range of U = Y 2 is SU = {1, 4, 9}. The PMF of U can
be found by observing that
√ √
P [U = u] = P Y 2 = u = P Y = u + P Y = − u (2)
√
Since Y is never negative, PU (u) = PY ( u). Hence,
PU (1) = PY (1) = 1/4 PU (4) = PY (2) = 1/4 PU (9) = PY (3) = 1/2 (3)
For all other values of u, PU (u) = 0. The complete expression for the PMF of U is
⎧
⎪
⎪ 1/4 u = 1
⎨
1/4 u = 4
PU (u) = (4)
⎪
⎪ 1/2 u = 9
⎩
0 otherwise
50
(b) From the PMF, it is straighforward to write down the CDF.
⎧
⎪
⎪ 0 u<1
⎨
1/4 1 ≤ u < 4
FU (u) = (5)
⎪
⎪ 1/2 4 ≤ u < 9
⎩
1 u≥9
(c) From Definition 2.14, the expected value of U is

E [U ] = uPU (u) = 1(1/4) + 4(1/4) + 9(1/2) = 5.75 (6)
u
From Theorem 2.10, we can calculate the expected value of U as

2
E [U ] = E Y 2 = y PY (y) = 12 (1/4) + 22 (1/4) + 32 (1/2) = 5.75 (7)
y
As we expect, both methods yield the same answer.

⎧
⎪
⎪ 0.2 x = −1
⎨
0.5 x=0
PX (x) = (1)
⎪
⎪ 0.3 x=1
⎩
0 otherwise
(a) The PMF of V = |X| satisfies
PV (v) = P [|X| = v] = PX (v) + PX (−v) (2)
In particular,
PV (0) = PX (0) = 0.5 PV (1) = PX (−1) + PX (1) = 0.5 (3)
The complete expression for the PMF of V is

⎧
⎨ 0.5 v = 0
PV (v) = 0.5 v = 1 (4)
⎩
0 otherwise
(b) From the PMF, we can construct the staircase CDF of V .

⎧
⎨ 0 v<0
FV (v) = 0.5 0 ≤ v < 1 (5)
⎩
1 v≥1
(c) From the PMF PV (v), the expected value of V is

E [V ] = PV (v) = 0(1/2) + 1(1/2) = 1/2 (6)
v
You can also compute E[V ] directly by using Theorem 2.10.
51
⎧
⎪
⎪ 0.4 x = −3
⎨
0.4 x=5
PX (x) = (1)
⎪
⎪ 0.2 x=7
⎩
0 otherwise
(a) The PMF of W = −X satisfies
PW (w) = P [−X = w] = PX (−w) (2)
This implies
PW (−7) = PX (7) = 0.2 PW (−5) = PX (5) = 0.4 PW (3) = PX (−3) = 0.4 (3)
The complete PMF for W is

⎧
⎪
⎪ 0.2 w = −7
⎨
0.4 w = −5
PW (w) = (4)
⎪
⎪ 0.4 w=3
⎩
0 otherwise
(b) From the PMF, the CDF of W is

⎧
⎪
⎪ 0 w < −7
⎨
0.2 −7 ≤ w < −5
FW (w) = (5)
⎪
⎪ 0.6 −5 ≤ w < 3
⎩
1 w≥3
(c) From the PMF, W has expected value

E [W ] = wPW (w) = −7(0.2) + −5(0.4) + 3(0.4) = −2.2 (6)
w

A tree for the experiment is
1/3 D=99.75 •C=100074.75

XXX 1/3 D=100 •C=10100
XXX
1/3 X D=100.25 •C=10125.13
Thus C has three equally likely outcomes. The PMF of C is

1/3 c = 100,074.75, 10,100, 10,125.13
PC (c) = (1)
0 otherwise
52
(a) The source continues to transmit packets until one is received correctly. Hence, the total
number of times that a packet is transmitted is X = x if the first x − 1 transmissions were in
error. Therefore the PMF of X is
x−1
q (1 − q) x = 1, 2, . . .
PX (x) = (1)
0 otherwise
(b) The time required to send a packet is a millisecond and the time required to send an acknowl-
edgment back to the source takes another millisecond. Thus, if X transmissions of a packet
are needed to send the packet correctly, then the packet is correctly received after T = 2X − 1
milliseconds. Therefore, for an odd integer t > 0, T = t iff X = (t + 1)/2. Thus,
(t−1)/2
q (1 − q) t = 1, 3, 5, . . .
PT (t) = PX ((t + 1)/2) = (2)
0 otherwise

The cellular calling plan charges a flat rate of $20 per month up to and including the 30th minute,
and an additional 50 cents for each minute over 30 minutes. Knowing that the time you spend on
the phone is a geometric random variable M with mean 1/p = 30, the PMF of M is

(1 − p)m−1 p m = 1, 2, . . .
PM (m) = (1)
0 otherwise
The monthly cost, C obeys

30
PC (20) = P [M ≤ 30] = (1 − p)m−1 p = 1 − (1 − p)30 (2)
m=1
When M ≥ 30, C = 20 + (M − 30)/2 or M = 2C − 10. Thus,
PC (c) = PM (2c − 10) c = 20.5, 21, 21.5, . . . (3)
The complete PMF of C is

1 − (1 − p)30 c = 20
PC (c) = (4)
(1 − p)2c−10−1 p c = 20.5, 21, 21.5, . . .

From the solution to Quiz 2.6, we found that T = 120 − 15N . By Theorem 2.10,

E [T ] = (120 − 15n)PN (n) (1)
n∈SN
= 0.1(120) + 0.3(120 − 15) + 0.3(120 − 30) + 0.3(120 − 45) = 93 (2)
Also from the solution to Quiz 2.6, we found that

⎧
⎨ 0.3 t = 75, 90, 105
PT (t) = 0.1 t = 120 (3)
⎩
0 otherwise
53
Using Definition 2.14,

E [T ] = tPT (t) = 0.3(75) + 0.3(90) + 0.3(105) + 0.1(120) = 93 (4)
t∈ST
As expected, the two calculations give the same answer.

Whether a lottery ticket is a winner is a Bernoulli trial with a success probability of 0.001. If we
buy one every day for 50 years for a total of 50 · 365 = 18250 tickets, then the number of winning
tickets T is a binomial random variable with mean
E [T ] = 18250(0.001) = 18.25 (1)
Since each winning ticket grosses $1000, the revenue we collect over 50 years is R = 1000T dollars.
The expected revenue is
E [R] = 1000E [T ] = 18250 (2)
But buying a lottery ticket everyday for 50 years, at $2.00 a pop isn’t cheap and will cost us a total
of 18250 · 2 = $36500. Our net profit is then Q = R − 36500 and the result of our loyal 50 year
patronage of the lottery system, is disappointing expected loss of
E [Q] = E [R] − 36500 = −18250 (3)

Let X denote the number of points the shooter scores. If the shot is uncontested, the expected
number of points scored is
E [X] = (0.6)2 = 1.2 (1)
If we foul the shooter, then X is a binomial random variable with mean E[X] = 2p. If 2p > 1.2,
then we should not foul the shooter. Generally, p will exceed 0.6 since a free throw is usually
even easier than an uncontested shot taken during the action of the game. Furthermore, fouling
the shooter ultimately leads to the the detriment of players possibly fouling out. This suggests
that fouling a player is not a good idea. The only real exception occurs when facing a player like
Shaquille O’Neal whose free throw probability p is lower than his field goal percentage during a
game.

Given the distributions of D, the waiting time in days and the resulting cost, C, we can answer the
following questions.
(a) The expected waiting time is simply the expected value of D.

4
E [D] = d · PD (d) = 1(0.2) + 2(0.4) + 3(0.3) + 4(0.1) = 2.3 (1)
d=1
(b) The expected deviation from the waiting time is
E [D − µD ] = E [D] − E [µd ] = µD − µD = 0 (2)
54
(c) C can be expressed as a function of D in the following manner.
⎧
⎪
⎪ 90 D = 1
⎨
70 D = 2
C(D) = (3)
⎪
⎪ 40 D = 3
⎩
40 D = 4
(d) The expected service charge is
E [C] = 90(0.2) + 70(0.4) + 40(0.3) + 40(0.1) = 62 dollars (4)

As a function of the number of minutes used, M , the monthly cost is

20 M ≤ 30
C(M ) = (1)
20 + (M − 30)/2 M ≥ 30
The expected cost per month is

∞

30 ∞

E [C] = C(m)PM (m) = 20PM (m) + (20 + (m − 30)/2)PM (m) (2)
m=1 m=1 m=31
∞ ∞
1
= 20 PM (m) + (m − 30)PM (m) (3)
2
m=1 m=31

∞
Since m=1 PM (m) = 1 and since PM (m) = (1 − p)m−1 p for m ≥ 1, we have
∞
(1 − p)30
E [C] = 20 + (m − 30)(1 − p)m−31 p (4)
2
m=31
Making the substitution j = m − 30 yields

∞
(1 − p)30 (1 − p)30
E [C] = 20 + j(1 − p)j−1 p = 20 + (5)
2 2p
j=1

Since our phone use is a geometric random variable M with mean value 1/p,

(1 − p)m−1 p m = 1, 2, . . .
PM (m) = (1)
0 otherwise
For this cellular billing plan, we are given no free minutes, but are charged half the flat fee. That
is, we are going to pay 15 dollars regardless and $1 for each minute we use the phone. Hence
C = 15 + M and for c ≥ 16, P [C = c] = P [M = c − 15]. Thus we can construct the PMF of the
cost C
(1 − p)c−16 p c = 16, 17, . . .
PC (c) = (2)
0 otherwise
55
Since C = 15 + M , the expected cost per month of the plan is
E [C] = E [15 + M ] = 15 + E [M ] = 15 + 1/p (3)
In Problem 2.7.5, we found that that the expected cost of the plan was
E [C] = 20 + [(1 − p)30 ]/(2p) (4)
In comparing the expected costs of the two plans, we see that the new plan is better (i.e. cheaper)
if
15 + 1/p ≤ 20 + [(1 − p)30 ]/(2p) (5)
A simple plot will show that the new plan is better if p ≤ p0 ≈ 0.2.

We define random variable W such that W = 1 if the circuit works or W = 0 if the circuit is
defective. (In the probability literature, W is called an indicator random variable.) Let Rs denote
the profit on a circuit with standard devices. Let Ru denote the profit on a circuit with ultrareliable
devices. We will compare E[Rs ] and E[Ru ] to decide which circuit implementation offers the highest
expected profit.
The circuit with standard devices works with probability (1 − q)10 and generates revenue of k
dollars if all of its 10 constituent devices work. We observe that we can we can express Rs as a
function rs (W ) and that we can find the PMF PW (w):
⎧
⎨ 1 − (1 − q)10 w = 0,
−10 W = 0,
Rs = rs (W ) = PW (w) = (1 − q)10 w = 1, (1)
k − 10 W = 1, ⎩
0 otherwise.
Thus we can express the expected profit as

1
E [rs (W )] = PW (w) rs (w) (2)
w=0
= PW (0) (−10) + PW (1) (k − 10) (3)
= (1 − (1 − q)10 )(−10) + (1 − q)10 (k − 10) = (0.9)10 k − 10. (4)
For the ultra-reliable case,

⎧
⎨ 1 − (1 − q/2)10 w = 0,
−30 W = 0,
Ru = ru (W ) = PW (w) = (1 − q/2)10 w = 1, (5)
k − 30 W = 1, ⎩
0 otherwise.
Thus we can express the expected profit as

1
E [ru (W )] = PW (w) ru (w) (6)
w=0
= PW (0) (−30) + PW (1) (k − 30) (7)
= (1 − (1 − q/2) )(−30) + (1 − q/2) (k − 30) = (0.95) k − 30
10 10 10
(8)
To determine which implementation generates the most profit, we solve E[Ru ] ≥ E[Rs ], yielding
k ≥ 20/[(0.95)10 − (0.9)10 ] = 80.21. So for k < $80.21 using all standard devices results in greater
56
revenue, while for k > $80.21 more revenue will be generated by implementing all ultra-reliable
devices. That is, when the price commanded for a working circuit is sufficiently high, we should
build more-expensive higher-reliability circuits.
If you have read ahead to Section 2.9 and learned about conditional expected values, you might prefer
the following solution. If not, you might want to come back and review this alternate approach after
reading Section 2.9.
Let W denote the event that a circuit works. The circuit works and generates revenue of k
dollars if all of its 10 constituent devices work. For each implementation, standard or ultra-reliable,
let R denote the profit on a device. We can express the expected profit as
E [R] = P [W ] E [R|W ] + P [W c ] E [R|W c ] (9)
Let’s first consider the case when only standard devices are used. In this case, a circuit works
with probability P [W ] = (1 − q)10 . The profit made on a working device is k − 10 dollars while a
nonworking circuit has a profit of -10 dollars. That is, E[R|W ] = k − 10 and E[R|W c ] = −10. Of
course, a negative profit is actually a loss. Using Rs to denote the profit using standard circuits,
the expected profit is
E [Rs ] = (1 − q)10 (k − 10) + (1 − (1 − q)10 )(−10) = (0.9)10 k − 10 (10)
And for the ultra-reliable case, the circuit works with probability P [W ] = (1−q/2)10 . The profit per
working circuit is E[R|W ] = k − 30 dollars while the profit for a nonworking circuit is E[R|W c ] =
−30 dollars. The expected profit is
E [Ru ] = (1 − q/2)10 (k − 30) + (1 − (1 − q/2)10 )(−30) = (0.95)10 k − 30 (11)
Not surprisingly, we get the same answers for E[Ru ] and E[Rs ] as in the first solution by performing
essentially the same calculations. it should be apparent that indicator random variable W in the
first solution indicates the occurrence of the conditioning event W in the second solution. That is,
indicators are a way to track conditioning events.
46
(a) There are 6 equally likely winning combinations so that
1 1
q = 46 = ≈ 1.07 × 10−7 (1)
6
9,366,819
(b) Assuming each ticket is chosen randomly, each of the 2n − 1 other tickets is independently a
winner with probability q. The number of other winning tickets Kn has the binomial PMF
2n−1 k
k q (1 − q)2n−1−k k = 0, 1, . . . , 2n − 1
PKn (k) = (2)
0 otherwise
(c) Since there are Kn + 1 winning tickets in all, the value of your winning ticket is Wn =
n/(Kn + 1) which has mean
1
E [Wn ] = nE (3)
Kn + 1
57
Calculating the expected value
1
2n−1
1
E = PKn (k) (4)
Kn + 1 k+1
k=0
is fairly complicated. The trick is to express the sum in terms of the sum of a binomial
PMF.
2n−1
1
1 (2n − 1)!
E = q k (1 − q)2n−1−k (5)
Kn + 1 k + 1 k!(2n − 1 − k)!
k=0
1
2n−1
(2n)!
= q k (1 − q)2n−(k+1) (6)
2n (k + 1)!(2n − (k + 1))!
k=0
By factoring out 1/q, we obtain

2n−1
1 1 2n
E = q k+1 (1 − q)2n−(k+1) (7)
Kn + 1 2nq k+1
k=0
2n
1 2n j
= q (1 − q)2n−j (8)
2nq j
j=1

A
We observe that the above sum labeled A is the sum of a binomial PMF for 2n trials and
success probability q over all possible values except j = 0. Thus

2n 0
A=1− q (1 − q)2n−0 = 1 − (1 − q)2n (9)
0
This implies
1 1 − (1 − q)2n
E = (10)
Kn + 1 2nq
Our expected return on a winning ticket is

1 1 − (1 − q)2n
E [Wn ] = nE = (11)
Kn + 1 2q
Note that when nq 1, we can use the approximation that (1 − q)2n ≈ 1 − 2nq to show that
1 − (1 − 2nq)
E [Wn ] ≈ =n (nq 1) (12)
2q
However, in the limit as the value of the prize n approaches infinity, we have
1
lim E [Wn ] = ≈ 4.683 × 106 (13)
n→∞ 2q
That is, as the pot grows to infinity, the expected return on a winning ticket doesn’t approach
infinity because there is a corresponding increase in the number of other winning tickets. If
it’s not clear how large n must be for this effect to be seen, consider the following table:
n 106 107 108
(14)
E [Wn ] 9.00 × 105 4.13 × 106 4.68 × 106
When the pot is $1 million, our expected return is $900,000. However, we see that when the
pot reaches $100 million, our expected return is very close to 1/(2q), less than $5 million!
58
46
(a) There are 6 equally likely winning combinations so that
1 1
q = 46 = ≈ 1.07 × 10−7 (1)
6
9,366,819
(b) Assuming each ticket is chosen randomly, each of the 2n − 1 other tickets is independently a
winner with probability q. The number of other winning tickets Kn has the binomial PMF
2n−1 k
k q (1 − q)2n−1−k k = 0, 1, . . . , 2n − 1
PKn (k) = (2)
0 otherwise
Since the pot has n + r dollars, the expected amount that you win on your ticket is

n+r 1
E [V ] = 0(1 − q) + qE = q(n + r)E (3)
Kn + 1 Kn + 1
Note that E[1/Kn + 1] was also evaluated in Problem 2.7.8. For completeness, we repeat
those steps here.
2n−1
1
1 (2n − 1)!
E = q k (1 − q)2n−1−k (4)
Kn + 1 k + 1 k!(2n − 1 − k)!
k=0
1
2n−1
(2n)!
= q k (1 − q)2n−(k+1) (5)
2n (k + 1)!(2n − (k + 1))!
k=0
By factoring out 1/q, we obtain

2n−1
1 1 2n
E = q k+1 (1 − q)2n−(k+1) (6)
Kn + 1 2nq k+1
k=0
2n
1 2n j
= q (1 − q)2n−j (7)
2nq j
j=1

A
We observe that the above sum labeled A is the sum of a binomial PMF for 2n trials and
success probability q over all possible values except j = 0. Thus A = 1 − 2n
0 q (1 − q)
0 2n−0 ,
which implies
1 A 1 − (1 − q)2n
E = = (8)
Kn + 1 2nq 2nq
The expected value of your ticket is
q(n + r)[1 − (1 − q)2n ] 1 r

E [V ] = = 1+ [1 − (1 − q)2n ] (9)
2nq 2 n
Each ticket tends to be more valuable when the carryover pot r is large and the number of
new tickets sold, 2n, is small. For any fixed number n, corresponding to 2n tickets sold,
59
a sufficiently large pot r will guarantee that E[V ] > 1. For example if n = 107 , (20 million
tickets sold) then r
E [V ] = 0.44 1 + 7 (10)
10
If the carryover pot r is 30 million dollars, then E[V ] = 1.76. This suggests that buying a
one dollar ticket is a good idea. This is an unusual situation because normally a carryover
pot of 30 million dollars will result in far more than 20 million tickets being sold.
(c) So that we can use the results of the previous part, suppose there were 2n − 1 tickets sold
before you must make your decision. If you buy one of each possible ticket, you are guaranteed
to have one winning ticket. From the other 2n − 1 tickets, there will be Kn winners. The
total number of winning tickets will be Kn + 1. In the previous part we found that

1 1 − (1 − q)2n
E = (11)
Kn + 1 2nq
Let R denote the expected return from buying one of each possible ticket. The pot had
r dollars beforehand. The 2n − 1 other tickets are sold add n − 1/2 dollars to the pot.
Furthermore, you must buy 1/q tickets, adding 1/(2q) dollars to the pot. Since the cost of
the tickets is 1/q dollars, your expected profit

r + n − 1/2 + 1/(2q) 1
E [R] = E − (12)
Kn + 1 q

q(2r + 2n − 1) + 1 1 1
= E − (13)
2q Kn + 1 q
[q(2r + 2n − 1) + 1](1 − (1 − q)2n ) 1
= − (14)
4nq 2 q
For fixed n, sufficiently large r will make E[R] > 0. On the other hand, for fixed r,
limn→∞ E[R] = −1/(2q). That is, as n approaches infinity, your expected loss will be quite
large.

Given the following PMF ⎧
⎪
⎪ 0.2 n=0
⎨
0.7 n=1
PN (n) = (1)
⎪ 0.1
⎪ n=2
⎩
0 otherwise
(a) E[N ] = (0.2)0 + (0.7)1 + (0.1)2 = 0.9
(b) E[N 2 ] = (0.2)02 + (0.7)12 + (0.1)22 = 1.1
(c) Var[N ] = E[N 2 ] − E[N ]2 = 1.1 − (0.9)2 = 0.29

√
(d) σN = Var[N ] = 0.29
60
⎧
⎪
⎪ 1/4 y=1
⎨
1/4 y=2
PY (y) = (1)
⎪
⎪ 1/2 y=3
⎩
0 otherwise

E [Y ] = yPY (y) = 1(1/4) + 2(1/4) + 3(1/2) = 9/4 (2)
y
The expected value of Y 2 is

2
E Y2 = y PY (y) = 12 (1/4) + 22 (1/4) + 32 (1/2) = 23/4 (3)
y
The variance of Y is

Var[Y ] = E Y 2 − (E [Y ])2 = 23/4 − (9/4)2 = 11/16 (4)

⎧
⎪
⎪ 0.2 x = −1
⎨
0.5 x=0
PX (x) = (1)
⎪
⎪ 0.3 x=1
⎩
0 otherwise

E [X] = xPX (x) = (−1)(0.2) + 0(0.5) + 1(0.3) = 0.1 (2)
x
The expected value of X 2 is

2
E X2 = x PX (x) = (−1)2 (0.2) + 02 (0.5) + 12 (0.3) = 0.5 (3)
x
The variance of X is

Var[X] = E X 2 − (E [X])2 = 0.5 − (0.1)2 = 0.49 (4)

⎧
⎪
⎪ 0.4 x = −3
⎨
0.4 x=5
PX (x) = (1)
⎪
⎪ 0.2 x=7
⎩
0 otherwise
61

E [X] = xPX (x) = −3(0.4) + 5(0.4) + 7(0.2) = 2.2 (2)
x

2
E X2 = x PX (x) = (−3)2 (0.4) + 52 (0.4) + 72 (0.2) = 23.4 (3)
x

Var[X] = E X 2 − (E [X])2 = 23.4 − (2.2)2 = 18.56 (4)
(a) The expected value of X is

4
4 1 4 1 4 1 4 1 4 1
E [X] = xPX (x) = 0 4
+1 4
+2 4
+3 4
+4 (1)
0 2 1 2 2 2 3 2 4 24
x=0
= [4 + 12 + 12 + 4]/24 = 2 (2)

2
4
2 4 12 2 4 1 2 4 1 2 4 1 2 4 1
E X = x PX (x) = 0 +1 +2 +3 +4 (3)
0 24 1 24 2 24 3 24 4 24
x=0
= [4 + 24 + 36 + 16]/24 = 5 (4)
Var[X] = E X 2 − (E [X])2 = 5 − 22 = 1 (5)

Thus, X has standard deviation σX = Var[X] = 1.
(b) The probability that X is within one standard deviation of its expected value is
P [µX − σX ≤ X ≤ µX + σX ] = P [2 − 1 ≤ X ≤ 2 + 1] = P [1 ≤ X ≤ 3] (6)
This calculation is easy using the PMF of X.
P [1 ≤ X ≤ 3] = PX (1) + PX (2) + PX (3) = 7/8 (7)

5
E [X] = xPX (x) (1)
x=0

5 1 5 1 5 1 5 1 5 1 5 1
=0 5
+1 5
+2 5
+3 5
+4 5
+5 (2)
0 2 1 2 2 2 3 2 4 2 5 25
= [5 + 20 + 30 + 20 + 5]/25 = 5/2 (3)
62

5
2
E X = x2 PX (x) (4)
x=0

5 1
2 2 5 1 2 5 1 2 5 1 2 5 1 2 5 1
=0 +1 +2 +3 +4 +5 (5)
0 25 1 25 2 25 3 25 4 25 5 25
= [5 + 40 + 90 + 80 + 25]/25 = 240/32 = 15/2 (6)

Var[X] = E X 2 − (E [X])2 = 15/2 − 25/4 = 5/4 (7)

By taking the square root of the variance, the standard deviation of X is σX = 5/4 ≈ 1.12.
(b) The probability that X is within one standard deviation of its mean is
P [µX − σX ≤ X ≤ µX + σX ] = P [2.5 − 1.12 ≤ X ≤ 2.5 + 1.12] (8)

= P [1.38 ≤ X ≤ 3.62] (9)
= P [2 ≤ X ≤ 3] (10)
By summing the PMF over the desired range, we obtain
P [2 ≤ X ≤ 3] = PX (2) + PX (3) = 10/32 + 10/32 = 5/8 (11)

For Y = aX + b, we wish to show that Var[Y ] = a2 Var[X]. We begin by noting that Theorem 2.12
says that E[aX + b] = aE[X] + b. Hence, by the definition of variance.

Var [Y ] = E (aX + b − (aE [X] + b))2 (1)
2
= E a (X − E [X])2 (2)

= a E (X − E [X])
2 2
(3)
Since E[(X − E[X])2 ] = Var[X], the assertion is proved.

Given the following description of the random variable Y ,
1
Y = (X − µX ) (1)
σx
we can use the linearity property of the expectation operator to find the mean value
E [X − µX ] E [X] − E [X]
E [Y ] = = =0 (2)
σX σX
Using the fact that Var[aX + b] = a2 Var[X], the variance of Y is found to be
1
Var [Y ] = 2 Var [X] = 1 (3)
σX
63
With our measure of jitter being σT , and the fact that T = 2X − 1, we can express the jitter as a
function of q by realizing that
4q
Var[T ] = 4 Var[X] = (1)
(1 − q)2
Therefore, our maximum permitted jitter is
√
2 q
σT = = 2 msec (2)
(1 − q)
Solving for q yields q 2 − 3q + 1 = 0. By solving this quadratic equation, we obtain
√
3± 5 √
q= = 3/2 ± 5/2 (3)
2
√
Since q must be a value between 0 and 1, we know that a value of q = 3/2 − 5/2 ≈ 0.382 will
ensure a jitter of at most 2 milliseconds.

We wish to minimize the function
e(x̂) = E (X − x̂)2 (1)
with respect to x̂. We can expand the square and take the expectation while treating x̂ as a
constant. This yields

e(x̂) = E X 2 − 2x̂X + x̂2 = E X 2 − 2x̂E [X] + x̂2 (2)
Solving for the value of x̂ that makes the derivative de(x̂)/dx̂ equal to zero results in the value of
x̂ that minimizes e(x̂). Note that when we take the derivative with respect to x̂, both E[X 2 ] and
E[X] are simply constants.
d 2
E X − 2x̂E [X] + x̂2 = 2E [X] − 2x̂ = 0 (3)
dx̂
Hence we see that x̂ = E[X]. In the sense of mean squared error, the best guess for a random
variable is the mean value. In Chapter 9 this idea is extended to develop minimum mean squared
error estimation.

The PMF of K is the Poisson PMF

λk e−λ /k! k = 0, 1, . . .
PK (k) = (1)
0 otherwise
The mean of K is
∞
∞

λk e−λ λk−1 e−λ
E [K] = k =λ =λ (2)
k! (k − 1)!
k=0 k=1
To find E[K 2 ], we use the hint and first find

∞
∞
λk e−λ λk e−λ
E [K(K − 1)] = k(k − 1) = (3)
k! (k − 2)!
k=0 k=2
64
By factoring out λ2 and substituting j = k − 2, we obtain
∞
λj e−λ
E [K(K − 1)] = λ2 = λ2 (4)
j!
j=0

1
The above sum equals 1 because it is the sum of a Poisson PMF over all possible values. Since
E[K] = λ, the variance of K is

Var[K] = E K 2 − (E [K])2 (5)
= E [K(K − 1)] + E [K] − (E [K])2 (6)
= λ2 + λ − λ2 = λ (7)

The standard deviation can be expressed as

σD = Var[D] = E [D2 ] − E [D]2 (1)
where
2
4
E D = d2 PD (d) = 0.2 + 1.6 + 2.7 + 1.6 = 6.1 (2)
d=1
So finally we have √
σD = 6.1 − 2.32 = 0.81 = 0.9 (3)

⎧
⎪
⎪ 1/4 y=1
⎨
1/4 y=2
PY (y) = (1)
⎪
⎪ 1/2 y=3
⎩
0 otherwise
The probability of the event B = {Y < 3} is P [B] = 1 − P [Y = 3] = 1/2. From Theorem 2.17, the
conditional PMF of Y given B is
⎧
PY (y) ⎨ 1/2 y = 1
P [B] y ∈ B
PY |B (y) = = 1/2 y = 2 (2)
0 otherwise ⎩
0 otherwise
The conditional first and second moments of Y are

E [Y |B] = yPY |B (y) = 1(1/2) + 2(1/2) = 3/2 (3)
y
2
E Y 2 |B = y PY |B (y) = 12 (1/2) + 22 (1/2) = 5/2 (4)
y
The conditional variance of Y is

Var[Y |B] = E Y 2 |B − (E [Y |B])2 = 5/2 − 9/4 = 1/4 (5)
65
⎧
⎪
⎪ 0.2 x = −1
⎨
0.5 x=0
PX (x) = (1)
⎪
⎪ 0.3 x=1
⎩
0 otherwise
The event B = {|X| > 0} has probability P [B] = P [X = 0] = 0.5. From Theorem 2.17, the
conditional PMF of X given B is
⎧
PX (x) ⎨ 0.4 x = −1
P [B] x∈B
PX|B (x) = = 0.6 x=1 (2)
0 otherwise ⎩
0 otherwise
The conditional first and second moments of X are

E [X|B] = xPX|B (x) = (−1)(0.4) + 1(0.6) = 0.2 (3)
x
2
E X 2 |B = x PX|B (x) = (−1)2 (0.4) + 12 (0.6) = 1 (4)
x
The conditional variance of X is

Var[X|B] = E X 2 |B − (E [X|B])2 = 1 − (0.2)2 = 0.96 (5)

⎧
⎪
⎪ 0.4 x = −3
⎨
0.4 x=5
PX (x) = (1)
⎪
⎪ 0.2 x=7
⎩
0 otherwise
The event B = {X > 0} has probability P [B] = PX (5) + PX (7) = 0.6. From Theorem 2.17, the
conditional PMF of X given B is
⎧
PX (x) ⎨ 2/3 x=5
P [B] x ∈ B
PX|B (x) = = 1/3 x=7 (2)
0 otherwise ⎩
0 otherwise

E [X|B] = xPX|B (x) = 5(2/3) + 7(1/3) = 17/3 (3)
x
2
E X 2 |B = x PX|B (x) = 52 (2/3) + 72 (1/3) = 33 (4)
x

Var[X|B] = E X 2 |B − (E [X|B])2 = 33 − (17/3)2 = 8/9 (5)
66
The event B = {X = 0} has probability P [B] = 1 − P [X = 0] = 15/16. The conditional PMF of
X given B is
4 1
PX (x)
P [B] x∈B x 15 x = 1, 2, 3, 4
PX|B (x) = = (1)
0 otherwise 0 otherwise

4
4 1 4 1 4 1 4 1
E [X|B] = xPX|B (x) = 1 2 +3 +4 (2)
1 15 2 15 3 15 4 15
x=1
= [4 + 12 + 12 + 4]/15 = 32/15 (3)

4
2 4 1 2 4 1 2 4 1 2 4 1
E X 2 |B = 2
x PX|B (x) = 1 2 +3 +4 (4)
1 15 2 15 3 15 4 15
x=1
= [4 + 24 + 36 + 16]/15 = 80/15 (5)

Var[X|B] = E X 2 |B − (E [X|B])2 = 80/15 − (32/15)2 = 176/225 ≈ 0.782 (6)

The probability of the event B is
P [B] = P [X ≥ µX ] = P [X ≥ 3] = PX (3) + PX (4) + PX (5) (1)

5 5 5
+ 4 + 5
= 3 = 21/32 (2)
32
The conditional PMF of X given B is
5
PX (x)
P [B] x∈B 1
x 21 x = 3, 4, 5
PX|B (x) = = (3)

5
5 1 5 1 5 1
E [X|B] = xPX|B (x) = 3 +4 +5 (4)
3 21 4 21 5 21
x=3
= [30 + 20 + 5]/21 = 55/21 (5)

2 5
2 5 1 2 5 1 2 5 1
E X |B = 2
x PX|B (x) = 3 +4 +5 (6)
3 21 4 21 5 21
x=3
= [90 + 80 + 25]/21 = 195/21 = 65/7 (7)

Var[X|B] = E X 2 |B − (E [X|B])2 = 65/7 − (55/21)2 = 1070/441 = 2.43 (8)
67
(a) Consider each circuit test as a Bernoulli trial such that a failed circuit is called a success.
The number of trials until the first success (i.e. a failed circuit) has the geometric PMF

(1 − p)n−1 p n = 1, 2, . . .
PN (n) = (1)
0 otherwise
(b) The probability there are at least 20 tests is

∞

P [B] = P [N ≥ 20] = PN (n) = (1 − p)19 (2)
n=20
Note that (1 − p)19 is just the probability that the first 19 circuits pass the test, which is
what we would expect since there must be at least 20 tests if the first 19 circuits pass. The
conditional PMF of N given B is

PN (n)
P [B] n∈B (1 − p)n−20 p n = 20, 21, . . .
PN |B (n) = = (3)
(c) Given the event B, the conditional expectation of N is
∞

E [N |B] = nPN |B (n) = n(1 − p)n−20 p (4)
n n=20
Making the substitution j = n − 19 yields

∞

E [N |B] = (j + 19)(1 − p)j−1 p = 1/p + 19 (5)
j=1
We see that in the above sum, we effectively have the expected value of J + 19 where J is
geometric random variable with parameter p. This is not surprising since the N ≥ 20 iff we
observed 19 successful tests. After 19 successful tests, the number of additional tests needed
to find the first failure is still a geometric random variable with mean 1/p.
(a) The PMF of M , the number of miles run on an arbitrary day is

q(1 − q)m m = 0, 1, . . .
PM (m) = (1)
0 otherwise
And we can see that the probability that M > 0, is
P [M > 0] = 1 − P [M = 0] = 1 − q (2)
68
(b) The probability that we run a marathon on any particular day is the probability that M ≥ 26.
∞

r = P [M ≥ 26] = q(1 − q)m = (1 − q)26 (3)
m=26
(c) We run a marathon on each day with probability equal to r, and we do not run a marathon
with probability 1 − r. Therefore in a year we have 365 tests of our jogging resolve, and thus
365 chances to run a marathon. So the PMF of the number of marathons run in a year, J,
can be expressed as
365 j
j r (1 − r)
365−j j = 0, 1, . . . , 365
PJ (j) = (4)
0 otherwise
(d) The random variable K is defined as the number of miles we run above that required for a
marathon, K = M − 26. Given the event, A, that we have run a marathon, we wish to know
how many miles in excess of 26 we in fact ran. So we want to know the conditional PMF
PK|A (k).
P [K = k, A] P [M = 26 + k]
PK|A (k) = = (5)
P [A] P [A]
Since P [A] = r, for k = 0, 1, . . .,
(1 − q)26+k q
PK|A (k) = = (1 − q)k q (6)
(1 − q)26
The complete expression of for the conditional PMF of K is

(1 − q)k q k = 0, 1, . . .
PK|A (k) = (7)
0 otherwise

Recall that the PMF of the number of pages in a fax is
⎧
⎨ 0.15 x = 1, 2, 3, 4
PX (x) = 0.1 x = 5, 6, 7, 8 (1)
⎩
0 otherwise
(a) The event that a fax was sent to machine A can be expressed mathematically as the event
that the number of pages X is an even number. Similarly, the event that a fax was sent
to B is the event that X is an odd number. Since SX = {1, 2, . . . , 8}, we define the set
A = {2, 4, 6, 8}. Using this definition for A, we have that the event that a fax is sent to A is
equivalent to the event X ∈ A. The event A has probability
P [A] = PX (2) + PX (4) + PX (6) + PX (8) = 0.5 (2)
Given the event A, the conditional PMF of X is

⎧
PX (x) ⎨ 0.3 x = 2, 4
P [A] x∈A
PX|A (x) = = 0.2 x = 6, 8 (3)
0 otherwise ⎩
0 otherwise
69
The conditional first and second moments of X given A is

E [X|A] = xPX|A (x) = 2(0.3) + 4(0.3) + 6(0.2) + 8(0.2) = 4.6 (4)
x
2
E X 2 |A = x PX|A (x) = 4(0.3) + 16(0.3) + 36(0.2) + 64(0.2) = 26 (5)
x
The conditional variance and standard deviation are

Var[X|A] = E X 2 |A − (E [X|A])2 = 26 − (4.6)2 = 4.84 (6)

σX|A = Var[X|A] = 2.2 (7)
(b) Let the event B denote the event that the fax was sent to B and that the fax had no more
than 6 pages. Hence, the event B = {1, 3, 5} has probability

P B = PX (1) + PX (3) + PX (5) = 0.4 (8)
The conditional PMF of X given B is
⎧
⎨ 3/8 x = 1, 3
PX (x)
P [B ] x∈ B
PX|B (x) = = 1/4 x = 5 (9)
0 otherwise ⎩
0 otherwise
Given the event B , the conditional first and second moments are

E X|B = xPX|B (x) = 1(3/8) + 3(3/8) + 5(1/4)+ = 11/4 (10)
x
2
E X 2 |B = x PX|B (x) = 1(3/8) + 9(3/8) + 25(1/4) = 10 (11)
x
The conditional variance and standard deviation are

Var[X|B ] = E X 2 |B − (E X|B )2 = 10 − (11/4)2 = 39/16 (12)
√
σX|B = Var[X|B ] = 39/4 ≈ 1.56 (13)

For a binomial (n, p) random variable X, the solution in terms of math is
√
n

P [E2 ] = PX x2 (1)
x=0
In terms of Matlab, the efficient solution is to generate the vector of perfect squares x =
[0 1 4 9 16 ...] and then to pass that vector to the binomialpmf.m. In this case, the val-
ues of the binomial PMF are calculated only once. Here is the code:
function q=perfectbinomial(n,p);
i=0:floor(sqrt(n));
x=i.^2;
q=sum(binomialpmf(n,p,x));
For a binomial (100, 0.5) random variable X, the probability X is a perfect square is
>> perfectbinomial(100,0.5)
ans =
0.0811
70
The random variable X given in Example 2.29 is just a finite random variable. We can generate
random samples using the finiterv function. The code is
function x=faxlength8(m);
sx=1:8;
p=[0.15*ones(1,4) 0.1*ones(1,4)];
x=finiterv(sx,p,m);

First we use faxlength8 from Problem 2.10.2 to generate m samples of the faqx length X. Next
we convert that to m samples of the fax cost Y . Summing these samples and dividing by m, we
obtain the average cost of m samples. Here is the code:
function y=avgfax(m);
x=faxlength8(m);
yy=cumsum([10 9 8 7 6]);
yy=[yy 50 50 50];
y=sum(yy(x))/m;
Each time we perform the experiment of executing the function avgfax, we generate m random
1
m
samples of X, and m corresponding samples of Y . The sum Y = m i=1 Yi is random. For m = 10,
four samples of Y are
>> [avgfax(10) avgfax(10) avgfax(10) avgfax(10)]
ans =
31.9000 31.2000 29.6000 34.1000
>>
For m = 100, the results are arguably more consistent:
ans =
34.5300 33.3000 29.8100 33.6900
>>
Finally, for m = 1000, we obtain results reasonably close to E[Y ]:
ans =
32.1740 31.8920 33.1890 32.8250
>>
In Chapter 7, we will develop techniques to show how Y converges to E[Y ] as m → ∞.

Suppose Xn is a Zipf (n, α = 1) random variable and thus has PMF

c(n)/x x = 1, 2, . . . , n
PX (x) = (1)
0 otherwise
The problem asks us to find the smallest value of k such that P [Xn ≤ k] ≥ 0.75. That is, if the
server caches the k most popular files, then with P [Xn ≤ k] the request is for one of the k cached
71
files. First, we might as well solve this problem for any probability p rather than just p = 0.75.
Thus, in math terms, we are looking for

k = min k |P Xn ≤ k ≥ p . (2)
What makes the Zipf distribution hard to analyze is that there is no closed form expression for
! n "−1
1
c(n) = . (3)
x
x=1
Thus, we use Matlab to grind through the calculations. The following simple program generates
the Zipf distributions and returns the correct value of k.
function k=zipfcache(n,p);
%Usage: k=zipfcache(n,p);
%for the Zipf (n,alpha=1) distribution, returns the smallest k
%such that the first k items have total probability p
pmf=1./(1:n);
pmf=pmf/sum(pmf); %normalize to sum to 1
cdf=cumsum(pmf);
k=1+sum(cdf<=p);
The program zipfcache generalizes 0.75 to be the probability p. Although this program is suffi-
cient, the problem asks us to find k for all values of n from 1 to 103 !. One way to do this is to call
zipfcache a thousand times to find k for each value of n. A better way is to use the properties of
the Zipf PDF. In particular,

k
1 c(n)

P Xn ≤ k = c(n) = (4)
x c(k )
x=1
Thus we wish to find

# #
c(n) 1 p
k = min k | ≥ p = min k | ≥ . (5)
c(k ) c(k ) c(n)
Note that the definition of k implies that
1 p
< , k = 1, . . . , k − 1. (6)
c(k ) c(n)
Using the notation |A| to denote the number of elements in the set A, we can write
$ #$
$ 1 p $
k = 1 + $$ k | < $ (7)
c(k ) c(n) $
This is the basis for a very short Matlab program:
function k=zipfcacheall(n,p);
%Usage: k=zipfcacheall(n,p);
%returns vector k such that the first
%k(m) items have total probability >= p
%for the Zipf(m,1) distribution.
c=1./cumsum(1./(1:n));
k=1+countless(1./c,p./c);
72
Note that zipfcacheall uses a short Matlab program countless.m that is almost the same as
count.m introduced in Example 2.47. If n=countless(x,y), then n(i) is the number of elements
of x that are strictly less than y(i) while count returns the number of elements less than or equal
to y(i).
In any case, the commands
k=zipfcacheall(1000,0.75);
plot(1:1000,k);
is sufficient to produce this figure of k as a function of m:
200
150
100
k
50
0
0 100 200 300 400 500 600 700 800 900 1000
n
We see in the figure that the number of files that must be cached grows slowly with the total
number of files n.
Finally, we make one last observation. It is generally desirable for Matlab to execute opera-
tions in parallel. The program zipfcacheall generally will run faster than n calls to zipfcache.
However, to do its counting all at once, countless generates and n × n array. When n is not too
large, say n ≤ 1000, the resulting array with n2 = 1,000,000 elements fits in memory. For much
large values of n, say n = 106 (as was proposed in the original printing of this edition of the text,
countless will cause an “out of memory” error.

We use poissonrv.m to generate random samples of a Poisson (α = 5) random variable. To
compare the Poisson PMF against the output of poissonrv, relative frequencies are calculated
using the hist function. The following code plots the relative frequency against the PMF.
function diff=poissontest(alpha,m)
x=poissonrv(alpha,m);
xr=0:ceil(3*alpha);
pxsample=hist(x,xr)/m;
pxsample=pxsample(:);
%pxsample=(countequal(x,xr)/m);
px=poissonpmf(alpha,xr);
plot(xr,pxsample,xr,px);
diff=sum((pxsample-px).^2);
For m = 100, 1000, 10000, here are sample plots comparing the PMF and the relative frequency.
The plots show reasonable agreement for m = 10000 samples.
73
0.2 0.2
0.1 0.1
0 0
0 5 10 15 0 5 10 15
(a) m = 100 (b) m = 100
0.2 0.2
0.1 0.1
0 0
0 5 10 15 0 5 10 15
(a) m = 1000 (b) m = 1000
0.2 0.2
0.1 0.1
0 0
0 5 10 15 0 5 10 15
(a) m = 10,000 (b) m = 10,000

We can compare the binomial and Poisson PMFs for (n, p) = (100, 0.1) using the following Matlab
code:
x=0:20;
p=poissonpmf(100,x);
b=binomialpmf(100,0.1,x);
plot(x,p,x,b);
For (n, p) = (10, 1), the binomial PMF has no randomness. For (n, p) = (100, 0.1), the approxima-
tion is reasonable:
1 0.2
0.8
0.15
0.6
0.1
0.4
0.05
0.2
0 0
0 5 10 15 20 0 5 10 15 20
(a) n = 10, p = 1 (b) n = 100, p = 0.1
Finally, for (n, p) = (1000, 0.01), and (n, p) = (10000, 0.001), the approximation is very good:
74
0.2 0.2
0.15 0.15
0.1 0.1
0.05 0.05
0 0
0 5 10 15 20 0 5 10 15 20
(a) n = 1000, p = 0.01 (b) n = 10000, p = 0.001

Following the Random Sample algorithm, we generate a sample value R = rand(1) and then we
find k ∗ such that
FK (k ∗ − 1) < R < FK (k ∗ ) . (1)
From Problem 2.4.4, we know for integers k ≥ 1 that geometric (p) random variable K has CDF
FK (k) = 1 − (1 − p)k . Thus,
∗ −1 ∗
1 − (1 − p)k < R ≤ 1 − (1 − p)k . (2)
Subtracting 1 from each side and then multiplying through by −1 (which reverses the inequalities),
we obtain
∗ ∗
(1 − p)k −1 > 1 − R ≥ (1 − p)k . (3)
Next we take the logarithm of each side. Since logarithms are monotonic functions, we have
(k ∗ − 1) ln(1 − p) > ln(1 − R) ≥ k ∗ ln(1 − p). (4)
Since 0 < p < 1, we have that ln(1 − p) < 0. Thus dividing through by ln(1 − p) reverses the
inequalities, yielding
ln(1 − R)
k∗ − 1 > ≤ k∗ . (5)
ln(1 − p)
Since k ∗ is an integer, it must be the smallest integer greater than or equal to ln(1 − R)/ ln(1 − p).
That is, following the last step of the random sample algorithm,
% &
∗ ln(1 − R)
K=k = (6)
ln(1 − p)
The Matlab algorithm that implements this operation is quite simple:
function x=geometricrv(p,m)
%Usage: x=geometricrv(p,m)
% returns m samples of a geometric (p) rv
r=rand(m,1);
x=ceil(log(1-r)/log(1-p));
75
For the PC version of Matlab employed for this test, poissonpmf(n,n) reported Inf for n =
n∗ = 714. The problem with the poissonpmf function in Example 2.44 is that the cumulative
product that calculated nk /k! can have an overflow. Following the hint, we can write an alternate
poissonpmf function as follows:
function pmf=poissonpmf(alpha,x)
%Poisson (alpha) rv X,
%out=vector pmf: pmf(i)=P[X=x(i)]
x=x(:);
if (alpha==0)
pmf=1.0*(x==0);
else
k=(1:ceil(max(x)))’;
logfacts =cumsum(log(k));
pb=exp([-alpha; ...
-alpha+ (k*log(alpha))-logfacts]);
okx=(x>=0).*(x==floor(x));
x=okx.*x;
pmf=okx.*pb(x+1);
end
%pmf(i)=0 for zero-prob x(i)
By summing logarithms, the intermediate terms are much less likely to overflow.
76

The CDF of X is ⎧
⎨ 0 x < −1
FX (x) = (x + 1)/2 −1 ≤ x < 1 (1)
⎩
1 x≥1
Each question can be answered by expressing the requested probability in terms of FX (x).
(a)
P [X > 1/2] = 1 − P [X ≤ 1/2] = 1 − FX (1/2) = 1 − 3/4 = 1/4 (2)
(b) This is a little trickier than it should be. Being careful, we can write
P [−1/2 ≤ X < 3/4] = P [−1/2 < X ≤ 3/4] + P [X = −1/2] − P [X = 3/4] (3)
Since the CDF of X is a continuous function, the probability that X takes on any specific
value is zero. This implies P [X = 3/4] = 0 and P [X = −1/2] = 0. (If this is not clear at this
point, it will become clear in Section 3.6.) Thus,
P [−1/2 ≤ X < 3/4] = P [−1/2 < X ≤ 3/4] = FX (3/4) − FX (−1/2) = 5/8 (4)
(c)
P [|X| ≤ 1/2] = P [−1/2 ≤ X ≤ 1/2] = P [X ≤ 1/2] − P [X < −1/2] (5)
Note that P [X ≤ 1/2] = FX (1/2) = 3/4. Since the probability that P [X = −1/2] = 0,
P [X < −1/2] = P [X ≤ 1/2]. Hence P [X < −1/2] = FX (−1/2) = 1/4. This implies
P [|X| ≤ 1/2] = P [X ≤ 1/2] − P [X < −1/2] = 3/4 − 1/4 = 1/2 (6)
(d) Since FX (1) = 1, we must have a ≤ 1. For a ≤ 1, we need to satisfy

a+1
P [X ≤ a] = FX (a) = = 0.8 (7)
2
Thus a = 0.6.

The CDF of V was given to be
⎧
⎨ 0 v < −5
FV (v) = c(v + 5) −5 ≤ v < 7
2 (1)
⎩
1 v≥7
(a) For V to be a continuous random variable, FV (v) must be a continuous function. This occurs
if we choose c such that FV (v) doesn’t have a discontinuity at v = 7. We meet this requirement
if c(7 + 5)2 = 1. This implies c = 1/144.
77
(b)
P [V > 4] = 1 − P [V ≤ 4] = 1 − FV (4) = 1 − 81/144 = 63/144 (2)
(c)
P [−3 < V ≤ 0] = FV (0) − FV (−3) = 25/144 − 4/144 = 21/144 (3)
(d) Since 0 ≤ FV (v) ≤ 1 and since FV (v) is a nondecreasing function, it must be that −5 ≤ a ≤ 7.
In this range,
P [V > a] = 1 − FV (a) = 1 − (a + 5)2 /144 = 2/3 (4)
√
The unique solution in the range −5 ≤ a ≤ 7 is a = 4 3 − 5 = 1.928.

In this problem, the CDF of W is
⎧
⎪
⎪ 0 w < −5
⎪
⎪
⎨ (w + 5)/8 −5 ≤ w < −3
FW (w) = 1/4 −3 ≤ w < 3 (1)
⎪
⎪
⎪
⎪ 1/4 + 3(w − 3)/8 3≤w<5
⎩
1 w ≥ 5.
Each question can be answered directly from this CDF.
(a)
P [W ≤ 4] = FW (4) = 1/4 + 3/8 = 5/8. (2)
(b)
P [−2 < W ≤ 2] = FW (2) − FW (−2) = 1/4 − 1/4 = 0. (3)
(c)
P [W > 0] = 1 − P [W ≤ 0] = 1 − FW (0) = 3/4 (4)
(d) By inspection of FW (w), we observe that P [W ≤ a] = FW (a) = 1/2 for a in the range
3 ≤ a ≤ 5. In this range,
FW (a) = 1/4 + 3(a − 3)/8 = 1/2 (5)
This implies a = 11/3.
(a) By definition, nx is the smallest integer that is greater than or equal to nx. This implies
nx ≤ nx ≤ nx + 1.
(b) By part (a),
nx nx nx + 1
≤ ≤ (1)
n n n
That is,
nx 1
x≤ ≤x+ (2)
n n
This implies
nx 1
x ≤ lim ≤ lim x + = x (3)
n→∞ n n→∞ n
78
(c) In the same way, nx is the largest integer that is less than or equal to nx. This implies
nx − 1 ≤ nx ≤ nx. It follows that
nx − 1 nx nx
≤ ≤ (4)
n n n
That is,
1 nx
x− ≤ ≤x (5)
n n
This implies
1 nx
lim x − = x ≤ lim ≤x (6)
n→∞ n n→∞ n

cx 0 ≤ x ≤ 2
fX (x) = (1)
0 otherwise
(a) From the above PDF we can determine the value of c by integrating the PDF and setting it
equal to 1. ' 2
cx dx = 2c = 1 (2)
0
Therefore c = 1/2.
(1
(b) P [0 ≤ X ≤ 1] = 0 x
dx = 1/4
2
( 1/2
(c) P [−1/2 ≤ X ≤ 1/2] = 0 x2 dx = 1/16
(d) The CDF of X is found by integrating the PDF from 0 to x.
⎧
' x
⎨ 0 2 x<0
FX (x) = fX x dx = x /4 0≤x≤2 (3)
0 ⎩
1 x>2

From the CDF, we can find the PDF by direct differentiation. The CDF and correponding PDF
are
⎧
⎨ 0 x < −1
1/2 −1 ≤ x ≤ 1
FX (x) = (x + 1)/2 −1 ≤ x ≤ 1 fX (x) = (1)
⎩ 0 otherwise
1 x>1

We find the PDF by taking the derivative of FU (u) on each piece that FU (u) is defined. The CDF
and corresponding PDF of U are
⎧ ⎧
⎪
⎪ 0 u < −5 ⎪
⎪ 0 u < −5
⎪
⎪ ⎪
⎪
⎨ (u + 5)/8 −5 ≤ u < −3 ⎨ 1/8 −5 ≤ u < −3
FU (u) = 1/4 −3 ≤ u < 3 fU (u) = 0 −3 ≤ u < 3 (1)
⎪
⎪ ⎪
⎪
⎪
⎪ 1/4 + 3(u − 3)/8 3 ≤ u < 5 ⎪
⎪ 3/8 3 ≤ u < 5
⎩ ⎩
1 u ≥ 5. 0 u ≥ 5.
79
For x < 0, FX (x) = 0. For x ≥ 0,
' x
FX (x) = fX (y) dy (1)
'
0
x
2 2
= a2 ye−a y /2 dy (2)
0
$x
2 2 $ 2 2
= −e−a y /2 $ = 1 − e−a x /2 (3)
0
A complete expression for the CDF of X is

0 x<0
FX (x) = −a 2 x2 /2 (4)
1−e x≥0

ax2 + bx 0 ≤ x ≤ 1
fX (x) = (1)
0 otherwise
First, we note that a and b must be chosen such that the above PDF integrates to 1.
' 1
(ax2 + bx) dx = a/3 + b/2 = 1 (2)
0
Hence, b = 2 − 2a/3 and our PDF becomes
fX (x) = x(ax + 2 − 2a/3) (3)
For the PDF to be non-negative for x ∈ [0, 1], we must have ax + 2 − 2a/3 ≥ 0 for all x ∈ [0, 1].
This requirement can be written as
a(2/3 − x) ≤ 2 (0 ≤ x ≤ 1) (4)
For x = 2/3, the requirement holds for all a. However, the problem is tricky because we must
consider the cases 0 ≤ x < 2/3 and 2/3 < x ≤ 1 separately because of the sign change of the
inequality. When 0 ≤ x < 2/3, we have 2/3 − x > 0 and the requirement is most stringent at
x = 0 where we require 2a/3 ≤ 2 or a ≤ 3. When 2/3 < x ≤ 1, we can write the constraint as
a(x − 2/3) ≥ −2. In this case, the constraint is most stringent at x = 1, where we must have
a/3 ≥ −2 or a ≥ −6. Thus a complete expression for our requirements are
−6 ≤ a ≤ 3 b = 2 − 2a/3 (5)
As we see in the following plot, the shape of the PDF fX (x) varies greatly with the value of a.
80
3
a=−6
a=−3
2.5 a=0
a=3
1.5
0.5
0
0 0.2 0.4 0.6 0.8 1

1/4 −1 ≤ x ≤ 3
fX (x) = (1)
0 otherwise
We recognize that X is a uniform random variable from [-1,3].
(3+1)2
(a) E[X] = 1 and Var[X] = 12 = 4/3.
(b) The new random variable Y is defined as Y = h(X) = X 2 . Therefore
h(E [X]) = h(1) = 1 (2)
and
E [h(X)] = E X 2 = Var [X] + E [X]2 = 4/3 + 1 = 7/3 (3)
(c) Finally

E [Y ] = E [h(X)] = E X 2 = 7/3 (4)
' 3 4
4 2 2 x 49 61 49
Var [Y ] = E X − E X = dx − = − (5)
−1 4 9 5 9
(a) Since the PDF is uniform over [1,9]
1+9 (9 − 1)2 16
E [X] = =5 Var [X] = = (1)
2 12 3
√
(b) Define h(X) = 1/ X then
√
h(E [X]) = 1/ 5 (2)
' 9 −1/2
x
E [h(X)] = dx = 1/2 (3)
1 8
81
(c)
E [Y ] = E [h(X)] = 1/2 (4)

'
2 9
x−1 ln 9
Var [Y ] = E Y − (E [Y ]) =
2
dx − E [X]2 = − 1/4 (5)
1 8 8

The CDF of X is ⎧
⎨ 0 x<0
FX (x) = x/2 0 ≤ x < 2 (1)
⎩
1 x≥2
(a) To find E[X], we first find the PDF by differentiating the above CDF.

1/2 0 ≤ x ≤ 2
fX (x) = (2)
0 otherwise
The expected value is then ' 2

x
E [X] = dx = 1 (3)
0 2
(b)
'
x2 2
E X2 = dx = 8/3 (4)
0 2

Var[X] = E X 2 − E [X]2 = 8/3 − 1 = 5/3 (5)

We can find the expected value of X by direct integration of the given PDF.

y/2 0 ≤ y ≤ 2
fY (y) = (1)
0 otherwise
The expectation is ' 2

y2
E [Y ] = dy = 4/3 (2)
0 2
To find the variance, we first find the second moment
' 2 3
2 y
E Y = dy = 2. (3)
0 2
The variance is then Var[Y ] = E[Y 2 ] − E[Y ]2 = 2 − (4/3)2 = 2/9.

The CDF of Y is ⎧
⎨ 0 y < −1
FY (y) = (y + 1)/2 −1 ≤ y < 1 (1)
⎩
1 y≥1
82
(a) We can find the expected value of Y by first differentiating the above CDF to find the PDF

1/2 −1 ≤ y ≤ 1,
fY (y) = (2)
0 otherwise.
It follows that ' 1

E [Y ] = y/2 dy = 0. (3)
−1
(b)
'
y2 1
E Y2 = dy = 1/3 (4)
−1 2

Var[Y ] = E Y 2 − E [Y ]2 = 1/3 − 0 = 1/3 (5)

To evaluate the moments of V , we need the PDF fV (v), which we find by taking the derivative of
the CDF FV (v). The CDF and corresponding PDF of V are
⎧ ⎧
⎨ 0 v < −5 ⎨ 0 v < −5
FV (v) = (v + 5) /144 −5 ≤ v < 7
2 fV (v) = (v + 5)/72 −5 ≤ v < 7 (1)
⎩ ⎩
1 v≥7 0 v≥7
(a) The expected value of V is

' ∞ ' 7
1
E [V ] = vfV (v) dv = (v 2 + 5v) dv (2)
−∞ 72 −5
$7
1 v 3 5v 2 $$ 1 343 245 125 125
= + = + + − =3 (3)
72 3 2 $−5 72 3 2 3 2
(b) To find the variance, we first find the second moment

' ∞ ' 7
1
E V2 = 2
v fV (v) dv = (v 3 + 5v 2 ) dv (4)
−∞ 72 −5
4 $7
1 v 3
5v $ $
= + = 6719/432 = 15.55 (5)
72 4 3 $−5
The variance is Var[V ] = E[V 2 ] − (E[V ])2 = 2831/432 = 6.55.
(c) The third moment of V is

' ∞ ' 7
3
3 1
E V = v fV (v) dv = (v 4 + 5v 3 ) dv (6)
−∞ 72 −5
$7
1 v 5 5v 4 $$
= + = 86.2 (7)
72 5 4 $−5
83
To find the moments, we first find the PDF of U by taking the derivative of FU (u). The CDF and
corresponding PDF are
⎧ ⎧
⎪
⎪ 0 u < −5 ⎪
⎪ 0 u < −5
⎪
⎪ ⎪
⎪
⎨ (u + 5)/8 −5 ≤ u < −3 ⎨ 1/8 −5 ≤ u < −3
FU (u) = 1/4 −3 ≤ u < 3 fU (u) = 0 −3 ≤ u < 3 (1)
⎪
⎪ ⎪
⎪
⎪
⎪ 1/4 + 3(u − 3)/8 3≤u<5 ⎪
⎪ 3/8 3≤u<5
⎩ ⎩
1 u ≥ 5. 0 u ≥ 5.
(a) The expected value of U is

' ∞ ' −3 ' 5
u 3u
E [U ] = ufU (u) du = du + du (2)
−∞ −5 8 3 8
$−3 $
u2 $
$ 3u $$2 5
= + =2 (3)
16 $ −5 16 $3
(b) The second moment of U is

' ∞ ' −3 ' 5 2
u2 3u
E U2 u2 fU (u) du = du + du (4)
−∞ −5 8 3 8
$−3 $5
u3 $$ u3 $$
= + = 49/3 (5)
24 $−5 8 $3
The variance of U is Var[U ] = E[U 2 ] − (E[U ])2 = 37/3.
(c) Note that 2U = e(ln 2)U . This implies that

' '
1 (ln 2)u 2u
2 du = e(ln 2)u du =
u
e = (6)
ln 2 ln 2
The expected value of 2U is then

' ∞ ' −3 ' 5
U u 2u 3 · 2u
E 2 = 2 fU (u) du = du + du (7)
−∞ −5 8 3 8
$ $
2u $$−3 3 · 2u $$5 2307
= $ + $ = = 13.001 (8)
8 ln 2 −5 8 ln 2 256 ln 2
3

The Pareto (α, µ) random variable has PDF

(α/µ) (x/µ)−(α+1) x ≥ µ
fX (x) = (1)
0 otherwise
The nth moment is

' ∞ −(α+1) ' ∞ −(α−n+1)
n nα x n α x
E [X ] = x dx = µ dx (2)
µ µ µ µ µ µ
84
With the variable substitution y = x/µ, we obtain
' ∞
n
E [X ] = αµ n
y −(α−n+1) dy (3)
1
We see that E[X n ] < ∞ if and only if α − n + 1 > 1, or, equivalently, n < α. In this case,
$y=∞
n αµn $
−(α−n+1)+1 $
E [X ] = y $ (4)
−(α − n + 1) + 1 y=1
$
−αµn −(α−n) $$y=∞ αµn
= y $ = (5)
α−n y=1 α−n

The reflected power Y has an exponential (λ = 1/P0 ) PDF. From Theorem 3.8, E[Y ] = P0 . The
probability that an aircraft is correctly identified is
' ∞
1 −y/P0
P [Y > P0 ] = e dy = e−1 . (1)
P0 P 0
Fortunately, real radar systems offer better performance.
From Appendix A, we observe that an exponential PDF Y with parameter λ > 0 has PDF

λe−λy y ≥ 0
fY (y) = (1)
0 otherwise
In addition, the mean and variance of Y are
1 1
E [Y ] = Var[Y ] = (2)
λ λ2
(a) Since Var[Y ] = 25, we must have λ = 1/5.
(b) The expected value of Y is E[Y ] = 1/λ = 5.
(c)
' ∞ $∞
$
P [Y > 5] = fY (y) dy = −e−y/5 $ = e−1 (3)
5 5

From Appendix A, an Erlang random variable X with parameters λ > 0 and n has PDF
n n−1 −λx
λ x e /(n − 1)! x ≥ 0
fX (x) = (1)
0 otherwise
In addition, the mean and variance of X are
n n
E [X] = Var[X] = 2 (2)
λ λ
(a) Since λ = 1/3 and E[X] = n/λ = 15, we must have n = 5.
(b) Substituting the parameters n = 5 and λ = 1/3 into the given PDF, we obtain

(1/3)5 x4 e−x/3 /24 x ≥ 0
fX (x) = (3)
0 otherwise
(c) From above, we know that Var[X] = n/λ2 = 45.
85
Since Y is an Erlang random variable with parameters λ = 2 and n = 2, we find in Appendix A
that
4ye−2y y ≥ 0
fY (y) = (1)
0 otherwise
(a) Appendix A tells us that E[Y ] = n/λ = 1.
(b) Appendix A also tells us that Var[Y ] = n/λ2 = 1/2.
(c) The probability that 1/2 ≤ Y < 3/2 is
' 3/2 ' 3/2
P [1/2 ≤ Y < 3/2] = fY (y) dy = 4ye−2y dy (2)
1/2 1/2
( (
This integral is easily completed using the integration by parts formula u dv = uv − v du
with
u = 2y dv = 2e−2y
du = 2dy v = −e−2y
Making these substitutions, we obtain

'
$3/2 3/2
P [1/2 ≤ Y < 3/2] = −2ye−2y $1/2 + 2e−2y dy (3)
1/2
= 2e−1 − 4e−3 = 0.537 (4)
(a) The PDF of a continuous uniform (−5, 5) random variable is

1/10 −5 ≤ x ≤ 5
fX (x) = (1)
0 otherwise
(b) For x < −5, FX (x) = 0. For x ≥ 5, FX (x) = 1. For −5 ≤ x ≤ 5, the CDF is
' x
x+5
FX (x) = fX (τ ) dτ = (2)
−5 10
The complete expression for the CDF of X is
⎧
⎨ 0 x < −5
FX (x) = (x + 5)/10 5 ≤ x ≤ 5 (3)
⎩
1 x>5
(c) The expected value of X is

' $5
5
x x2 $$
dx = =0 (4)
−5 10 20 $−5
Another way to obtain this answer is to use Theorem 3.6 which says the expected value of X
is E[X] = (5 + −5)/2 = 0.
86
(d) The fifth moment of X is
' $5
5
x5 x6 $$
dx = =0 (5)
−5 10 60 $−5
(e) The expected value of eX is

' $
5
ex ex $$5 e5 − e−5
dx = $ = = 14.84 (6)
−5 10 10 −5 10

We know that X has a uniform PDF over [a, b) and has mean µX = 7 and variance Var[X] = 3.
All that is left to do is determine the values of the constants a and b, to complete the model of the
uniform PDF.
a+b (b − a)2
E [X] = =7 Var[X] = =3 (1)
2 12
Since we assume b > a, this implies
a + b = 14 b−a=6 (2)
Solving these two equations, we arrive at
a=4 b = 10 (3)
And the resulting PDF of X is,

1/6 4 ≤ x ≤ 10
fX (x) = (4)
0 otherwise

Given that
(1/2)e−x/2 x ≥ 0
fX (x) = (1)
0 otherwise
(a)
' 2
P [1 ≤ X ≤ 2] = (1/2)e−x/2 dx = e−1/2 − e−1 = 0.2387 (2)
1
(b) The CDF of X may be be expressed as

0 x<0 0 x<0
FX (x) = ( x −x/2 dτ = −x/2 (3)
0 (1/2)e x≥0 1−e x≥0
(c) X is an exponential random variable with parameter a = 1/2. By Theorem 3.8, the expected
value of X is E[X] = 1/a = 2.
(d) By Theorem 3.8, the variance of X is Var[X] = 1/a2 = 4.
87
Given the uniform PDF
1/(b − a) a ≤ u ≤ b
fU (u) = (1)
0 otherwise
The mean of U can be found by integrating
' b
b2 − a2 b+a
E [U ] = u/(b − a) du = = (2)
a 2(b − a) 2
Where we factored (b2 − a2 ) = (b − a)(b + a). The variance of U can also be found by finding E[U 2 ].
' b
2 (b3 − a3 )
E U = u2 /(b − a) du = (3)
a 3(b − a)
Therefore the variance is
2
(b3 − a3 ) b+a (b − a)2
Var[U ] = − = (4)
3(b − a) 2 12

Let X denote the holding time of a call. The PDF of X is

(1/τ )e−x/τ x ≥ 0
fX (x) = (1)
0 otherwise
We will use CA (X) and CB (X) to denote the cost of a call under the two plans. From the problem
statement, we note that CA (X) = 10X so that E[CA (X)] = 10E[X] = 10τ . On the other hand
CB (X) = 99 + 10(X − 20)+ (2)
where y + = y if y ≥ 0; otherwise y + = 0 for y < 0. Thus,

E [CB (X)] = E 99 + 10(X − 20)+ (3)

= 99 + 10E (X − 20)+ (4)

= 99 + 10E (X − 20)+ |X ≤ 20 P [X ≤ 20]

+ 10E (X − 20)+ |X > 20 P [X > 20] (5)
Given X ≤ 20, (X − 20)+ = 0. Thus E[(X − 20)+ |X ≤ 20] = 0 and
E [CB (X)] = 99 + 10E [(X − 20)|X > 20] P [X > 20] (6)
Finally, we observe that P [X > 20] = e−20/τ and that
E [(X − 20)|X > 20] = τ (7)
since given X ≥ 20, X −20 has a PDF identical to X by the memoryless property of the exponential
random variable. Thus,
E [CB (X)] = 99 + 10τ e−20/τ (8)
Some numeric comparisons show that E[CB (X)] ≤ E[CA (X)] if τ > 12.34 minutes. That is, the
flat price for the first 20 minutes is a good deal only if your average phone call is sufficiently long.
88
The integral I1 is ' ∞ $∞
$
I1 = λe−λx dx = −e−λx $ = 1 (1)
0 0
For n > 1, we have ' ∞

λn−1 xn−1 −λx
In = λe dt (2)
(n − 1)!
0
dv
u
(
We define
( u and dv as shown above in order to use the integration by parts formula u dv =
uv − v du. Since
λn−1 xn−1
du = dx v = −e−λx (3)
(n − 2)!
we can write
' ∞
∞
In = uv|0 − v du (4)
0
$∞ ' ∞ n−1 n−1
λn−1 xn−1 −λx $$ λ x
=− e $ + e−λx dx = 0 + In−1 (5)
(n − 1)! 0 0 (n − 2)!
Hence, In = 1 for all n ≥ 1.

For an Erlang (n, λ) random variable X, the kth moment is
' ∞
E Xk = xk fX (x) dt (1)
0
' ∞ n n+k−1 '
λ x −λx (n + k − 1)! ∞ λn+k xn+k−1 −λt
= e dt = k e dt (2)
(n − 1)! λ (n − 1)! 0 (n + k − 1)!
0

1
The above marked integral equals 1 since it is the integral of an Erlang PDF with parameters λ
and n + k over all possible values. Hence,
(n + k − 1)!
E Xk = k (3)
λ (n − 1)!
This implies that the first and second moments are
n! n (n + 1)! (n + 1)n
E [X] = = E X2 = 2 = (4)
(n − 1)!λ λ λ (n − 1)! λ2
It follows that the variance of X is n/λ2 .

In this problem, we prove Theorem 3.11 which says that for x ≥ 0, the CDF of an Erlang (n, λ)
random variable Xn satisfies
(λx)k e−λx
n−1
FXn (x) = 1 − . (1)
k!
k=0
We do this in two steps. First, we derive a relationship between FXn (x) and FXn−1 (x). Second, we
use that relationship to prove the theorem by induction.
89
(a) By Definition 3.7, the CDF of Erlang (n, λ) random variable Xn is
' x ' x n n−1 −λt
λ t e
FXn (x) = fXn (t) dt = dt. (2)
−∞ 0 (n − 1)!
(b) To use integration by parts, we define

tn−1
u= dv = λn e−λt dt (3)
(n − 1)!
tn−2
du = v = −λn−1 e−λt (4)
(n − 2)!
( (
Thus, using the integration by parts formula u dv = uv − v du, we have
' x n n−1 −λt $x ' x n−1 n−2 −λt
λ t e λn−1 tn−1 e−λt $$ λ t e
FXn (x) = dt = − $ + dt (5)
0 (n − 1)! (n − 1)! 0 0 (n − 2)!
λn−1 xn−1 e−λx
=− + FXn−1 (x) (6)
(n − 1)!
(c) Now we do proof by induction. For n = 1, the Erlang (n, λ) random variable X1 is simply
an exponential random variable. Hence for x ≥ 0, FX1 (x) = 1 − e−λx . Now we suppose the
claim is true for FXn−1 (x) so that

n−2
(λx)k e−λx
FXn−1 (x) = 1 − . (7)
k!
k=0
Using the result of part (a), we can write

(λx)n−1 e−λx
FXn (x) = FXn−1 (x) − (8)
(n − 1)!

n−2
(λx)k e−λx (λx)n−1 e−λx
=1− − (9)
k! (n − 1)!
k=0
which proves the claim.

For n = 1, we have the fact E[X] = 1/λ that is given in the problem statement. Now we assume that
E[X n−1 ] = (n − 1)!/λn−1 . To complete the proof, we show that this implies that E[X n ] = n!/λn .
Specifically, we write
'
E [X ] = xn λe−λx dx
n
(1)
0
( (
Now we use the integration by parts formula u dv = uv − v du with u = xn and dv = λe−λx dx.
This implies du = nxn−1 dx and v = −e−λx so that
$∞ ' ∞
n −λx $
E [X ] = −x e
n
$ + nxn−1 e−λx dx (2)
0
' 0
n ∞ n−1 −λx
=0+ x λe dx (3)
λ 0
n
= E X n−1 (4)
λ
90
By our induction hyothesis, E[X n−1 ] = (n − 1)!/λn−1 which implies
E [X n ] = n!/λn (5)
(a) Since fX (x) ≥ 0 and x ≥ r over the entire integral, we can write
' ∞ ' ∞
xfX (x) dx ≥ rfX (x) dx = rP [X > r] (1)
r r
(b) We can write the expected value of X in the form

' r ' ∞
E [X] = xfX (x) dx + xfX (x) dx (2)
0 r
Hence, ' ∞ ' r
rP [X > r] ≤ xfX (x) dx = E [X] − xfX (x) dx (3)
r 0
Allowing r to approach infinity yields
' r
lim rP [X > r] ≤ E [X] − lim xfX (x) dx = E [X] − E [X] = 0 (4)
r→∞ r→∞ 0
Since rP [X > r] ≥ 0 for all r ≥ 0, we must have limr→∞ rP [X > r] = 0.

( (
(c) We can use the integration by parts formula u dv = uv − v du by defining u = 1 − FX (x)
and dv = dx. This yields
' ∞ ' ∞
∞
[1 − FX (x)] dx = x[1 − FX (x)]|0 + xfX (x) dx (5)
0 0
By applying part (a), we now observe that
x [1 − FX (x)]|∞
0 = lim r[1 − FX (r)] − 0 = lim rP [X > r] (6)
r→∞ r→∞
By part (b), limr→∞ rP [X > r] = 0 and this implies x[1 − FX (x)]|∞

0 = 0. Thus,
' ∞ ' ∞
[1 − FX (x)] dx = xfX (x) dx = E [X] (7)
0 0

Given that the peak temperature, T , is a Gaussian random variable with mean 85 and standard
deviation 10 we can use the fact that FT (t) = Φ((t − µT )/σT ) and Table 3.1 on page 123 to evaluate
the following

100 − 85
P [T > 100] = 1 − P [T ≤ 100] = 1 − FT (100) = 1 − Φ (1)
10
= 1 − Φ(1.5) = 1 − 0.933 = 0.066 (2)

60 − 85
P [T < 60] = Φ = Φ(−2.5) (3)
10
= 1 − Φ(2.5) = 1 − .993 = 0.007 (4)
P [70 ≤ T ≤ 100] = FT (100) − FT (70) (5)
= Φ(1.5) − Φ(−1.5) = 2Φ(1.5) − 1 = .866 (6)
91
The standard normal Gaussian random variable Z has mean µ = 0 and variance σ 2 = 1. Making
these substitutions in Definition 3.8 yields
1 2
fZ (z) = √ e−z /2 (1)
2π

X is a Gaussian random variable with zero mean but unknown variance. We do know, however,
that
P [|X| ≤ 10] = 0.1 (1)
We can find the variance Var[X] by expanding the above probability in terms of the Φ(·) function.

10
P [−10 ≤ X ≤ 10] = FX (10) − FX (−10) = 2Φ −1 (2)
σX
This implies Φ(10/σX ) = 0.55. Using Table 3.1 for the Gaussian CDF, we find that 10/σX = 0.15
or σX = 66.6.

Repeating Definition 3.11, ' ∞
1 2 /2
Q(z) = √ e−u du (1)
2π z
√
Making the substitution x = u/ 2, we have
' ∞
1 −x2 1 z
Q(z) = √ e dx = erfc √ (2)
π z/√2 2 2

Moving to Antarctica, we find that the temperature, T is still Gaussian but with variance 225. We
also know that with probability 1/2, T exceeds 10 degrees. First we would like to find the mean
temperature, and we do so by looking at the second fact.

10 − µT
P [T > 10] = 1 − P [T ≤ 10] = 1 − Φ = 1/2 (1)
15
By looking at the table we find that if Φ(Γ) = 1/2, then Γ = 0. Therefore,

10 − µT
Φ = 1/2 (2)
15
implies that (10 − µT )/15 = 0 or µT = 10. Now we have a Gaussian T with mean 10 and standard
92
deviation 15. So we are prepared to answer the following problems.

32 − 10
P [T > 32] = 1 − P [T ≤ 32] = 1 − Φ (3)
15
= 1 − Φ(1.45) = 1 − 0.926 = 0.074 (4)

0 − 10
P [T < 0] = FT (0) = Φ (5)
15
= Φ(−2/3) = 1 − Φ(2/3) (6)
= 1 − Φ(0.67) = 1 − 0.749 = 0.251 (7)
P [T > 60] = 1 − P [T ≤ 60] = 1 − FT (60) (8)

60 − 10
=1−Φ = 1 − Φ(10/3) (9)
15
= Q(3.33) = 4.34 · 10−4 (10)

In this problem, we use Theorem 3.14 and the tables for the Φ and Q functions to answer the
questions. Since E[Y20 ] = 40(20) = 800 and Var[Y20 ] = 100(20) = 2000, we can write

Y20 − 800 1000 − 800
P [Y20 > 1000] = P √ > √ (1)
2000 2000

200
= P Z > √ = Q(4.47) = 3.91 × 10−6 (2)
20 5
The second part is a little trickier. Since E[Y25 ] = 1000, we know that the prof will spend around
$1000 in roughly 25 years. However, to be certain with probability 0.99 that the prof spends $1000
will require more than 25 years. In particular, we know that

Yn − 40n 1000 − 40n 100 − 4n
P [Yn > 1000] = P √ > √ =1−Φ √ = 0.99 (3)
100n 100n n
Hence, we must find n such that
100 − 4n
Φ √ = 0.01 (4)
n
Recall that Φ(x) = 0.01 for a negative value of x. This is consistent with our earlier observation that
we would need n > 25 corresponding to 100 − 4n < 0. Thus, we use the identity Φ(x) = 1 − Φ(−x)
to write
100 − 4n 4n − 100
Φ √ =1−Φ √ = 0.01 (5)
n n
Equivalently, we have
4n − 100
Φ √ = 0.99 (6)
n
√
From the table of the Φ function, we have that (4n − 100)/ n = 2.33, or
(n − 25)2 = (0.58)2 n = 0.3393n. (7)
Solving this quadratic yields n = 28.09. Hence, only after 28 years are we 99 percent sure that the
prof will have spent $1000. Note that a second root of the quadratic yields n = 22.25. This root is
not a valid solution to our problem. Mathematically, it is a solution of our quadratic in which we
√
choose the negative root of n. This would correspond to assuming the standard deviation of Yn
is negative.
93
We are given that there are 100,000,000 men in the United States and 23,000 of them are at least 7
feet tall, and the heights of U.S men are independent Gaussian random variables with mean 5 10 .
(a) Let H denote the height in inches of a U.S male. To find σX , we look at the fact that the
probability that P [H ≥ 84] is the number of men who are at least 7 feet tall divided by the
total number of men (the frequency interpretation of probability). Since we measure H in
inches, we have

23,000 70 − 84
P [H ≥ 84] = =Φ = 0.00023 (1)
100,000,000 σX
Since Φ(−x) = 1 − Φ(x) = Q(x),
Q(14/σX ) = 2.3 · 10−4 (2)
From Table 3.2, this implies 14/σX = 3.5 or σX = 4.
(b) The probability that a randomly chosen man is at least 8 feet tall is

96 − 70
P [H ≥ 96] = Q = Q(6.5) (3)
4
Unfortunately, Table 3.2 doesn’t include Q(6.5), although it should be apparent that the
probability is very small. In fact, Q(6.5) = 4.0 × 10−11 .
(c) First we need to find the probability that a man is at least 7’6”.

90 − 70
P [H ≥ 90] = Q = Q(5) ≈ 3 · 10−7 = β (4)
4
Although Table 3.2 stops at Q(4.99), if you’re curious, the exact value is Q(5) = 2.87 · 10−7 .
Now we can begin to find the probability that no man is at least 7’6”. This can be modeled
as 100,000,000 repetitions of a Bernoulli trial with parameter 1 − β. The probability that no
man is at least 7’6” is
(1 − β)100,000,000 = 9.4 × 10−14 (5)
(d) The expected value of N is just the number of trials multiplied by the probability that a man
is at least 7’6”.
E [N ] = 100,000,000 · β = 30 (6)

This problem is in the wrong section since the erf(·) function is defined later on in Section 3.9 as
' x
2 2
erf(x) = √ e−u du. (1)
π 0
94
√
(a) Since Y is Gaussian (0, 1/ 2), Y has variance 1/2 and
1 2 /[2(1/2)] 1 2
fY (y) = e−y = √ e−y . (2)
2π(1/2) π
(y (y
For y ≥ 0, FY (y) = −∞ fY (u) du = 1/2 + 0 fY (u) du. Substituting fY (u) yields
' y
1 1 2 1
FY (y) = + √ e−u du = + erf(y). (3)
2 π 0 2
√ √ √
(b) Since Y is Gaussian (0, 1/ 2), Z = 2Y is Gaussian with expected value E[Z] = 2E[Y ] = 0
and variance Var[Z] = 2 Var[Y ] = 1. Thus Z is Gaussian (0, 1) and
√
z z 1 z
Φ(z) = FZ (z) = P 2Y ≤ z = P Y ≤ √ = FY √ = + erf √ (4)
2 2 2 2

First we note that since W has an N [µ, σ 2 ] distribution, the integral we wish to evaluate is
' ∞ ' ∞
1 2 2
I= fW (w) dw = √ e−(w−µ) /2σ dw (1)
−∞ 2
2πσ −∞
(a) Using the substitution x = (w − µ)/σ, we have dx = dw/σ and
' ∞
1 2
I=√ e−x /2 dx (2)
2π −∞
(b) When we write I 2 as the product of integrals, we use y to denote the other variable of
integration so that
' ∞ ' ∞
1 −x2 /2 1 −y 2 /2
2
I = √ e dx √ e dy (3)
2π 2π −∞
' ∞ −∞
' ∞
1 2 2
= e−(x +y )/2 dx dy (4)
2π −∞ −∞
(c) By changing to polar coordinates, x2 + y 2 = r2 and dx dy = r dr dθ so that

' 2π ' ∞
1 2
2
I = e−r /2 r dr dθ (5)
2π 0 0
' 2π $∞ ' 2π
1 2 $ 1
= −e−r /2 $ dθ = dθ = 1 (6)
2π 0 0 2π 0

This problem is mostly calculus and only a little probability. From the problem statement, the
SNR Y is an exponential (1/γ) random variable with PDF

(1/γ)e−y/γ y ≥ 0,
fY (y) = (1)
0 otherwise.
95
Thus, from the problem statement, the BER is
' ∞ ' ∞ y
P e = E [Pe (Y )] = Q( 2y)fY (y) dy = Q( 2y) e−y/γ dy (2)
−∞ 0 γ
Like most integrals with exponential factors, its a good idea to try integration by parts. Before
doing so, we recall that if X is a Gaussian (0, 1) random variable with CDF FX (x), then
Q(x) = 1 − FX (x) . (3)
It follows that Q(x) has derivative

dQ(x) dFX (x) 1 2
Q (x) = =− = −fX (x) = − √ e−x /2 (4)
dx dx 2π
(b (b
To solve the integral, we use the integration by parts formula a u dv = uv|ba − a v du, where
1 −y/γ
u = Q( 2y) dv = e dy (5)
γ
1 e−y
du = Q ( 2y) √ = − √ v = −e−y/γ (6)
2y 2 πy
From integration by parts, it follows that
' ∞ $∞ ' ∞ 1
∞ −y/γ $
P e = uv|0 − v du = −Q( 2y)e $ − √ e−y[1+(1/γ)] dy (7)
0 0 0 y
' ∞
1
= 0 + Q(0)e−0 − √ y −1/2 e−y/γ̄ dy (8)
2 π 0
where γ̄ = γ/(1 + γ). Next, recalling that Q(0) = 1/2 and making the substitution t = y/γ̄, we
obtain
) '
1 1 γ̄ ∞ −1/2 −t
Pe = − t e dt (9)
2 2 π 0
From Math Fact B.11, we see that the remaining integral is the Γ(z) function evaluated z = 1/2.
√
Since Γ(1/2) = π,
) )
1 1 γ̄ 1 √ 1 γ
Pe = − Γ(1/2) = 1 − γ̄ = 1− (10)
2 2 π 2 2 1+γ
(a) Using the given CDF

P [X < −1] = FX −1− = 0 (1)
P [X ≤ −1] = FX (−1) = −1/3 + 1/3 = 0 (2)
Where FX (−1− ) denotes the limiting value of the CDF found by approaching −1 from the
left. Likewise, FX (−1+ ) is interpreted to be the value of the CDF found by approaching
−1 from the right. We notice that these two probabilities are the same and therefore the
probability that X is exactly −1 is zero.
96
(b)

P [X < 0] = FX 0− = 1/3 (3)
P [X ≤ 0] = FX (0) = 2/3 (4)
Here we see that there is a discrete jump at X = 0. Approached from the left the CDF yields
a value of 1/3 but approached from the right the value is 2/3. This means that there is a
non-zero probability that X = 0, in fact that probability is the difference of the two values.
P [X = 0] = P [X ≤ 0] − P [X < 0] = 2/3 − 1/3 = 1/3 (5)
(c)

P [0 < X ≤ 1] = FX (1) − FX 0+ = 1 − 2/3 = 1/3 (6)

P [0 ≤ X ≤ 1] = FX (1) − FX 0− = 1 − 1/3 = 2/3 (7)
The difference in the last two probabilities above is that the first was concerned with the
probability that X was strictly greater then 0, and the second with the probability that X
was greater than or equal to zero. Since the the second probability is a larger set (it includes
the probability that X = 0) it should always be greater than or equal to the first probability.
The two differ by the probability that X = 0, and this difference is non-zero only when the
random variable exhibits a discrete jump in the CDF.

Similar to the previous problem we find
(a)

P [X < −1] = FX −1− = 0 P [X ≤ −1] = FX (−1) = 1/4 (1)
Here we notice the discontinuity of value 1/4 at x = −1.
(b)

P [X < 0] = FX 0− = 1/2 P [X ≤ 0] = FX (0) = 1/2 (2)
Since there is no discontinuity at x = 0, FX (0− ) = FX (0+ ) = FX (0).
(c)
P [X > 1] = 1 − P [X ≤ 1] = 1 − FX (1) = 0 (3)

P [X ≥ 1] = 1 − P [X < 1] = 1 − FX 1− = 1 − 3/4 = 1/4 (4)
Again we notice a discontinuity of size 1/4, here occurring at x = 1.
(a) By taking the derivative of the CDF FX (x) given in Problem 3.6.2, we obtain the PDF
δ(x+1)
4 + 1/4 + δ(x−1)
4 −1 ≤ x ≤ 1
fX (x) = (1)
0 otherwise
97
(b) The first moment of X is
' ∞
E [X] = xfX (x) dx (2)
−∞
$1
= x/4|x=−1 + x2 /8$−1 + x/4|x=1 = −1/4 + 0 + 1/4 = 0. (3)
(c) The second moment of X is

' ∞
2
E X = x2 fX (x) dx (4)
−∞
$ $1 $
= x2 /4$x=−1 + x3 /12$−1 + x2 /4$x=1 = 1/4 + 1/6 + 1/4 = 2/3. (5)
Since E[X] = 0, Var[X] = E[X 2 ] = 2/3.

The PMF of a Bernoulli random variable with mean p is
⎧
⎨ 1−p x=0
PX (x) = p x=1 (1)
⎩
0 otherwise
The corresponding PDF of this discrete random variable is
fX (x) = (1 − p)δ(x) + pδ(x − 1) (2)

The PMF of a geometric random variable with mean 1/p is

p(1 − p)x−1 x = 1, 2, . . .
PX (x) = (1)
0 otherwise
The corresponding PDF is
fX (x) = pδ(x − 1) + p(1 − p)δ(x − 2) + · · · (2)
∞
= p(1 − p)j−1 δ(x − j) (3)
j=1
(a) Since the conversation time cannot be negative, we know that FW (w) = 0 for w < 0. The
conversation time W is zero iff either the phone is busy, no one answers, or if the conversation
time X of a completed call is zero. Let A be the event that the call is answered. Note that
the event Ac implies W = 0. For w ≥ 0,
FW (w) = P [Ac ] + P [A] FW |A (w) = (1/2) + (1/2)FX (w) (1)
Thus the complete CDF of W is

0 w<0
FW (w) = (2)
1/2 + (1/2)FX (w) w ≥ 0
98
(b) By taking the derivative of FW (w), the PDF of W is

(1/2)δ(w) + (1/2)fX (w)
fW (w) = (3)
0 otherwise
Next, we keep in mind that since X must be nonnegative, fX (x) = 0 for x < 0. Hence,
fW (w) = (1/2)δ(w) + (1/2)fX (w) (4)
(c) From the PDF fW (w), calculating the moments is straightforward.

' ∞ ' ∞
E [W ] = wfW (w) dw = (1/2) wfX (w) dw = E [X] /2 (5)
−∞ −∞
The second moment is

' ∞ ' ∞
2
2

E W = w fW (w) dw = (1/2) w2 fX (w) dw = E X 2 /2 (6)
−∞ −∞
The variance of W is

Var[W ] = E W 2 − (E [W ])2 = E X 2 /2 − (E [X] /2)2 (7)
2
= (1/2) Var[X] + (E [X]) /4 (8)

The professor is on time 80 percent of the time and when he is late his arrival time is uniformly
distributed between 0 and 300 seconds. The PDF of T , is

0.8δ(t − 0) + 300
0.2
0 ≤ t ≤ 300
fT (t) = (1)
0 otherwise
The CDF can be found be integrating

⎧
⎨ 0 t < −1
FT (t) = 0.8 + 0.2t
0 ≤ t < 300 (2)
⎩ 300
1 t ≥ 300

Let G denote the event that the throw is good, that is, no foul occurs. The CDF of D obeys
FD (y) = P [D ≤ y|G] P [G] + P [D ≤ y|Gc ] P [Gc ] (1)
Given the event G,
P [D ≤ y|G] = P [X ≤ y − 60] = 1 − e−(y−60)/10 (y ≥ 60) (2)
Of course, for y < 60, P [D ≤ y|G] = 0. From the problem statement, if the throw is a foul, then
D = 0. This implies
P [D ≤ y|Gc ] = u(y) (3)
99
where u(·) denotes the unit step function. Since P [G] = 0.7, we can write
FD (y) = P [G] P [D ≤ y|G] + P [Gc ] P [D ≤ y|Gc ] (4)

0.3u(y) y < 60
= (5)
0.3 + 0.7(1 − e−(y−60)/10 ) y ≥ 60
Another way to write this CDF is
FD (y) = 0.3u(y) + 0.7u(y − 60)(1 − e−(y−60)/10 ) (6)
However, when we take the derivative, either expression for the CDF will yield the PDF. However,
taking the derivative of the first expression perhaps may be simpler:

0.3δ(y) y < 60
fD (y) = (7)
0.07e−(y−60)/10 y ≥ 60
Taking the derivative of the second expression for the CDF is a little tricky because of the product
of the exponential and the step function. However, applying the usual rule for the differentation of
a product does give the correct answer:
fD (y) = 0.3δ(y) + 0.7δ(y − 60)(1 − e−(y−60)/10 ) + 0.07u(y − 60)e−(y−60)/10 (8)

−(y−60)/10
= 0.3δ(y) + 0.07u(y − 60)e (9)
The middle term δ(y − 60)(1 − e−(y−60)/10 ) dropped out because at y = 60, e−(y−60)/10 = 1.

The professor is on time and lectures the full 80 minutes with probability 0.7. In terms of math,
P [T = 80] = 0.7. (1)
Likewise when the professor is more than 5 minutes late, the students leave and a 0 minute lecture
is observed. Since he is late 30% of the time and given that he is late, his arrival is uniformly
distributed between 0 and 10 minutes, the probability that there is no lecture is
P [T = 0] = (0.3)(0.5) = 0.15 (2)
The only other possible lecture durations are uniformly distributed between 75 and 80 minutes,
because the students will not wait longer then 5 minutes, and that probability must add to a total
of 1 − 0.7 − 0.15 = 0.15. So the PDF of T can be written as
⎧
⎪
⎪ 0.15δ(t) t=0
⎨
0.03 75 ≤ 7 < 80
fT (t) = (3)
⎪
⎪ 0.7δ(t − 80) t = 80
⎩
0 otherwise

Since 0 ≤ X ≤ 1, Y = X 2 satisfies 0 ≤ Y ≤ 1. We can conclude that FY (y) = 0 for y < 0 and that
FY (y) = 1 for y ≥ 1. For 0 ≤ y < 1,
√
FY (y) = P X 2 ≤ y = P [X ≤ y] (1)
100
Since fX (x) = 1 for 0 ≤ x ≤ 1, we see that for 0 ≤ y < 1,
' √
√ y
√
P [X ≤ y] = dx = y (2)
0
Hence, the CDF of Y is ⎧

⎨ 0 y<0
√
FY (y) = y 0≤y<1 (3)
⎩
1 y≥1
By taking the derivative of the CDF, we obtain the PDF
√
1/(2 y) 0 ≤ y < 1
fY (y) = (4)
0 otherwise
Problem √
3.7.2 Solution
Since Y = X, the fact that X is nonegative and that we asume the squre root is always positive
implies FY (y) = 0 for y < 0. In addition, for y ≥ 0, we can find the CDF of Y by writing
√
FY (y) = P [Y ≤ y] = P X ≤ y = P X ≤ y 2 = FX y 2 (1)
For x ≥ 0, FX (x) = 1 − e−λx . Thus,

2
1 − e−λy y≥0
FY (y) = (2)
0 otherwise
By taking the derivative with respect to y, it follows that the PDF of Y is

2
2λye−λy y ≥ 0
fY (y) = (3)
0 otherwise
In comparing this result to the √

Rayleigh PDF given in Appendix A, we observe that Y is a Rayleigh
(a) random variable with a = 2λ.

Since X is non-negative, W = X 2 is also non-negative. Hence for w < 0, fW (w) = 0. For w ≥ 0,

FW (w) = P [W ≤ w] = P X 2 ≤ w (1)
= P [X ≤ w] (2)
√
−λ w
=1−e (3)
√ √
Taking the derivative with respect to w yields fW (w) = λe−λ w /(2 w). The complete expression
for the PDF is −λ√w
λe √
2 w
w≥0
fW (w) = (4)
0 otherwise
101
From Problem 3.6.1, random variable X has CDF
⎧
⎪
⎪ 0 x < −1
⎨
x/3 + 1/3 −1 ≤ x < 0
FX (x) = (1)
⎪
⎪ x/3 + 2/3 0≤x<1
⎩
1 1≤x
(a) We can find the CDF of Y , FY (y) by noting that Y can only take on two possible values, 0
and 100. And the probability that Y takes on these two values depends on the probability
that X < 0 and X ≥ 0, respectively. Therefore
⎧
⎨ 0 y<0
FY (y) = P [Y ≤ y] = P [X < 0] 0 ≤ y < 100 (2)
⎩
1 y ≥ 100
The probabilities concerned with X can be found from the given CDF FX (x). This is the
general strategy for solving problems of this type: to express the CDF of Y in terms of the
CDF of X. Since P [X < 0] = FX (0− ) = 1/3, the CDF of Y is
⎧
⎨ 0 y<0
FY (y) = P [Y ≤ y] = 1/3 0 ≤ y < 100 (3)
⎩
1 y ≥ 100
(b) The CDF FY (y) has jumps of 1/3 at y = 0 and 2/3 at y = 100. The corresponding PDF of
Y is
fY (y) = δ(y)/3 + 2δ(y − 100)/3 (4)
(c) The expected value of Y is

' ∞
1 2
E [Y ] = yfY (y) dy = 0 · + 100 · = 66.66 (5)
−∞ 3 3

Before solving for the PDF, it is helpful to have a sketch of the function X = − ln(1 − U ).
2
X
0
0 0.5 1
U
(a) From the sketch, we observe that X will be nonnegative. Hence FX (x) = 0 for x < 0. Since
U has a uniform distribution on [0, 1], for 0 ≤ u ≤ 1, P [U ≤ u] = u. We use this fact to find
the CDF of X. For x ≥ 0,

FX (x) = P [− ln(1 − U ) ≤ x] = P 1 − U ≥ e−x = P U ≤ 1 − e−x (1)
102
For x ≥ 0, 0 ≤ 1 − e−x ≤ 1 and so

FX (x) = FU 1 − e−x = 1 − e−x (2)
The complete CDF can be written as

0 x<0
FX (x) = (3)
1 − e−x x ≥ 0
(b) By taking the derivative, the PDF is

e−x x ≥ 0
fX (x) = (4)
0 otherwise
Thus, X has an exponential PDF. In fact, since most computer languages provide uniform
[0, 1] random numbers, the procedure outlined in this problem provides a way to generate
exponential random variables from uniform random variables.
(c) Since X is an exponential random variable with parameter a = 1, E[X] = 1.

We wish to find a transformation that takes a uniformly distributed random variable on [0,1] to
the following PDF for Y . 2
3y 0 ≤ y ≤ 1
fY (y) = (1)
0 otherwise
We begin by realizing that in this case the CDF of Y must be
⎧
⎨ 0 y<0
FY (y) = y3 0≤y≤1 (2)
⎩
1 otherwise
Therefore, for 0 ≤ y ≤ 1,
P [Y ≤ y] = P [g(X) ≤ y] = y 3 (3)
Thus, using g(X) = X 1/3 , we see that for 0 ≤ y ≤ 1,

P [g(X) ≤ y] = P X 1/3 ≤ y = P X ≤ y 3 = y 3 (4)
which is the desired answer.

Since the microphone voltage V is uniformly distributed between -1 and 1 volts, V has PDF and
CDF ⎧
⎨ 0 v < −1
1/2 −1 ≤ v ≤ 1
fV (v) = FV (v) = (v + 1)/2 −1 ≤ v ≤ 1 (1)
0 otherwise ⎩
1 v>1
The voltage is processed by a limiter whose output magnitude is given by below

|V | |V | ≤ 0.5
L= (2)
0.5 otherwise
103
(a)
P [L = 0.5] = P [|V | ≥ 0.5] = P [V ≥ 0.5] + P [V ≤ −0.5] (3)
= 1 − FV (0.5) + FV (−0.5) (4)
= 1 − 1.5/2 + 0.5/2 = 1/2 (5)
(b) For 0 ≤ l ≤ 0.5,

FL (l) = P [|V | ≤ l] = P [−l ≤ v ≤ l] = FV (l) − FV (−l) (6)
= 1/2(l + 1) − 1/2(−l + 1) = l (7)
So the CDF of L is ⎧
⎨ 0 l<0
FL (l) = l 0 ≤ l < 0.5 (8)
⎩
1 l ≥ 0.5
(c) By taking the derivative of FL (l), the PDF of L is

1 + (0.5)δ(l − 0.5) 0 ≤ l ≤ 0.5
fL (l) = (9)
0 otherwise
The expected value of L is
' ∞ ' 0.5 ' 0.5
E [L] = lfL (l) dl = l dl + 0.5 l(0.5)δ(l − 0.5) dl = 0.375 (10)
−∞ 0 0

Let X denote the position of the pointer and Y denote the area within the arc defined by the
stopping position of the pointer.
(a) If the disc has radius r, then the area of the disc is πr2 . Since the circumference of the disc
is 1 and X is measured around the circumference, Y = πr2 X. For example, when X = 1, the
shaded area is the whole disc and Y = πr2 . Similarly, if X = 1/2, then Y = πr2 /2 is half the
area of the disc. Since the disc has circumference 1, r = 1/(2π) and
X
Y = πr2 X = (1)
4π
(b) The CDF of Y can be expressed as

X
FY (y) = P [Y ≤ y] = P ≤ y = P [X ≤ 4πy] = FX (4πy) (2)
4π
Therefore the CDF is ⎧
⎨ 0 y<0
FY (y) = 4πy 0 ≤ y ≤ 1
(3)
⎩ 4π
1 y ≥ 4π
1
(c) By taking the derivative of the CDF, the PDF of Y is

4π 0 ≤ y ≤ 4π
1
fY (y) = (4)
0 otherwise
( 1/(4π)
(d) The expected value of Y is E[Y ] = 0 4πy dy = 1/(8π).
104
The uniform (0, 2) random variable U has PDF and CDF
⎧
⎨ 0 u < 0,
1/2 0 ≤ u ≤ 2,
fU (u) = FU (u) = u/2 0 ≤ u < 2, (1)
0 otherwise, ⎩
1 u > 2.
The uniform random variable U is subjected to the following clipper.

U U ≤1
W = g(U ) = (2)
1 U >1
To find the CDF of the output of the clipper, W , we remember that W = U for 0 ≤ U ≤ 1
while W = 1 for 1 ≤ U ≤ 2. First, this implies W is nonnegative, i.e., FW (w) = 0 for w < 0.
Furthermore, for 0 ≤ w ≤ 1,
FW (w) = P [W ≤ w] = P [U ≤ w] = FU (w) = w/2 (3)
Lastly, we observe that it is always true that W ≤ 1. This implies FW (w) = 1 for w ≥ 1. Therefore
the CDF of W is ⎧
⎨ 0 w<0
FW (w) = w/2 0≤w<1 (4)
⎩
1 w≥1
From the jump in the CDF at w = 1, we see that P [W = 1] = 1/2. The corresponding PDF can
be found by taking the derivative and using the delta function to model the discontinuity.

1/2 + (1/2)δ(w − 1) 0 ≤ w ≤ 1
fW (w) = (5)
0 otherwise
The expected value of W is

' ∞ ' 1
E [W ] = wfW (w) dw = w[1/2 + (1/2)δ(w − 1)] dw (6)
−∞ 0
= 1/4 + 1/2 = 3/4. (7)

Given the following function of random variable X,

10 X<0
Y = g(X) = (1)
−10 X ≥ 0
we follow the same procedure as in Problem 3.7.4. We attempt to express the CDF of Y in terms
of the CDF of X. We know that Y is always less than −10. We also know that −10 ≤ Y < 10
when X ≥ 0, and finally, that Y = 10 when X < 0. Therefore
⎧
⎨ 0 y < −10
FY (y) = P [Y ≤ y] = P [X ≥ 0] = 1 − FX (0) −10 ≤ y < 10 (2)
⎩
1 y ≥ 10
105
The PDF of U is
1/2 −1 ≤ u ≤ 1
fU (u) = (1)
0 otherwise
Since W ≥ 0, we see that FW (w) = 0 for w < 0. Next, we observe that the rectifier output W is a
mixed random variable since
' 0
P [W = 0] = P [U < 0] = fU (u) du = 1/2 (2)
−1
The above facts imply that
FW (0) = P [W ≤ 0] = P [W = 0] = 1/2 (3)
Next, we note that for 0 < w < 1,

' w
FW (w) = P [U ≤ w] = fU (u) du = (w + 1)/2 (4)
−1
Finally, U ≤ 1 implies W ≤ 1, which implies FW (w) = 1 for w ≥ 1. Hence, the complete expression
for the CDF is ⎧
⎨ 0 w<0
FW (w) = (w + 1)/2 0 ≤ w ≤ 1 (5)
⎩
1 w>1
By taking the derivative of the CDF, we find the PDF of W ; however, we must keep in mind that
the discontinuity in the CDF at w = 0 yields a corresponding impulse in the PDF.

(δ(w) + 1)/2 0 ≤ w ≤ 1
fW (w) = (6)
0 otherwise
From the PDF, we can calculate the expected value

' 1 ' 1
E [W ] = w(δ(w) + 1)/2 dw = 0 + (w/2) dw = 1/4 (7)
0 0
Perhaps an easier way to find the expected value is to use Theorem 2.10. In this case,
' ∞ ' 1
E [W ] = g(u)fW (w) du = u(1/2) du = 1/4 (8)
−∞ 0
As we expect, both approaches give the same answer.

Theorem 3.19 states that for a constant a > 0, Y = aX has CDF and PDF
1
FY (y) = FX (y/a) fY (y) = fX (y/a) (1)
a
(a) If X is uniform (b, c), then Y = aX has PDF

a(c−b) b ≤ y/a ≤ c ab ≤ y ≤ ac
1 1
1 ac−ab
fY (y) = fX (y/a) = = (2)
a 0 otherwise 0 otherwise
Thus Y has the PDF of a uniform (ab, ac) random variable.
106
(b) Using Theorem 3.19, the PDF of Y = aX is
λ −λ(y/a)
1 ae y/a ≥ 0
fY (y) = fX (y/a) = (3)
a 0 otherwise
−(λ/a)y
(λ/a)e y≥0
= (4)
0 otherwise
Hence Y is an exponential (λ/a) exponential random variable.
(c) Using Theorem 3.19, the PDF of Y = aX is
n
λ (y/a)n−1 e−λ(y/a)
1 a(n−1)! y/a ≥ 0
fY (y) = fX (y/a) = (5)
a 0 otherwise
n n−1 −(λ/a)y
(λ/a) y e
(n−1)! y ≥ 0,
= (6)
0 otherwise,
which is an Erlang (n, λ) PDF.
(d) If X is a Gaussian (µ, σ) random variable, then Y = aX has PDF
1 2 2
fY (y) = fX (y/a) = √ e−((y/a)−µ) /2σ (7)
a 2πσ 2
1 2 2 2
=√ e−(y−aµ) /2(a σ ) (8)
2
2πa σ 2
(9)
Thus Y is a Gaussian random variable with expected value E[Y ] = aµ and Var[Y ] = a2 σ 2 .
That is, Y is a Gaussian (aµ, aσ) random variable.

If X has a uniform distribution from 0 to 1 then the PDF and corresponding CDF of X are
⎧
⎨ 0 x<0
1 0≤x≤1
fX (x) = FX (x) = x 0≤x≤1 (1)
0 otherwise ⎩
1 x>1
For b − a > 0, we can find the CDF of the function Y = a + (b − a)X
FY (y) = P [Y ≤ y] = P [a + (b − a)X ≤ y] (2)

y−a
=P X≤ (3)
b−a

y−a y−a
= FX = (4)
b−a b−a
Therefore the CDF of Y is ⎧
⎨ 0 y<a
FY (y) = y−a
a≤y≤b (5)
⎩ b−a
1 y≥b
By differentiating with respect to y we arrive at the PDF

1/(b − a) a ≤ x ≤ b
fY (y) = (6)
0 otherwise
which we recognize as the PDF of a uniform (a, b) random variable.
107
Since X = F −1 (U ), it is desirable that the function F −1 (u) exist for all 0 ≤ u ≤ 1. However, for the
continuous uniform random variable U , P [U = 0] = P [U = 1] = 0. Thus, it is a zero probability
event that F −1 (U ) will be evaluated at U = 0 or U = 1. Asa result, it doesn’t matter whether
F −1 (u) exists at u = 0 or u = 1.

The relationship between X and Y is shown in the following figure:
3
2
Y
1
0
0 1 2 3
X
(a) Note that Y = 1/2 if and only if 0 ≤ X ≤ 1. Thus,

' 1 ' 1
P [Y = 1/2] = P [0 ≤ X ≤ 1] = fX (x) dx = (x/2) dx = 1/4 (1)
0 0
(b) Since Y ≥ 1/2, we can conclude that FY (y) = 0 for y < 1/2. Also, FY (1/2) = P [Y = 1/2] =
1/4. Similarly, for 1/2 < y ≤ 1,
FY (y) = P [0 ≤ X ≤ 1] = P [Y = 1/2] = 1/4 (2)
Next, for 1 < y ≤ 2, ' y

FY (y) = P [X ≤ y] = fX (x) dx = y 2 /4 (3)
0
Lastly, since Y ≤ 2, FY (y) = 1 for y ≥ 2. The complete expression of the CDF is
⎧
⎪
⎪ 0 y < 1/2
⎨
1/4 1/2 ≤ y ≤ 1
FY (y) = (4)
⎪
⎪ y 2 /4 1 < y < 2
⎩
1 y≥2

We can prove the assertion by considering the cases where a > 0 and a < 0, respectively. For the
case where a > 0 we have

y−b y−b
FY (y) = P [Y ≤ y] = P X ≤ = FX (1)
a a
Therefore by taking the derivative we find that

1 y−b
fY (y) = fX a>0 (2)
a a
108
Similarly for the case when a < 0 we have

y−b y−b
FY (y) = P [Y ≤ y] = P X ≥ = 1 − FX (3)
a a
And by taking the derivative, we find that for negative a,

1 y−b
fY (y) = − fX a<0 (4)
a a
A valid expression for both positive and negative a is

1 y−b
fY (y) = fX (5)
|a| a
Therefore the assertion is proved.

Understanding this claim may be harder than completing the proof. Since 0 ≤ F (x) ≤ 1, we know
that 0 ≤ U ≤ 1. This implies FU (u) = 0 for u < 0 and FU (u) = 1 for u ≥ 1. Moreover, since F (x)
is an increasing function, we can write for 0 ≤ u ≤ 1,

FU (u) = P [F (X) ≤ u] = P X ≤ F −1 (u) = FX F −1 (u) (1)
Since FX (x) = F (x), we have for 0 ≤ u ≤ 1,
FU (u) = F (F −1 (u)) = u (2)
Hence the complete CDF of U is

⎧
⎨ 0 u<0
FU (u) = u 0≤u<1 (3)
⎩
1 u≥1
That is, U is a uniform [0, 1] random variable.
(a) Given FX (x) is a continuous function, there exists x0 such that FX (x0 ) = u. For each
value of u, the corresponding x0 is unique. To see this, suppose there were also x1 such
that FX (x1 ) = u. Without loss of generality, we can assume x1 > x0 since otherwise we
could exchange the points x0 and x1 . Since FX (x0 ) = FX (x1 ) = u, the fact that FX (x)
is nondecreasing implies FX (x) = u for all x ∈ [x0 , x1 ], i.e., FX (x) is flat over the interval
[x0 , x1 ], which contradicts the assumption that FX (x) has no flat intervals. Thus, for any
u ∈ (0, 1), there is a unique x0 such that FX (x) = u. Moreiver, the same x0 is the minimum
of all x such that FX (x ) ≥ u. The uniqueness of x0 such that FX (x)x0 = u permits us to
define F̃ (u) = x0 = FX−1 (u).
(b) In this part, we are given that FX (x) has a jump discontinuity at x0 . That is, there exists
u− − + + − + − +
0 = FX (x0 ) and u0 = FX (x0 ) with u0 < u0 . Consider any u in the interval [u0 , u0 ].
Since FX (x0 ) = FX (x+
0 ) and FX (x) is nondecreasing,
FX (x) ≥ FX (x0 ) = u+
0, x ≥ x0 . (1)
109
Moreover,
FX (x) < FX x− −
0 = u0 , x < x0 . (2)
Thus for any u satisfying u−
o ≤ u ≤ u0 , FX (x) < u for x < x0 and FX (x) ≥ u for x ≥ x0 .
+
Thus, F̃ (u) = min{x|FX (x) ≥ u} = x0 .
(c) We note that the first two parts of this problem were just designed to show the properties of
F̃ (u). First, we observe that

P X̂ ≤ x = P F̃ (U ) ≤ x = P min x |FX x ≥ U ≤ x . (3)
To prove the claim, we define, for any x, the events

A : min x |FX x ≥ U ≤ x, (4)
B : U ≤ FX (x) . (5)
Note that P [A] = P [X̂ ≤ x]. In addition, P [B] = P [U ≤ FX (x)] = FX (x) since P [U ≤ u] = u
for any u ∈ [0, 1].
We will show that the events A and B are the same. This fact implies

P X̂ ≤ x = P [A] = P [B] = P [U ≤ FX (x)] = FX (x) . (6)
All that remains is to show A and B are the same. As always, we need to show that A ⊂ B
and that B ⊂ A.
• To show A ⊂ B, suppose A is true and min{x |FX (x ) ≥ U } ≤ x. This implies there
exists x0 ≤ x such that FX (x0 ) ≥ U . Since x0 ≤ x, it follows from FX (x) being
nondecreasing that FX (x0 ) ≤ FX (x). We can thus conclude that
U ≤ FX (x0 ) ≤ FX (x) . (7)
That is, event B is true.

• To show B ⊂ A, we suppose event B is true so that U ≤ FX (x). We define the set

L = x |FX x ≥ U . (8)
We note x ∈ L. It follows that the minimum element min{x |x ∈ L} ≤ x. That is,

min x |FX x ≥ U ≤ x, (9)
which is simply event A.

The PDF of X is
1/10 −5 ≤ x ≤ 5
fX (x) = (1)
0 otherwise
110
(a) The event B has probability
' 3
1 3
P [B] = P [−3 ≤ X ≤ 3] = dx = (2)
−3 10 5
From Definition 3.15, the conditional PDF of X given B is

fX (x) /P [B] x ∈ B 1/6 |x| ≤ 3
fX|B (x) = = (3)
(b) Given B, we see that X has a uniform PDF over [a, b] with a = −3 and b = 3. From
Theorem 3.6, the conditional expected value of X is E[X|B] = (a + b)/2 = 0.
(c) From Theorem 3.6, the conditional variance of X is Var[X|B] = (b − a)2 /12 = 3.

From Definition 3.6, the PDF of Y is

(1/5)e−y/5 y ≥ 0
fY (y) = (1)
0 otherwise
(a) The event A has probability

' 2 $2
−y/5 −y/5 $
P [A] = P [Y < 2] = (1/5)e dy = −e $ = 1 − e−2/5 (2)
0 0
From Definition 3.15, the conditional PDF of Y given A is

fY (y) /P [A] x ∈ A
fY |A (y) = (3)
0 otherwise

(1/5)e−y/5 /(1 − e−2/5 ) 0 ≤ y < 2
= (4)
0 otherwise
(b) The conditional expected value of Y given A is

' ∞ ' 2
1/5
E [Y |A] = yfY |A (y) dy = ye−y/5 dy (5)
−∞ 1 − e−2/5 0
( (
Using the integration by parts formula u dv = uv − v du with u = y and dv = e−y/5 dy
yields
$2 ' 2
1/5 −y/5 $ −y/5
E [Y |A] = −5ye $ + 5e dy (6)
1 − e−2/5 0
0
$2
1/5 −2/5 −y/5 $
= −10e − 25e $ (7)
1 − e−2/5 0
5 − 7e−2/5
= (8)
1 − e−2/5
111
The condition right side of the circle is R = [0, 1/2]. Using the PDF in Example 3.5, we have
' 1/2 ' 1/2
P [R] = fY (y) dy = 3y 2 dy = 1/8 (1)
0 0
Therefore, the conditional PDF of Y given event R is

24y 2 0 ≤ y ≤ 1/2
fY |R (y) = (2)
0 otherwise
The conditional expected value and mean square value are

' ∞ ' 1/2
E [Y |R] = yfY |R (y) dy = 24y 3 dy = 3/8 meter (3)
−∞ 0
' ∞ '
1/2
E Y |R =
2 2
y fY |R (y) dy = 24y 4 dy = 3/20 m2 (4)
−∞ 0
The conditional variance is

2
3 3
Var [Y |R] = E Y |R − (E [Y |R]) =
2
− 2
= 3/320 m2 (5)
20 8

The conditional standard deviation is σY |R = Var[Y |R] = 0.0968 meters.

From Definition 3.8, the PDF of W is
1 2
fW (w) = √ e−w /32 (1)
32π
(a) Since W has expected value µ = 0, fW (w) is symmetric about w = 0. Hence P [C] =
P [W > 0] = 1/2. From Definition 3.15, the conditional PDF of W given C is
−w2 /32 √
fW (w) /P [C] w ∈ C 2e / 32π w > 0
fW |C (w) = = (2)
(b) The conditional expected value of W given C is

' ∞ ' ∞
2 2
E [W |C] = wfW |C (w) dw = √ we−w /32 dw (3)
−∞ 4 2π 0
Making the substitution v = w2 /32, we obtain

' ∞
32 32
E [W |C] = √ e−v dv = √ (4)
32π 0 32π
(c) The conditional second moment of W is

' ∞ ' ∞
2
E W |C = 2
w fW |C (w) dw = 2 w2 fW (w) dw (5)
−∞ 0
112
We observe that w2 fW (w) is an even function. Hence
' ∞

E W 2 |C = 2 w2 fW (w) dw (6)
' ∞0

= w2 fW (w) dw = E W 2 = σ 2 = 16 (7)
−∞
Lastly, the conditional variance of W given C is

Var[W |C] = E W 2 |C − (E [W |C])2 = 16 − 32/π = 5.81 (8)
(a) We first find the conditional PDF of T . The PDF of T is

100e−100t t ≥ 0
fT (t) = (1)
0 otherwise
The conditioning event has probability
' ∞
$∞
P [T > 0.02] = fT (t) dt = −e−100t $0.02 = e−2 (2)
0.02
From Definition 3.15, the conditional PDF of T is

100e−100(t−0.02) t ≥ 0.02
fT (t)
P [T >0.02] t ≥ 0.02
fT |T >0.02 (t) = = (3)
The conditional expected value of T is

' ∞
E [T |T > 0.02] = t(100)e−100(t−0.02) dt (4)
0.02
The substitution τ = t − 0.02 yields

' ∞
E [T |T > 0.02] = (τ + 0.02)(100)e−100τ dτ (5)
'0 ∞
= (τ + 0.02)fT (τ ) dτ = E [T + 0.02] = 0.03 (6)
0
(b) The conditional second moment of T is

' ∞

E T 2 |T > 0.02 = t2 (100)e−100(t−0.02) dt (7)
0.02
The substitution τ = t − 0.02 yields

' ∞

E T 2 |T > 0.02 = (τ + 0.02)2 (100)e−100τ dτ (8)
'0 ∞
= (τ + 0.02)2 fT (τ ) dτ (9)
0

= E (T + 0.02)2 (10)
113
Now we can calculate the conditional variance.

Var[T |T > 0.02] = E T 2 |T > 0.02 − (E [T |T > 0.02])2 (11)

= E (T + 0.02)2 − (E [T + 0.02])2 (12)
= Var[T + 0.02] (13)
= Var[T ] = 0.01 (14)
(a) In Problem 3.6.8, we found that the PDF of D is

0.3δ(y) y < 60
fD (y) = (1)
0.07e−(y−60)/10 y ≥ 60
First, we observe that D > 0 if the throw is good so that P [D > 0] = 0.7. A second way to
find this probability is ' ∞
P [D > 0] = fD (y) dy = 0.7 (2)
0+
From Definition 3.15, we can write

(1/10)e−(y−60)/10 y ≥ 60
fD (y)
P [D>0] y>0
fD|D>0 (y) = = (3)
(b) If instead we learn that D ≤ 70, we can calculate the conditional PDF by first calculating
' 70
P [D ≤ 70] = fD (y) dy (4)
0
' 60 ' 70
= 0.3δ(y) dy + 0.07e−(y−60)/10 dy (5)
0 60
$70
−(y−60)/10 $
= 0.3 + −0.7e $ = 1 − 0.7e−1 (6)
60
The conditional PDF is

fD (y)
P [D≤70] y ≤ 70
fD|D≤70 (y) = (7)
0 otherwise
⎧
⎨
0.3
1−0.7e−1
δ(y) 0 ≤ y < 60
= 0.07
1−0.7e−1
e−(y−60)/10 60 ≤ y ≤ 70 (8)
⎩
0 otherwise
(a) Given that a person is healthy, X is a Gaussian (µ = 90, σ = 20) random variable. Thus,
1 2 2 1 2
fX|H (x) = √ e−(x−µ) /2σ = √ e−(x−90) /800 (1)
σ 2π 20 2π
114
(b) Given the event H, we use the conditional PDF fX|H (x) to calculate the required probabilities

P T + |H = P [X ≥ 140|H] = P [X − 90 ≥ 50|H] (2)

X − 90
=P ≥ 2.5|H = 1 − Φ(2.5) = 0.006 (3)
20
Similarly,

P T − |H = P [X ≤ 110|H] = P [X − 90 ≤ 20|H] (4)

X − 90
=P ≤ 1|H = Φ(1) = 0.841 (5)
20
(c) Using Bayes Theorem, we have
P [T − |H] P [H] P [T − |H] P [H]

P H|T − = = (6)
P [T − ] P [T − |D] P [D] + P [T − |H] P [H]
In the denominator, we need to calculate

P T − |D = P [X ≤ 110|D] = P [X − 160 ≤ −50|D] (7)

X − 160
=P ≤ −1.25|D (8)
40
= Φ(−1.25) = 1 − Φ(1.25) = 0.106 (9)
Thus,
P [T − |H] P [H]
P H|T − = (10)
P [T − |D] P [D] + P [T − |H] P [H]
0.841(0.9)
= = 0.986 (11)
0.106(0.1) + 0.841(0.9)
(d) Since T − , T 0 , and T + are mutually exclusive and collectively exhaustive,

P T 0 |H = 1 − P T − |H − P T + |H = 1 − 0.841 − 0.006 = 0.153 (12)
We say that a test is a failure if the result is T 0 . Thus, given the event H, each test has
conditional failure probability of q = 0.153, or success probability p = 1 − q = 0.847. Given
H, the number of trials N until a success is a geometric (p) random variable with PMF

(1 − p)n−1 p n = 1, 2, . . . ,
PN |H (n) = (13)
0 otherwise.
115
(a) The event Bi that Y = ∆/2 + i∆ occurs if and only if i∆ ≤ X < (i + 1)∆. In particular,
since X has the uniform (−r/2, r/2) PDF

1/r −r/2 ≤ x < r/2,
fX (x) = (1)
0 otherwise,
we observe that ' (i+1)∆
1 ∆
P [Bi ] = dx = (2)
i∆ r r
In addition, the conditional PDF of X given Bi is

fX (x) /P [B] x ∈ Bi 1/∆ i∆ ≤ x < (i + 1)∆
fX|Bi (x) = = (3)
It follows that given Bi , Z = X − Y = X − ∆/2 − i∆, which is a uniform (−∆/2, ∆/2)
random variable. That is,

1/∆ −∆/2 ≤ z < ∆/2
fZ|Bi (z) = (4)
0 otherwise
(b) We observe that fZ|Bi (z) is the same for every i. Thus, we can write

fZ (z) = P [Bi ] fZ|Bi (z) = fZ|B0 (z) P [Bi ] = fZ|B0 (z) (5)
i i
Thus, Z is a uniform (−∆/2, ∆/2) random variable. From the definition of a uniform (a, b)
random variable, Z has mean and variance
(∆/2 − (−∆/2))2 ∆2
E [Z] = 0, Var[Z] = = . (6)
12 12

For this problem, almost any non-uniform random variable X will yield a non-uniform random
variable Z. For example, suppose X has the “triangular” PDF

8x/r2 0 ≤ x ≤ r/2
fX (x) = (1)
0 otherwise
In this case, the event Bi that Y = i∆ + ∆/2 occurs if and only if i∆ ≤ X < (i + 1)∆. Thus
' (i+1)∆
8x 8∆(i∆ + ∆/2)
P [Bi ] = 2
dx = (2)
i∆ r r2
It follows that the conditional PDF of X given Bi is

fX (x)
P [B ] x ∈ Bi x
∆(i∆+∆/2) i∆ ≤ x < (i + 1)∆
fX|Bi (x) = i = (3)
Given event Bi , Y = i∆ + ∆/2, so that Z = X − Y = X − i∆ − ∆/2. This implies

z+i∆+∆/2
fZ|Bi (z) = fX|Bi (z + i∆ + ∆/2) = ∆(i∆+∆/2) −∆/2 ≤ z < ∆/2 (4)
0 otherwise
We observe that the PDF of Z depends on which event Bi occurs. Moreover, fZ|Bi (z) is non-uniform
for all Bi .
116
Taking the derivative of the CDF FY (y) in Quiz 3.1, we obtain

1/4 0 ≤ y ≤ 4
fY (y) = (1)
0 otherwise
We see that Y is a uniform (0, 4) random variable. By Theorem 3.20, if X is a uniform (0, 1)
random variable, then Y = 4X is a uniform (0, 4) random variable. Using rand as Matlab’s
uniform (0, 1) random variable, the program quiz31rv is essentially a one line program:
function y=quiz31rv(m)
%Usage y=quiz31rv(m)
%Returns the vector y holding m
%samples of the uniform (0,4) random
%variable Y of Quiz 3.1
y=4*rand(m,1);

The modem receiver voltage is genrated by taking a ±5 voltage representing data, and adding to
it a Gaussian (0, 2) noise variable. Although siuations in which two random variables are added
together are not analyzed until Chapter 4, generating samples of the receiver voltage is easy in
Matlab. Here is the code:
function x=modemrv(m);
%Usage: x=modemrv(m)
%generates m samples of X, the modem
%receiver voltage in Exampe 3.32.
%X=+-5 + N where N is Gaussian (0,2)
sb=[-5; 5]; pb=[0.5; 0.5];
b=finiterv(sb,pb,m);
noise=gaussrv(0,2,m);
x=b+noise;
The commands
x=modemrv(10000); hist(x,100);
generate 10,000 sample of the modem receiver voltage and plots the relative frequencies using 100
bins. Here is an example plot:
300
250
Relative Frequency
200
150
100
50
0
−15 −10 −5 0 5 10 15
x
As expected, the result is qualitatively similar (“hills” around X = −5 and X = 5) to the sketch
in Figure 3.3.
117
The code for Q̂(z) is the Matlab function
function p=qapprox(z);
%approximation to the Gaussian
% (0,1) complementary CDF Q(z)
t=1./(1.0+(0.231641888.*z(:)));
a=[0.127414796; -0.142248368; 0.7107068705; ...
-0.7265760135; 0.5307027145];
p=([t t.^2 t.^3 t.^4 t.^5]*a).*exp(-(z(:).^2)/2);
This code generates two plots of the relative error e(z) as a function of z:
z=0:0.02:6;
q=1.0-phi(z(:));
qhat=qapprox(z);
e=(q-qhat)./q;
plot(z,e); figure;
semilogy(z,abs(e));
Here are the output figures of qtest.m:

−3
x 10
1 −2
10
0
−4
10
−1
e(z)
−6
−2 10
−3 −8
10
−4
0 2 4 6 −10
10
z 0 2 4 6
The left side plot graphs e(z) versus z. It appears that the e(z) = 0 for z ≤ 3. In fact, e(z) is
nonzero over that range, but the relative error is so small that it isn’t visible in comparison to
e(6) ≈ −3.5 × 10−3 . To see the error for small z, the right hand graph plots |e(z)| versus z in log
scale where we observe very small relative errors on the order of 10−7 .

By Theorem 3.9, if X is an exponential (λ) random variable, then K = X is a geometric (p)
random variable with p = 1 − e−λ . Thus, given p, we can write λ = − ln(1 − p) and X is a
geometric (p) random variable. Here is the Matlab function that implements this technique:
function k=georv(p,m);
lambda= -log(1-p);
k=ceil(exponentialrv(lambda,m));
To compare this technique with that use in geometricrv.m, we first examine the code for exponentialrv.m:
function x=exponentialrv(lambda,m)
x=-(1/lambda)*log(1-rand(m,1));
118
To analyze how m = 1 random sample is generated, let R = rand(1,1). In terms of mathematics,
exponentialrv(lambda,1) generates the random variable
ln(1 − R)
X=− (1)
λ
For λ = − ln(1 − p), we have that
% &
ln(1 − R)
K = X = (2)
ln(1 − p)
This is precisely the same function implemented by geometricrv.m. In short, the two methods for
generating geometric (p) random samples are one in the same.

Given 0 ≤ u ≤ 1, we need to find the “inverse” function that finds the value of w satisfying
u = FW (w). The problem is that for u = 1/4, any w in the interval [−3, 3] satisfies FW (w) = 1/4.
However, in terms of generating samples of random variable W , this doesn’t matter. For a uniform
(0, 1) random variable U , P [U = 1/4] = 0. Thus we can choose any w ∈ [−3, 3]. In particular, we
define the inverse CDF as

−1 8u − 5 0 ≤ u ≤ 1/4
w = FW (u) = (1)
(8u + 7)/3 1/4 < u ≤ 1
−1
Note that because 0 ≤ FW (w) ≤ 1, the inverse FW (u) is defined only for 0 ≤ u ≤ 1. Careful
inspection will show that u = (w + 5)/8 for −5 ≤ w < −3 and that u = 1/4 + 3(w − 3)/8 for
−1
−3 ≤ w ≤ 5. Thus, for a uniform (0, 1) random variable U , the function W = FW (U ) produces a
random variable with CDF FW (w). To implement this solution in Matlab, we define
function w=iwcdf(u);
w=((u>=0).*(u <= 0.25).*(8*u-5))+...
((u > 0.25).*(u<=1).*((8*u+7)/3));
so that the Matlab code W=icdfrv(@iwcdf,m) generates m samples of random variable W .
(a) To test the exponential random variables, the following code
function exponentialtest(lambda,n)
delta=0.01;
x=exponentialrv(lambda,n);
xr=(0:delta:(5.0/lambda))’;
fxsample=(histc(x,xr)/(n*delta));
fx=exponentialpdf(lambda,xr);
plot(xr,fx,xr,fxsample);
generates n samples of an exponential λ random variable and plots the relative frequency
ni /(n∆) against the corresponding exponential PDF. Note that the histc function generates
a histogram using xr to define the edges of the bins. Two representative plots for n = 1,000
and n = 100,000 samples appear in the following figure:
119
1.5 1.5
1 1
0.5 0.5
0 0
0 1 2 3 4 5 0 1 2 3 4 5
exponentialtest(1,1000) exponentialtest(1,100000)
For n = 1,000, the jaggedness of the relative frequency occurs because δ is sufficiently small
that the number of sample of X in each bin i∆ < X ≤ (i+1)∆ is fairly small. For n = 100,000,
the greater smoothness of the curve demonstrates how the relative frequency is becoming a
better approximation to the actual PDF.
(b) Similar results hold for Gaussian random variables. The following code generates the same
comparison between the Gaussian PDF and the relative frequency of n samples.
function gausstest(mu,sigma2,n)
delta=0.01;
x=gaussrv(mu,sigma2,n);
xr=(0:delta:(mu+(3*sqrt(sigma2))))’;
fxsample=(histc(x,xr)/(n*delta));
fx=gausspdf(mu,sigma2,xr);
plot(xr,fx,xr,fxsample);
Here are two typical plots produced by gaussiantest.m:
1 0.5
0.8 0.4
0.6 0.3
0.4 0.2
0.2 0.1
0 0
0 2 4 6 0 2 4 6
gausstest(3,1,1000) gausstest(3,1,100000)

First we need to build a uniform (−r/2, r/2) b-bit quantizer. The function uquantize does this.
function y=uquantize(r,b,x)
%uniform (-r/2,r/2) b bit quantizer
n=2^b;
delta=r/n;
x=min(x,(r-delta/2)/2);
x=max(x,-(r-delta/2)/2);
y=(delta/2)+delta*floor(x/delta);
120
Note that if |x| > r/2, then x is truncated so that the quantizer output has maximum amplitude.
Next, we generate Gaussian samples, quantize them and record the errors:
function stdev=quantizegauss(r,b,m)
x=gaussrv(0,1,m);
x=x((x<=r/2)&(x>=-r/2));
y=uquantize(r,b,x);
z=x-y;
hist(z,100);
stdev=sqrt(sum(z.^2)/length(z));
For a Gaussian random variable X, P [|X| > r/2] > 0 for any value of r. When we generate enough
Gaussian samples, we will always see some quantization errors due to the finite (−r/2, r/2) range.
To focus our attention on the effect of b bit quantization, quantizegauss.m eliminates Gaussian
samples outside the range (−r/2, r/2). Here are outputs of quantizegauss for b = 1, 2, 3 bits.
15000 15000 15000
10000 10000 10000
5000 5000 5000
0 0 0
−2 0 2 −1 0 1 −0.5 0 0.5
b=1 b=2 b=3
It is obvious that for b = 1 bit quantization, the error is decidely not uniform. However, it appears
that the error is uniform for b = 2 and b = 3. You can verify that uniform errors is a reasonable
model for larger values of b.

To solve this problem, we want to use Theorem 3.22. One complication is that in the theorem,
U denotes the uniform random variable while X is the derived random variable. In this problem,
we are using U for the random variable we want to derive. As a result, we will use Theorem 3.22
with the roles of X and U reversed. Given U with CDF FU (u) = F (u), we need to find the inverse
functon F −1 (x) = FU−1 (x) so that for a uniform (0, 1) random variable X, U = F −1 (X).
Recall that random variable U defined in Problem
⎧ 3.3.7 has CDF
1 ⎪
⎪ 0 u < −5
⎪
⎪
⎨ −5 ≤ u < −3
FU(u)
(u + 5)/8
0.5
FU (u) = 1/4 −3 ≤ u < 3 (1)
⎪
⎪
0 ⎪
⎪ 1/4 + 3(u − 3)/8 3 ≤ u < 5
−5 0 5 ⎩
u 1 u ≥ 5.
At x = 1/4, there are multiple values of u such that FU (u) = 1/4. However, except for x = 1/4,
the inverse FU−1 (x) is well defined over 0 < x < 1. At x = 1/4, we can arbitrarily define a value for
FU−1 (1/4) because when we produce sample values of FU−1 (X), the event X = 1/4 has probability
zero. To generate the inverse CDF, given a value of x, 0 < x < 1, we ave to find the value of u
such that x = FU (u). From the CDF we see that
1 u+5
0≤x≤ ⇒x= (2)
4 8
1 1 3
<x≤1 ⇒ x = + (u − 3) (3)
4 4 8
(4)
121
These conditions can be inverted to express u as a function of x.

−1 8x − 5 0 ≤ x ≤ 1/4
u = F (x) = (5)
(8x + 7)/3 1/4 < x ≤ 1
In particular, when X is a uniform (0, 1) random variable, U = F −1 (X) will generate samples of
the rndom variable U . A Matlab program to implement this solution is now straightforward:
function u=urv(m)
%Usage: u=urv(m)
%Generates m samples of the random
%variable U defined in Problem 3.3.7
x=rand(m,1);
u=(x<=1/4).*(8*x-5);
u=u+(x>1/4).*(8*x+7)/3;
To see that this generates the correct output, we can generate a histogram of a million sample
values of U using the commands
u=urv(1000000); hist(u,100);
The output
4 is shown in the following graph, alongside the corresponding PDF of U .
x 10
4
⎧
3 ⎪
⎪ 0 u < −5
⎪
⎪
⎨ 1/8 −5 ≤ u < −3
Count
2
fU (u) = 0 −3 ≤ u < 3 (6)
⎪
⎪
1 ⎪
⎪ 3/8 3≤u<5
⎩
0 0 u ≥ 5.
−5 0 5
u
Note that the scaling constant 104 on the histogram plot comes from the fact that the histogram
was generated using 106 sample points and 100 bins. The width of each bin is ∆ = 10/100 = 0.1.
Consider a bin of idth ∆ centered at u0 . A sample value of U would fall in that bin with probability
fU (u0 )∆. Given that we generate m = 106 samples, we would expect about mfU (u0 )∆ = 105 fU (u0 )
samples in each bin. For −5 < u0 < −3, we would expect to see about 1.25 × 104 samples in each
bin. For 3 < u0 < 5, we would expect to see about 3.75 × 104 samples in each bin. As can be seen,
these conclusions are consistent with the histogam data.
Finally, we comment that if you generate histograms for a range of values of m, the number of
samples, you will see that the histograms will become more and more similar to a scaled version of
the PDF. This gives the (false) impression that any bin centered on u0 has a number of samples
increasingly close to mfU (u0 )∆. Because the histpgram is always the same height, what is actually
happening is that the vertical axis is effectively scaled by 1/m and the height of a histogram bar is
proportional to the fraction of m samples that land in that bin. We will see in Chapter 7 that the
fraction of samples in a bin does converge to the probability of a sample being in that bin as the
number of samples m goes to infinity.

From Quiz 3.6, random variable X has CDF The CDF of X is
1
⎧
⎨ 0 x < −1,
FX(x)
0.5
FX (x) = (x + 1)/4 −1 ≤ x < 1, (1)
⎩
0 1 x ≥ 1.
−2 0 2
x
122
Following the procedure outlined in Problem 3.7.18, we define for 0 < u ≤ 1,
F̃ (u) = min {x|FX (x) ≥ u} . (2)
We observe that if 0 < u < 1/4, then we can choose x so that FX (x) = u. In this case, (x+1)/4 = u,
or equivalently, x = 4u − 1. For 1/4 ≤ u ≤ 1, the minimum x that satisfies FX (x) ≥ u is x = 1.
These facts imply
4u − 1 0 < u < 1/4
F̃ (u) = (3)
1 1/4 ≤ u ≤ 1
It follows that if U is a uniform (0, 1) random variable, then F̃ (U ) has the same CDF as X. This
is trivial to implement in Matlab.
function x=quiz36rv(m)
%Usage x=quiz36rv(m)
%Returns the vector x holding m samples
%of the random variable X of Quiz 3.6
u=rand(m,1);
x=((4*u-1).*(u< 0.25))+(1.0*(u>=0.25));
123
(a) The probability P [X ≤ 2, Y ≤ 3] can be found be evaluating the joint CDF FX,Y (x, y) at
x = 2 and y = 3. This yields
P [X ≤ 2, Y ≤ 3] = FX,Y (2, 3) = (1 − e−2 )(1 − e−3 ) (1)
(b) To find the marginal CDF of X, FX (x), we simply evaluate the joint CDF at y = ∞.

1 − e−x x ≥ 0
FX (x) = FX,Y (x, ∞) = (2)
0 otherwise
(c) Likewise for the marginal CDF of Y , we evaluate the joint CDF at X = ∞.

1 − e−y y ≥ 0
FY (y) = FX,Y (∞, y) = (3)
0 otherwise
(a) Because the probability that any random variable is less than −∞ is zero, we have
FX,Y (x, −∞) = P [X ≤ x, Y ≤ −∞] ≤ P [Y ≤ −∞] = 0 (1)
(b) The probability that any random variable is less than infinity is always one.
FX,Y (x, ∞) = P [X ≤ x, Y ≤ ∞] = P [X ≤ x] = FX (x) (2)
(c) Although P [Y ≤ ∞] = 1, P [X ≤ −∞] = 0. Therefore the following is true.
FX,Y (−∞, ∞) = P [X ≤ −∞, Y ≤ ∞] ≤ P [X ≤ −∞] = 0 (3)
(d) Part (d) follows the same logic as that of part (a).
FX,Y (−∞, y) = P [X ≤ −∞, Y ≤ y] ≤ P [X ≤ −∞] = 0 (4)
(e) Analogous to Part (b), we find that
FX,Y (∞, y) = P [X ≤ ∞, Y ≤ y] = P [Y ≤ y] = FY (y) (5)
124
We wish to find P [x1 ≤ X ≤ x2 ] or P [y1 ≤ Y ≤ y2 ]. We define events A = {y1 ≤ Y ≤ y2 } and
B = {y1 ≤ Y ≤ y2 } so that
P [A ∪ B] = P [A] + P [B] − P [AB] (1)
Keep in mind that the intersection of events A and B are all the outcomes such that both A and
B occur, specifically, AB = {x1 ≤ X ≤ x2 , y1 ≤ Y ≤ y2 }. It follows that
P [A ∪ B] = P [x1 ≤ X ≤ x2 ] + P [y1 ≤ Y ≤ y2 ]
− P [x1 ≤ X ≤ x2 , y1 ≤ Y ≤ y2 ] . (2)
By Theorem 4.5,
P [x1 ≤ X ≤ x2 , y1 ≤ Y ≤ y2 ]
= FX,Y (x2 , y2 ) − FX,Y (x2 , y1 ) − FX,Y (x1 , y2 ) + FX,Y (x1 , y1 ) . (3)
Expressed in terms of the marginal and joint CDFs,
P [A ∪ B] = FX (x2 ) − FX (x1 ) + FY (y2 ) − FY (y1 ) (4)
− FX,Y (x2 , y2 ) + FX,Y (x2 , y1 ) + FX,Y (x1 , y2 ) − FX,Y (x1 , y1 ) (5)

Its easy to show that the properties of Theorem 4.1 are satisfied. However, those properties are
necessary but not sufficient to show F (x, y) is a CDF. To convince ourselves that F (x, y) is a valid
CDF, we show that for all x1 ≤ x2 and y1 ≤ y2 ,
P [x1 < X1 ≤ x2 , y1 < Y ≤ y2 ] ≥ 0 (1)
In this case, for x1 ≤ x2 and y1 ≤ y2 , Theorem 4.5 yields
P [x1 < X ≤ x2 , y1 < Y ≤ y2 ] = F (x2 , y2 ) − F (x1 , y2 ) − F (x2 , y1 ) + F (x1 , y1 ) (2)
= FX (x2 ) FY (y2 ) − FX (x1 ) FY (y2 ) (3)
− FX (x2 ) FY (y1 ) + FX (x1 ) FY (y1 ) (4)
= [FX (x2 ) − FX (x1 )][FY (y2 ) − FY (y1 )] (5)
≥0 (6)
Hence, FX (x)FY (y) is a valid joint CDF.

In this problem, we prove Theorem 4.5 which states
P [x1 < X ≤ x2 , y1 < Y ≤ y2 ] = FX,Y (x2 , y2 ) − FX,Y (x2 , y1 ) (1)
− FX,Y (x1 , y2 ) + FX,Y (x1 , y1 ) (2)
(a) The events A, B, and C are
Y Y Y
y2 y2 y2
y1 y1 y1
(3)
X X X
x1 x1 x2 x1 x2
A B C
125
(b) In terms of the joint CDF FX,Y (x, y), we can write
P [A] = FX,Y (x1 , y2 ) − FX,Y (x1 , y1 ) (4)

P [B] = FX,Y (x2 , y1 ) − FX,Y (x1 , y1 ) (5)
P [A ∪ B ∪ C] = FX,Y (x2 , y2 ) − FX,Y (x1 , y1 ) (6)
(c) Since A, B, and C are mutually exclusive,
P [A ∪ B ∪ C] = P [A] + P [B] + P [C] (7)
However, since we want to express
P [C] = P [x1 < X ≤ x2 , y1 < Y ≤ y2 ] (8)
in terms of the joint CDF FX,Y (x, y), we write
P [C] = P [A ∪ B ∪ C] − P [A] − P [B] (9)

= FX,Y (x2 , y2 ) − FX,Y (x1 , y2 ) − FX,Y (x2 , y1 ) + FX,Y (x1 , y1 ) (10)
which completes the proof of the theorem.

The given function is
1 − e−(x+y) x, y ≥ 0
FX,Y (x, y) = (1)
0 otherwise
First, we find the CDF FX (x) and FY (y).

1 x≥0
FX (x) = FX,Y (x, ∞) = (2)
0 otherwise

1 y≥0
FY (y) = FX,Y (∞, y) = (3)
0 otherwise
Hence, for any x ≥ 0 or y ≥ 0,
P [X > x] = 0 P [Y > y] = 0 (4)
For x ≥ 0 and y ≥ 0, this implies
P [{X > x} ∪ {Y > y}] ≤ P [X > x] + P [Y > y] = 0 (5)
However,
P [{X > x} ∪ {Y > y}] = 1 − P [X ≤ x, Y ≤ y] = 1 − (1 − e−(x+y) ) = e−(x+y) (6)
Thus, we have the contradiction that e−(x+y) ≤ 0 for all x, y ≥ 0. We can conclude that the given
function is not a valid CDF.
126
In this problem, it is helpful to label points with nonzero probability on the X, Y plane:
y P
X,Y (x, y)
4 6
3 •3c •6c •12c
2
1 •c •2c •4c
0 - x
0 1 2 3 4
(a) We must choose c so the PMF sums to one:

PX,Y (x, y) = c x y (1)
x=1,2,4 y=1,3 x=1,2,4 y=1,3
= c [1(1 + 3) + 2(1 + 3) + 4(1 + 3)] = 28c (2)
Thus c = 1/28.
(b) The event {Y < X} has probability

1(0) + 2(1) + 4(1 + 3) 18
P [Y < X] = PX,Y (x, y) = = (3)
28 28
x=1,2,4 y<x
(c) The event {Y > X} has probability

1(3) + 2(3) + 4(0) 9
P [Y > X] = PX,Y (x, y) = = (4)
28 28
x=1,2,4 y>x
(d) There are two ways to solve this part. The direct way is to calculate
1(1) + 2(0) 1
P [Y = X] = PX,Y (x, y) = = (5)
28 28
x=1,2,4 y=x
The indirect way is to use the previous results and the observation that
P [Y = X] = 1 − P [Y < X] − P [Y > X] = (1 − 18/28 − 9/28) = 1/28 (6)
(e)
(1)(3) + (2)(3) + (4)(3) 21 3
P [Y = 3] = PX,Y (x, 3) = = = (7)
28 28 4
x=1,2,4

On the X, Y plane, the joint PMF is
127
y
PX,Y (x, y) 6
•c 1 •c •3c
•2c •2c - x
1 2
•3c •c •c
?
(a) To find c, we sum the PMF over all possible values of X and Y . We choose c so the sum
equals one.

PX,Y (x, y) = c |x + y| = 6c + 2c + 6c = 14c (1)
x y x=−2,0,2 y=−1,0,1
Thus c = 1/14.
(b)
P [Y < X] = PX,Y (0, −1) + PX,Y (2, −1) + PX,Y (2, 0) + PX,Y (2, 1) (2)
= c + c + 2c + 3c = 7c = 1/2 (3)
(c)
P [Y > X] = PX,Y (−2, −1) + PX,Y (−2, 0) + PX,Y (−2, 1) + PX,Y (0, 1) (4)
= 3c + 2c + c + c = 7c = 1/2 (5)
(d) From the sketch of PX,Y (x, y) given above, P [X = Y ] = 0.
(e)
P [X < 1] = PX,Y (−2, −1) + PX,Y (−2, 0) + PX,Y (−2, 1)

+ PX,Y (0, −1) + PX,Y (0, 1) (6)
= 8c = 8/14. (7)

Let r (reject) and a (accept) denote the result of each test. There are four possible outcomes:
rr, ra, ar, aa. The sample tree is
p2

p r •rr

XXX
p r
XX
1−p X a •ra
p(1−p)

HH
HH p r •ar p(1−p)
1−pHH a
XXX
XX
1−p X a •aa (1−p)2
128
Now we construct a table that maps the sample outcomes to values of X and Y .
outcome P [·] X Y
rr p 2 1 1
ra p(1 − p) 1 0 (1)
ar p(1 − p) 0 1
aa (1 − p)2 0 0
This table is esentially the joint PMF PX,Y (x, y).

⎧ 2
⎪
⎪ p x = 1, y = 1
⎪
⎪
⎨ p(1 − p) x = 0, y = 1
PX,Y (x, y) = p(1 − p) x = 1, y = 0 (2)
⎪
⎪
⎪
⎪ (1 − p)2 x = 0, y = 0
⎩
0 otherwise

The sample space is the set S = {hh, ht, th, tt} and each sample point has probability 1/4. Each
sample outcome specifies the values of X and Y as given in the following table
outcome X Y
hh 0 1
ht 1 0 (1)
th 1 1
tt 2 0
The joint PMF can represented by the table
PX,Y (x, y) y = 0 y = 1
x=0 0 1/4
(2)
x=1 1/4 1/4
x=2 1/4 0

As the problem statement says, reasonable arguments can be made for the labels being X and Y or
x and y. As we see in the arguments below, the lowercase choice of the text is somewhat arbitrary.
• Lowercase axis labels: For the lowercase labels, we observe that we are depicting the masses
associated with the joint PMF PX,Y (x, y) whose arguments are x and y. Since the PMF
function is defined in terms of x and y, the axis labels should be x and y.
• Uppercase axis labels: On the other hand, we are depicting the possible outcomes (labeled with
their respective probabilities) of the pair of random variables X and Y . The corresponding
axis labels should be X and Y just as in Figure 4.2. The fact that we have labeled the
possible outcomes by their probabilities is irrelevant. Further, since the expression for the
PMF PX,Y (x, y) given in the figure could just as well have been written PX,Y (·, ·), it is clear
that the lowercase x and y are not what matter.
129
As the problem statement indicates, Y = y < n if and only if
A: the first y tests are acceptable, and
B: test y + 1 is a rejection.
Thus P [Y = y] = P [AB]. Note that Y ≤ X since the number of acceptable tests before the first
failure cannot exceed the number of acceptable circuits. Moreover, given the occurrence of AB, the
event X = x < n occurs if and only if there are x − y acceptable circuits in the remaining n − y − 1
tests. Since events A, B and C depend on disjoint sets of tests, they are independent events. Thus,
for 0 ≤ y ≤ x < n,
PX,Y (x, y) = P [X = x, Y = y] = P [ABC] (1)

= P [A] P [B] P [C] (2)

n − y − 1 x−y
= p (1 − p)
y
p (1 − p)n−y−1−(x−y) (3)
x−y
P [A] P [B]
P [C]

n−y−1 x
= p (1 − p)n−x (4)
x−y
The case y = x = n occurs when all n tests are acceptable and thus PX,Y (n, n) = pn .

The joint PMF of X and K is PK,X (k, x) = P [K = k, X = x], which is the probability that K = k
and X = x. This means that both events must be satisfied. The approach we use is similar to that
used in finding the Pascal PMF in Example 2.15. Since X can take on only the two values 0 and
1, let’s consider each in turn. When X = 0 that means that a rejection occurred on the last test
and that the other k − 1 rejections must have occurred in the previous n − 1 tests. Thus,

n−1
PK,X (k, 0) = (1 − p)k−1 pn−1−(k−1) (1 − p) k = 1, . . . , n (1)
k−1
When X = 1 the last test was acceptable and therefore we know that the K = k ≤ n − 1 tails must
have occurred in the previous n − 1 tests. In this case,

n−1
PK,X (k, 1) = (1 − p)k pn−1−k p k = 0, . . . , n − 1 (2)
k
We can combine these cases into a single complete expression for the joint PMF.
⎧ n−1
⎨ k−1(1 − p)k pn−k x = 0, k = 1, 2, . . . , n
PK,X (k, x) = n−1
(1 − p)k pn−k x = 1, k = 0, 1, . . . , n − 1 (3)
⎩ k
0 otherwise

Each circuit test produces an acceptable circuit with probability p. Let K denote the number of
rejected circuits that occur in n tests and X is the number of acceptable circuits before the first re-
ject. The joint PMF, PK,X (k, x) = P [K = k, X = x] can be found by realizing that {K = k, X = x}
occurs if and only if the following events occur:
130
A The first x tests must be acceptable.
B Test x+1 must be a rejection since otherwise we would have x+1 acceptable at the beginnning.
C The remaining n − x − 1 tests must contain k − 1 rejections.
Since the events A, B and C are independent, the joint PMF for x + k ≤ r, x ≥ 0 and k ≥ 0 is

n−x−1
PK,X (k, x) = p (1 − p)
x
(1 − p)k−1 pn−x−1−(k−1) (1)
k−1
P [A] P [B]
P [C]
After simplifying, a complete expression for the joint PMF is

n−x−1 n−k
k−1 p (1 − p)k x + k ≤ n, x ≥ 0, k ≥ 0
PK,X (k, x) = (2)
0 otherwise

On the X, Y plane, the joint PMF PX,Y (x, y) is
y
6 PX,Y (x, y)
4
3 • •
3c 6c
•
12c
2
1 •c 2c
• •
4c
0 - x
0 1 2 3 4
By choosing c = 1/28, the PMF sums to one.
(a) The marginal PMFs of X and Y are

⎧
⎪
⎪ 4/28 x = 1
⎨
8/28 x = 2
PX (x) = PX,Y (x, y) = (1)
⎪
⎪ 16/28 x = 4
y=1,3 ⎩
0 otherwise
⎧
⎨ 7/28 y = 1
PY (y) = PX,Y (x, y) = 21/28 y = 3 (2)
⎩
x=1,2,4 0 otherwise
(b) The expected values of X and Y are

E [X] = xPX (x) = (4/28) + 2(8/28) + 4(16/28) = 3 (3)
x=1,2,4

E [Y ] = yPY (y) = 7/28 + 3(21/28) = 5/2 (4)
y=1,3
131
(c) The second moments are

E X2 = xPX (x) = 12 (4/28) + 22 (8/28) + 42 (16/28) = 73/7 (5)
x=1,2,4

2
E Y = yPY (y) = 12 (7/28) + 32 (21/28) = 7 (6)
y=1,3
The variances are

Var[X] = E X 2 − (E [X])2 = 10/7 Var[Y ] = E Y 2 − (E [Y ])2 = 3/4 (7)

The standard deviations are σX = 10/7 and σY = 3/4.

On the X, Y plane, the joint PMF is
y
PX,Y (x, y) 6
•c 1 •c •3c
•2c •2c - x
1 2
•3c •c •c
?
The PMF sums to one when c = 1/14.

(a) The marginal PMFs of X and Y are
⎧
⎨ 6/14 x = −2, 2
PX (x) = PX,Y (x, y) = 2/14 x=0 (1)
⎩
y=−1,0,1 0 otherwise
⎧
⎨ 5/14 y = −1, 1
PY (y) = PX,Y (x, y) = 4/14 y=0 (2)
⎩
x=−2,0,2 0 otherwise
(b) The expected values of X and Y are

E [X] = xPX (x) = −2(6/14) + 2(6/14) = 0 (3)
x=−2,0,2

E [Y ] = yPY (y) = −1(5/14) + 1(5/14) = 0 (4)
y=−1,0,1
(c) Since X and Y both have zero mean, the variances are

Var[X] = E X 2 = x2 PX (x) = (−2)2 (6/14) + 22 (6/14) = 24/7 (5)
x=−2,0,2

Var[Y ] = E Y 2 = y 2 PY (y) = (−1)2 (5/14) + 12 (5/14) = 5/7 (6)
y=−1,0,1

The standard deviations are σX = 24/7 and σY = 5/7.
132
We recognize that the given joint PMF is written as the product of two marginal PMFs PN (n) and
PK (k) where

100 100n e−100
n! n = 0, 1, . . .
PN (n) = PN,K (n, k) = (1)
0 otherwise
k=0
∞ 100
k pk (1 − p)100−k k = 0, 1, . . . , 100
PK (k) = PN,K (n, k) = (2)
0 otherwise
n=0

The joint PMF of N, K is
⎧
⎨ (1 − p)n−1 p/n k = 1, 2, . . . , n
PN,K (n, k) = n = 1, 2 . . . (1)
⎩
o otherwise
For n ≥ 1, the marginal PMF of N is

n
n
PN (n) = PN,K (n, k) = (1 − p)n−1 p/n = (1 − p)n−1 p (2)
k=1 k=1
The marginal PMF of K is found by summing PN,K (n, k) over all possible N . Note that if K = k,
then N ≥ k. Thus,
∞
1
PK (k) = (1 − p)n−1 p (3)
n
n=k
Unfortunately, this sum cannot be simplified.

For n = 0, 1, . . ., the marginal PMF of N is

n
100n e−100 100n e−100
PN (n) = PN,K (n, k) = = (1)
(n + 1)! n!
k k=0
For k = 0, 1, . . ., the marginal PMF of K is

∞
∞
100n e−100 1 100n+1 e−100
PK (k) = = (2)
(n + 1)! 100 (n + 1)!
n=k n=k
∞
1
= PN (n + 1) (3)
100
n=k
= P [N > k] /100 (4)
133
(a) The joint PDF of X and Y is
Y
1
Y+X=1

c x + y ≤ 1, x, y ≥ 0
fX,Y (x, y) = (1)
0 otherwise
X
1
To find the constant c we integrate over the region shown. This gives
' 1 ' 1−x
cx $$1 c
c dy dx = cx − $ = = 1 (2)
0 0 2 0 2
Therefore c = 2.
(b) To find the P [X ≤ Y ] we look to integrate over the area indicated by the graph
Y
1 X£Y ' 1/2 ' 1−x
X=Y P [X ≤ Y ] = dy dx (3)
0 x
' 1/2
= (2 − 4x) dx (4)
0
X
= 1/2 (5)
1
(c) The probability P [X + Y ≤ 1/2] can be seen in the figure. Here we can set up the following
integrals
Y
Y+X=1 '
1 1/2 ' 1/2−x
P [X + Y ≤ 1/2] = 2 dy dx (6)
Y+X=½ 0 0
' 1/2
= (1 − 2x) dx (7)
X 0
= 1/2 − 1/4 = 1/4 (8)
1

Given the joint PDF
cxy 2 0 ≤ x, y ≤ 1
fX,Y (x, y) = (1)
0 otherwise
(a) To find the constant c integrate fX,Y (x, y) over the all possible values of X and Y to get
' 1' 1
1= cxy 2 dx dy = c/6 (2)
0 0
Therefore c = 6.
134
(b) The probability P [X ≥ Y ] is the integral of the joint PDF fX,Y (x, y) over the indicated shaded
region.
Y
1 ' 1' x
P [X ≥ Y ] = 6xy 2 dy dx (3)
0 0
' 1
= 2x4 dx (4)
0
X = 2/5 (5)
1
Y
1 Similarly, to find P [Y ≤ X 2 ] we can integrate over the region
shown in the figure.
Y=X 2
' 1 ' x2

P Y ≤ X2 = 6xy 2 dy dx (6)
0 0
X
= 1/4 (7)
1
(c) Here we can choose to either integrate fX,Y (x, y) over the lighter shaded region, which would
require the evaluation of two integrals, or we can perform one integral over the darker region
by recognizing
min(X,Y) < ½
Y
min(X,Y) > ½
1 P [min(X, Y ) ≤ 1/2] = 1 − P [min(X, Y ) > 1/2] (8)
' 1 ' 1
=1− 6xy 2 dx dy (9)
1/2 1/2
' 1
9y 2 11
=1− dy = (10)
X 1/2 4 32
1
(d) The probability P [max(X, Y ) ≤ 3/4] can be found be integrating over the shaded region
shown below.
Y
1 max(X,Y) < ¾
P [max(X, Y ) ≤ 3/4] = P [X ≤ 3/4, Y ≤ 3/4] (11)
' 3' 3
4 4
= 6xy 2 dx dy (12)
0 $ 0 $
= x2 $0 y 3 $0
3/4 3/4
(13)
X
= (3/4)5 = 0.237 (14)
1

The joint PDF of X and Y is

6e−(2x+3y) x ≥ 0, y ≥ 0,
fX,Y (x, y) = (1)
0 otherwise.
135
(a) The probability that X ≥ Y is:
Y ' ∞' x
X³Y P [X ≥ Y ] = 6e−(2x+3y) dy dx (2)
'
0 0

∞ $y=x
= 2e−2x −e−3y $y=0 dx (3)
'
0
∞
X = [2e−2x − 2e−5x ] dx = 3/5 (4)
0
The P [X + Y ≤ 1] is found by integrating over the region where X + Y ≤ 1
' 1 ' 1−x
Y P [X + Y ≤ 1] = 6e−(2x+3y) dy dx (5)
1 0 0
'
X+Y≤ 1 1 $y=1−x
= 2e−2x −e−3y $y=0 dx (6)
0
' 1
= 2e−2x 1 − e−3(1−x) dx (7)
X 0
$1
1 = −e−2x − 2ex−3 $0 (8)
= 1 + 2e−3 − 3e−2 (9)
(b) The event min(X, Y ) ≥ 1 is the same as the event {X ≥ 1, Y ≥ 1}. Thus,
' ∞' ∞
P [min(X, Y ) ≥ 1] = 6e−(2x+3y) dy dx = e−(2+3) (10)
1 1
(c) The event max(X, Y ) ≤ 1 is the same as the event {X ≤ 1, Y ≤ 1} so that

' 1' 1
P [max(X, Y ) ≤ 1] = 6e−(2x+3y) dy dx = (1 − e−2 )(1 − e−3 ) (11)
0 0

The only difference between this problem and Example 4.5 is that in this problem we must integrate
the joint PDF over the regions to find the probabilities. Just as in Example 4.5, there are five cases.
We will use variable u and v as dummy variables for x and y.
• x < 0 or y < 0
Y
1
In this case, the region of integration doesn’t overlap the
region of nonzero probability and
y ' y ' x
FX,Y (x, y) = fX,Y (u, v) du dv = 0 (1)
X −∞ −∞
x 1
136
• 0<y≤x≤1
In this case, the region where the integral has a nonzero contribution is
Y
' y ' x
1
FX,Y (x, y) = fX,Y (u, v) dy dx (2)
'−∞ −∞
y' x
y
= 8uv du dv (3)
'0 y v
X
= 4(x2 − v 2 )v dv (4)
x 1 0
$v=y
= 2x2 v 2 − v 4 $v=0 = 2x2 y 2 − y 4 (5)
• 0 < x ≤ y and 0 ≤ x ≤ 1
Y
1 ' y ' x
FX,Y (x, y) = fX,Y (u, v) dv du (6)
y '−∞ −∞
x' u
= 8uv dv du (7)
'0 x 0
X
1 x
= 4u3 du = x4 (8)
0
• 0 < y ≤ 1 and x ≥ 1
Y
' y ' x
1
−∞ −∞
y ' y' 1
= 8uv du dv (10)
' 0
y
v
X = 4v(1 − v 2 ) dv (11)
x 1 0
= 2y 2 − y 4 (12)
• x ≥ 1 and y ≥ 1
137
Y
y
1
In this case, the region of integration completely covers the
region of nonzero probability and
' y ' x
FX,Y (x, y) = fX,Y (u, v) du dv (13)
−∞ −∞
X =1 (14)
1 x
The complete answer for the joint CDF is

⎧
⎪
⎪ 0 x < 0 or y < 0
⎪
⎪
⎨ 2x2 y 2 − y 4 0<y≤x≤1
FX,Y (x, y) = x4 0 ≤ x ≤ y, 0 ≤ x ≤ 1 (15)
⎪
⎪
⎪
⎪ 2y 2 − y 4 0 ≤ y ≤ 1, x ≥ 1
⎩
1 x ≥ 1, y ≥ 1
(a) The joint PDF (and the corresponding region of nonzero probability) are
Y

1/2 −1 ≤ x ≤ y ≤ 1
X fX,Y (x, y) = (1)
0 otherwise
1
-1
(b)
' 1' 1 '
1 1
1−x
P [X > 0] = dy dx = dx = 1/4 (2)
0 x 2 0 2
This result can be deduced by geometry. The shaded triangle of the X, Y plane corresponding
to the event X > 0 is 1/4 of the total shaded area.
(c) For x > 1 or x < −1, fX (x) = 0. For −1 ≤ x ≤ 1,

' ∞ ' 1
1
fX (x) = fX,Y (x, y) dy = dy = (1 − x)/2. (3)
−∞ x 2
The complete expression for the marginal PDF is

(1 − x)/2 −1 ≤ x ≤ 1,
fX (x) = (4)
0 otherwise.
138
(d) From the marginal PDF fX (x), the expected value of X is
' ∞ '
1 1
E [X] = xfX (x) dx = x(1 − x) dx (5)
−∞ 2 −1
$1
x2 x3 $$ 1
= − $ =− . (6)
4 6 −1 3

2 x + y ≤ 1, x, y ≥ 0
fX,Y (x, y) = (1)
0 otherwise
Using the figure to the left we can find the marginal PDFs by integrating
Y over the appropriate regions.
1
Y+X=1 ' 1−x
2(1 − x) 0 ≤ x ≤ 1
fX (x) = 2 dy = (2)
0 0 otherwise
Likewise for fY (y):

X
'
1
1−y
2(1 − y) 0 ≤ y ≤ 1
fY (y) = 2 dx = (3)
0 0 otherwise

Random variables X and Y have joint PDF

1/(πr2 ) 0 ≤ x2 + y 2 ≤ r2
fX,Y (x, y) = (1)
0 otherwise
(a) The marginal PDF of X is

' √ √
r 2 −x2 2 r 2 −x2
1
πr 2
−r ≤ x ≤ r,
fX (x) = 2 √ dy = (2)
− r 2 −x2 πr2 0 otherwise.
(b) Similarly, for random variable Y ,

' √ r 2 −y 2
√
2 r 2 −y 2
1 −r ≤ y ≤ r,
fY (y) = 2 √ πr2
dx = πr 2 (3)
− r 2 −y 2 0 otherwise.

The joint PDF of X and Y and the region of nonzero probability are
Y
1
5x2 /2 −1 ≤ x ≤ 1, 0 ≤ y ≤ x2
fX,Y (x, y) = (1)
0 otherwise
X
-1 1
We can find the appropriate marginal PDFs by integrating the joint PDF.
139
' x2
5x2 5x4 /2 −1 ≤ x ≤ 1
fX (x) = dy = (2)
0 2 0 otherwise
(b) Note that fY (y) = 0 for y > 1 or y < 0. For 0 ≤ y ≤ 1,

Y ' ∞
1 fY (y) = fX,Y (x, y) dx (3)
−∞
' −√y ' 1
y 5x2 5x2
= dx + √
dx (4)
X −1 2 y 2
-1 -Öy Öy 1 = 5(1 − y 3/2 )/3 (5)
The complete expression for the marginal CDF of Y is

5(1 − y 3/2 )/3 0 ≤ y ≤ 1
fY (y) = (6)
0 otherwise

In this problem, the joint PDF is

2 |xy| /r4 0 ≤ x2 + y 2 ≤ r2
fX,Y (x, y) = (1)
0 otherwise
(a) Since |xy| = |x||y|, for −r ≤ x ≤ r, we can write

' ' √
∞ r 2 −x2
2 |x|
fX (x) = fX,Y (x, y) dy = 4 √ |y| dy (2)
−∞ r − r 2 −x2
Since |y| is symmetric about the origin, we can simplify the integral to
' √ $√ 2 2
4 |x| r 2 −x2
2 |x| 2 $$ r −x 2 |x| (r2 − x2 )
fX (x) = 4 y dy = 4 y $ = (3)
r 0 r 0 r4
Note that for |x| > r, fX (x) = 0. Hence the complete expression for the PDF of X is

2|x|(r 2 −x2 )
r4
−r ≤ x ≤ r
fX (x) = (4)
0 otherwise
(b) Note that the joint PDF is symmetric in x and y so that fY (y) = fX (y).
140
(a) The joint PDF of X and Y and the region of nonzero probability are
Y
1

cy 0 ≤ y ≤ x ≤ 1
fX,Y (x, y) = (1)
0 otherwise
X
1
(b) To find the value of the constant, c, we integrate the joint PDF over all x and y.
' ' ' 1' x ' $1
∞ ∞ 1
cx2 cx3 $$ c
fX,Y (x, y) dx dy = cy dy dx = dx = $ = . (2)
−∞ −∞ 0 0 0 2 6 0 6
Thus c = 6.
(c) We can find the CDF FX (x) = P [X ≤ x] by integrating the joint PDF over the event X ≤ x.
For x < 0, FX (x) = 0. For x > 1, FX (x) = 1. For 0 ≤ x ≤ 1,
Y ''

1 FX (x) = fX,Y x , y dy dx (3)
x ≤x
' x ' x
= 6y dy dx (4)
' 0
x
0
X
x 1 = 3(x )2 dx = x3 . (5)
0
The complete expression for the joint CDF is
⎧
⎨ 0 x<0
FX (x) = x3 0 ≤ x ≤ 1 (6)
⎩
1 x≥1
(d) Similarly, we find the CDF of Y by integrating fX,Y (x, y) over the event Y ≤ y. For y < 0,
FY (y) = 0 and for y > 1, FY (y) = 1. For 0 ≤ y ≤ 1,
''
Y
FY (y) = fX,Y x , y dy dx (7)
1 y ≤y
' y' 1
y
= 6y dx dy (8)
0 y
' y
X 6y (1 − y ) dy
= (9)
1 $y0
= 3(y )2 − 2(y )3 $0 = 3y 2 − 2y 3 . (10)
The complete expression for the CDF of Y is
⎧
⎨ 0 y<0
FY (y) = 3y 2 − 2y 3 0 ≤ y ≤ 1 (11)
⎩
1 y>1
141
(e) To find P [Y ≤ X/2], we integrate the joint PDF fX,Y (x, y) over the region y ≤ x/2.
Y ' 1 ' x/2
1 P [Y ≤ X/2] = 6y dy dx (12)
0 0
' 1 $x/2
½ = 3y 2 $0 dx (13)
0
' 1
X 3x2
1 = dx = 1/4 (14)
0 4

In this problem, it is helpful to label possible points X, Y along with the corresponding values of
W = X − Y . From the statement of Problem 4.6.1,
y
4 6 PX,Y (x, y)
W =−2 W =−1 W =1
3/28 6/28 12/28
3 • • •
W =0 W =1 W =3
1/28 2/28 4/28
1 • • •
0 - x
0 1 2 3 4
(a) To find the PMF of W , we simply add the probabilities associated with each possible value
of W :
PW (−2) = PX,Y (1, 3) = 3/28 PW (−1) = PX,Y (2, 3) = 6/28 (1)

PW (0) = PX,Y (1, 1) = 1/28 PW (1) = PX,Y (2, 1) + PX,Y (4, 3) (2)
PW (3) = PX,Y (4, 1) = 4/28 = 14/28 (3)
For all other values of w, PW (w) = 0.
(b) The expected value of W is

E [W ] = wPW (w) (4)
w
= −2(3/28) + −1(6/28) + 0(1/28) + 1(14/28) + 3(4/28) = 1/2 (5)
(c) P [W > 0] = PW (1) + PW (3) = 18/28.
142
y In Problem 4.2.2, the joint PMF
6 PX,Y (x, y) PX,Y (x, y) is given in terms of the
c c 3c
W =0
• 1 • W =2
• W =4 parameter c. For this problem, we first
2c 2c need to find c. Before doing so, it is
W
=−2 -
• • W =2 x convenient to label each possible X, Y
1 2
point with the corresponding value of
• • c • c
3c
W =−4 W =−2 W =0 W = X + 2Y .
?
To find c, we sum the PMF over all possible values of X and Y . We choose c so the sum equals
one.

PX,Y (x, y) = c |x + y| (1)
x y x=−2,0,2 y=−1,0,1
= 6c + 2c + 6c = 14c (2)
Thus c = 1/14. Now we can solve the actual problem.
(a) From the above graph, we can calculate the probability of each possible value of w.
PW (−4) = PX,Y (−2, −1) = 3c (3)

PW (−2) = PX,Y (−2, 0) + PX,Y (0, −1) = 3c (4)
PW (0) = PX,Y (−2, 1) + PX,Y (2, −1) = 2c (5)
PW (2) = PX,Y (0, 1) + PX,Y (2, 0) = 3c (6)
PW (4) = PX,Y (2, 1) = 3c (7)
With c = 1/14, we can summarize the PMF as

⎧
⎨ 3/14 w = −4, −2, 2, 4
PW (w) = 2/14 w = 0 (8)
⎩
0 otherwise
(b) The expected value is now straightforward:

3 2
E [W ] = (−4 + −2 + 2 + 4) + 0 = 0. (9)
14 14
(c) Lastly, P [W > 0] = PW (2) + PW (4) = 3/7.

We observe that when X = x, we must have Y = w − x in order for W = w. That is,
∞
∞

PW (w) = P [X = x, Y = w − x] = PX,Y (x, w − x) (1)
x=−∞ x=−∞
143
Y
W>w The x, y pairs with nonzero probability are shown in the figure.
For w = 0, 1, . . . , 10, we observe that
w
P [W > w] = P [min(X, Y ) > w] (1)
= P [X > w, Y > w] (2)
= 0.01(10 − w) 2
(3)
X
w
To find the PMF of W , we observe that for w = 1, . . . , 10,
PW (w) = P [W > w − 1] − P [W > w] (4)

= 0.01[(10 − w − 1)2 − (10 − w)2 ] = 0.01(21 − 2w) (5)
The complete expression for the PMF of W is

0.01(21 − 2w) w = 1, 2, . . . , 10
PW (w) = (6)
0 otherwise

Y
The x, y pairs with nonzero probability are shown in the figure.

For v = 1, . . . , 11, we observe that
v V<v
P [V < v] = P [max(X, Y ) < v] (1)
= P [X < v, Y < v] (2)
= 0.01(v − 1) 2
(3)
X
v
To find the PMF of V , we observe that for v = 1, . . . , 10,
PV (v) = P [V < v + 1] − P [V < v] (4)

= 0.01[v − (v − 1) ]
2 2
(5)
= 0.01(2v − 1) (6)
The complete expression for the PMF of V is

0.01(2v − 1) v = 1, 2, . . . , 10
PV (v) = (7)
0 otherwise
144
(a) The minimum value of W is W = 0, which occurs when X = 0 and Y = 0. The maximum
value of W is W = 1, which occurs when X = 1 or Y = 1. The range of W is SW =
{w|0 ≤ w ≤ 1}.
(b) For 0 ≤ w ≤ 1, the CDF of W is

Y
1
FW (w) = P [max(X, Y ) ≤ w] (1)
w W<w
= P [X ≤ w, Y ≤ w] (2)
' w' w
X = fX,Y (x, y) dy dx (3)
w 0 0
1
Substituting fX,Y (x, y) = x + y yields
' w' w
FW (w) = (x + y) dy dx (4)
' w! $y=w "
0 0
' w
y 2 $$
= xy + $ dx = (wx + w2 /2) dx = w3 (5)
0 2 y=0 0
The complete expression for the CDF is

⎧
⎨ 0 w<0
FW (w) = w3 0 ≤ w ≤ 1 (6)
⎩
1 otherwise
The PDF of W is found by differentiating the CDF.

dFW (w) 3w2 0 ≤ w ≤ 1
fW (w) = = (7)
dw 0 otherwise
(a) Since the joint PDF fX,Y (x, y) is nonzero only for 0 ≤ y ≤ x ≤ 1, we observe that W =
Y − X ≤ 0 since Y ≤ X. In addition, the most negative value of W occurs when Y = 0 and
X = 1 and W = −1. Hence the range of W is SW = {w| − 1 ≤ w ≤ 0}.
(b) For w < −1, FW (w) = 0. For w > 0, FW (w) = 1. For −1 ≤ w ≤ 0, the CDF of W is
Y
FW (w) = P [Y − X ≤ w] (1)
1 ' 1 ' x+w
Y=X+w = 6y dy dx (2)
½ −w 0
' 1
= 3(x + w)2 dx (3)
X −w
-w 1 $1
= (x + w)3 $−w = (1 + w)3 (4)
145
Therefore, the complete CDF of W is
⎧
⎨ 0 w < −1
FW (w) = (1 + w) −1 ≤ w ≤ 0
3 (5)
⎩
1 w>0
By taking the derivative of fW (w) with respect to w, we obtain the PDF

3(w + 1)2 −1 ≤ w ≤ 0
fW (w) = (6)
0 otherwise

Y
1

2 0≤y≤x≤1
fX,Y (x, y) = (1)
0 otherwise
X
1
(a) Since X and Y are both nonnegative, W = Y /X ≥ 0. Since Y ≤ X, W = Y /X ≤ 1. Note

that W = 0 can occur if Y = 0. Thus the range of W is SW = {w|0 ≤ w ≤ 1}.
(b) For 0 ≤ w ≤ 1, the CDF of W is

Y
1 FW (w) = P [Y /X ≤ w] = P [Y ≤ wX] = w (2)
w
The complete expression for the CDF is
P[Y<wX] ⎧
⎨ 0 w<0
X
FW (w) = w 0≤w<1 (3)
1 ⎩
1 w≥1
By taking the derivative of the CDF, we find that the PDF of W is

1 0≤w<1
fW (w) = (4)
0 otherwise
We see that W has a uniform PDF over [0, 1]. Thus E[W ] = 1/2.

Y
1

2 0≤y≤x≤1
fX,Y (x, y) = (1)
0 otherwise
X
1
146
(a) Since fX,Y (x, y) = 0 for y > x, we can conclude that Y ≤ X and that W = X/Y ≥ 1. Since
Y can be arbitrarily small but positive, W can be arbitrarily large. Hence the range of W is
SW = {w|w ≥ 1}.
(b) For w ≥ 1, the CDF of W is
Y
1 P[Y<X/w]
FW (w) = P [X/Y ≤ w] (2)
1/w = 1 − P [X/Y > w] (3)
= 1 − P [Y < X/w] (4)
X
1
= 1 − 1/w (5)
Note that we have used the fact that P [Y < X/w] equals 1/2 times the area of the corre-
sponding triangle. The complete CDF is

0 w<1
FW (w) = (6)
1 − 1/w w ≥ 1
The PDF of W is found by differentiating the CDF.

dFW (w) 1/w2 w ≥ 1
fW (w) = = (7)
dw 0 otherwise
To find the expected value E[W ], we write
' ∞ ' ∞
dw
E [W ] = wfW (w) dw = . (8)
−∞ 1 w
However, the integral diverges and E[W ] is undefined.

The position of the mobile phone is equally likely to be anywhere in the area of a circle with radius
16 km. Let X and Y denote the position of the mobile. Since we are given that the cell has a
radius of 4 km, we will measure X and Y in kilometers. Assuming the base station is at the origin
of the X, Y plane, the joint PDF of X and Y is
1
16π x + y ≤ 16
2 2
fX,Y (x, y) = (1)
0 otherwise
√
Since the mobile’s radial distance from the base station is R = X 2 + Y 2 , the CDF of R is

FR (r) = P [R ≤ r] = P X 2 + Y 2 ≤ r2 (2)
By changing to polar coordinates, we see that for 0 ≤ r ≤ 4 km,
' 2π ' r
r
FR (r) = dr dθ = r2 /16 (3)
0 0 16π
So ⎧
⎨ 0 r<0
FR (r) = r2 /16 0 ≤ r < 4 (4)
⎩
1 r≥4
Then by taking the derivative with respect to r we arrive at the PDF

r/8 0 ≤ r ≤ 4
fR (r) = (5)
0 otherwise
147
Following the hint, we observe that either Y ≥ X or X ≥ Y , or, equivalently, (Y /X) ≥ 1 or
(X/Y ) ≥ 1. Hence, W ≥ 1. To find the CDF FW (w), we know that FW (w) = 0 for w < 1. For
w ≥ 1, we solve
FW (w) = P [max[(X/Y ), (Y /X)] ≤ w]

= P [(X/Y ) ≤ w, (Y /X) ≤ w]
= P [Y ≥ X/w, Y ≤ wX]
= P [X/w ≤ Y ≤ wX]
Y=wX
Y We note that in the middle of the above steps, nonnegativity of X and Y
a
a/w was essential. We can depict the given set {X/w ≤ Y ≤ wX} as the dark
region on the X, Y plane. Because the PDF is uniform over the square, it
Y=X/w
is easier to use geometry to calculate the probability. In particular, each
a/w of the lighter triangles that are not part of the region of interest has area
X a2 /2w.
a
This implies
a2 /2w + a2 /2w 1
P [X/w ≤ Y ≤ wX] = 1 − 2
=1− (1)
a w
The final expression for the CDF of W is

0 w<1
FW (w) = (2)
1 − 1/w w ≥ 1
By taking the derivative, we obtain the PDF

0 w<1
fW (w) = (3)
1/w2 w ≥ 1

y
In Problem 4.2.1, we found the joint PMF PX,Y (x, y) as
4 6 PX,Y (x, y)
shown. Also the expected values and variances were
3/28
3 • •6/28 •12/28
E [X] = 3 Var[X] = 10/7 (1)
2
1/28 E [Y ] = 5/2 Var[Y ] = 3/4 (2)
1 • •2/28 •4/28
0 - x We use these results now to solve this problem.
0 1 2 3 4
(a) Random variable W = Y /X has expected value

y
E [Y /X] = PX,Y (x, y) (3)
x
x=1,2,4 y=1,3
1 1 3 3 1 2 3 6 1 4 3 12
= + + + + + = 15/14 (4)
1 28 1 28 2 28 2 28 4 28 4 28
148
(b) The correlation of X and Y is

rX,Y = xyPX,Y (x, y) (5)
x=1,2,4 y=1,3
1 · 1 · 1 1 · 3 · 3 2 · 1 · 2 2 · 3 · 6 4 · 1 · 4 4 · 3 · 12
= + + + + + (6)
28 28 28 28 28 28
= 210/28 = 15/2 (7)
Recognizing that PX,Y (x, y) = xy/28 yields the faster calculation

(xy)2
rX,Y = E [XY ] = (8)
28
x=1,2,4 y=1,3
1 2 2
= x y (9)
28
x=1,2,4 y=1,3
1 210 15
= (1 + 22 + 42 )(12 + 32 ) = = (10)
28 28 2
(c) The covariance of X and Y is

15 5
Cov [X, Y ] = E [XY ] − E [X] E [Y ] = −3 =0 (11)
2 2
(d) Since X and Y have zero covariance, the correlation coefficient is
Cov [X, Y ]
ρX,Y = = 0. (12)
Var[X] Var[Y ]
(e) Since X and Y are uncorrelated, the variance of X + Y is

61
Var[X + Y ] = Var[X] + Var[Y ] = . (13)
28

y In Problem 4.2.1, we found the joint PMF PX,Y (x, y)
PX,Y (x, y) 6 shown here. The expected values and variances were
•1/14 1 •1/14 •3/14 found to be
•2/14 •2/14- x E [X] = 0 Var[X] = 24/7 (1)

1 2
E [Y ] = 0 Var[Y ] = 5/7 (2)
•3/14 •1/14 •1/14
We need these results to solve this problem.
?
149
(a) Random variable W = 2XY has expected value

E 2XY = 2xy PX,Y (x, y) (3)
x=−2,0,2 y=−1,0,1
3 2 1 1 1
= 2−2(−1) + 2−2(0) + 2−2(1) + 20(−1) + 20(1) (4)
14 14 14 14 14
2(−1) 1 2(0) 2 2(1) 3
+2 +2 +2 (5)
14 14 14
= 61/28 (6)
(b) The correlation of X and Y is

rX,Y = xyPX,Y (x, y) (7)
x=−2,0,2 y=−1,0,1
−2(−1)(3) −2(0)(2) −2(1)(1) 2(−1)(1) 2(0)(2) 2(1)(3)
= + + + + + (8)
14 14 14 14 14 14
= 4/7 (9)
(c) The covariance of X and Y is

Cov [X, Y ] = E [XY ] − E [X] E [Y ] = 4/7 (10)
(d) The correlation coefficient is

Cov [X, Y ] 2
ρX,Y = =√ (11)
Var[X] Var[Y ] 30
(e) By Theorem 4.16,

Var [X + Y ] = Var [X] + Var[Y ] + 2 Cov [X, Y ] (12)
24 5 4 37
= + +2 = . (13)
7 7 7 7

In the solution to Quiz 4.3, the joint PMF and the marginal PMFs are
PH,B (h, b) b = 0 b = 2 b = 4 PH (h)
h = −1 0 0.4 0.2 0.6
h=0 0.1 0 0.1 0.2 (1)
h=1 0.1 0.1 0 0.2
PB (b) 0.2 0.5 0.3
From the joint PMF, the correlation coefficient is

1
rH,B = E [HB] = hbPH,B (h, b) (2)
h=−1 b=0,2,4
= −1(2)(0.4) + 1(2)(0.1) + −1(4)(0.2) + 1(4)(0) (3)

= −1.4 (4)
150
since only terms in which both h and b are nonzero make a contribution. Using the marginal
PMFs, the expected values of X and Y are

1
E [H] = hPH (h) = −1(0.6) + 0(0.2) + 1(0.2) = −0.2 (5)
h=−1

E [B] = bPB (b) = 0(0.2) + 2(0.5) + 4(0.3) = 2.2 (6)
b=0,2,4
The covariance is
Cov [H, B] = E [HB] − E [H] E [B] = −1.4 − (−0.2)(2.2) = −0.96 (7)

From the joint PMF, PX (x)Y , found in Example 4.13, we can find the marginal PMF for X or Y
by summing over the columns or rows of the joint PMF.
⎧
⎪
⎪ 25/48 y = 1
⎪
⎪
⎨ 13/48 y = 2
1/4 x = 1, 2, 3, 4
PY (y) = 7/48 y = 3 PX (x) = (1)
⎪
⎪ 0 otherwise
⎪
⎪ 3/48 y = 4
⎩
0 otherwise
(a) The expected values are

4
25 13 7 3
E [Y ] = yPY (y) = 1 + 2 + 3 + 4 = 7/4 (2)
48 48 48 48
y=1

4
1
E [X] = xPX (x) = (1 + 2 + 3 + 4) = 5/2 (3)
4
x=1
(b) To find the variances, we first find the second moments.

4
25 13 7 3
2
E Y = y 2 PY (y) = 12 + 22 + 32 + 42 = 47/12 (4)
48 48 48 48
y=1
2
4
1 2
E X = x2 PX (x) = 1 + 22 + 32 + 42 = 15/2 (5)
4
x=1
Now the variances are

Var[Y ] = E Y 2 − (E [Y ])2 = 47/12 − (7/4)2 = 41/48 (6)

Var[X] = E X 2 − (E [X])2 = 15/2 − (5/2)2 = 5/4 (7)
151
(c) To find the correlation, we evaluate the product XY over all values of X and Y . Specifically,

4
x
rX,Y = E [XY ] = xyPX,Y (x, y) (8)
x=1 y=1
1 2 3 4 4 6 8 9 12 16
= + + + + + + + + + (9)
4 8 12 16 8 12 16 12 16 16
=5 (10)
(d) The covariance of X and Y is
Cov [X, Y ] = E [XY ] − E [X] E [Y ] = 5 − (7/4)(10/4) = 10/16 (11)
(e) The correlation coefficient is

Cov [W, V ] 10/16
ρX,Y = = ≈ 0.605 (12)
Var[W ] Var[V ] (41/48)(5/4)

For integers 0 ≤ x ≤ 5, the marginal PMF of X is

x
x+1
PX (x) = PX,Y (x, y) = (1/21) = (1)
y
21
y=0
Similarly, for integers 0 ≤ y ≤ 5, the marginal PMF of Y is

5
6−y
PY (y) = PX,Y (x, y) = (1/21) = (2)
x x=y
21
The complete expressions for the marginal PMFs are

(x + 1)/21 x = 0, . . . , 5
PX (x) = (3)
0 otherwise

(6 − y)/21 y = 0, . . . , 5
PY (y) = (4)
0 otherwise
The expected values are

5
x+1 70 10
5
6−y 35 5
E [X] = x = = E [Y ] = y = = (5)
21 21 3 21 21 3
x=0 y=0
To find the covariance, we first find the correlation

5
x
xy 1
5 x
1 2
5
280 20
E [XY ] = = x y= x (x + 1) = = (6)
21 21 42 42 3
x=0 y=0 x=1 y=1 x=1
The covariance of X and Y is

20 50 10
Cov [X, Y ] = E [XY ] − E [X] E [Y ] = − = (7)
3 9 9
152
y To solve this problem, we identify the

1/16 values of W = min(X, Y ) and V =
4 6 PX,Y (x, y) •
W =4
max(X, Y ) for each possible pair x, y.
V =4 Here we observe that W = Y and
1/12 1/16 V = X. This is a result of the under-
3 • •
W =3 W =3 lying experiment in that given X =
V =3 V =4
x, each Y ∈ {1, 2, . . . , x} is equally
1/8 1/12 1/16
2 • • • likely. Hence Y ≤ X. This implies
W =2 W =2 W =2
V =2 V =3 V =4 min(X, Y ) = Y and max(X, Y ) = X.
1/4 1/8 1/12 1/16
1 • • • •
W =1 W =1 W =1 W =1
V =1 V =2 V =3 V =4
0 - x
0 1 2 3 4
Using the results from Problem 4.7.4, we have the following answers.
(a) The expected values are
E [W ] = E [Y ] = 7/4 E [V ] = E [X] = 5/2 (1)
(b) The variances are

Var[W ] = Var[Y ] = 41/48 Var[V ] = Var[X] = 5/4 (2)
(c) The correlation is

rW,V = E [W V ] = E [XY ] = rX,Y = 5 (3)
(d) The covariance of W and V is
Cov [W, V ] = Cov [X, Y ] = 10/16 (4)
(e) The correlation coefficient is

10/16
ρW,V = ρX,Y = ≈ 0.605 (5)
(41/48)(5/4)

First, we observe that Y has mean µY = aµX + b and variance Var[Y ] = a2 Var[X]. The covariance
of X and Y is
Cov [X, Y ] = E [(X − µX )(aX + b − aµX − b)] (1)

= aE (X − µX )2 (2)
= a Var[X] (3)
The correlation coefficient is
Cov [X, Y ] a Var[X] a
ρX,Y = = = (4)
Var[X] Var[Y ] 2
Var[X] a Var[X] |a|
When a > 0, ρX,Y = 1. When a < 0, ρX,Y = 1.
153

(x + y)/3 0 ≤ x ≤ 1, 0 ≤ y ≤ 2
fX,Y (x, y) = (1)
0 otherwise
Before calculating moments, we first find the marginal PDFs of X and Y . For 0 ≤ x ≤ 1,
' ∞ ' 2 $y=2
x+y xy y 2 $$ 2x + 2
fX (x) = fX,Y (x, y) dy = dy = + $ = (2)
−∞ 0 3 3 6 y=0 3
For 0 ≤ y ≤ 2,
' ' 1
$x=1
∞
x y x2 xy $$ 2y + 1
fY (y) = fX,Y (x, y) dx = + dx = + $ = (3)
−∞ 0 3 3 6 3 x=0 6
Complete expressions for the marginal PDFs are
2x+2
3 0≤x≤1 2y+1
6 0≤y≤2
fX (x) = fY (y) = (4)
' ∞ ' $1
1
2x + 2 2x3 x2 $$ 5
E [X] = xfX (x) dx = x dx = + $ = (5)
−∞ 0 3 9 3 0 9
The second moment of X is
' ∞ ' $1
2 2
1
2 2x +2 x4 2x3 $$ 7
E X = x fX (x) dx = x dx = + = (6)
−∞ 0 3 6 9 $0 18
The variance of X is Var[X] = E[X 2 ] − (E[X])2 = 7/18 − (5/9)2 = 13/162.
(b) The expected value of Y is
' ∞ ' $2
2
2y + 1 y2 y 3 $$ 11
E [Y ] = yfY (y) dy = y dy = + $ = (7)
−∞ 0 6 12 9 0 9
The second moment of Y is
' ∞ ' $2
2 2
2
2 2y +1 y3 y 4 $$ 16
E Y = y fY (y) dy = y dy = + $ = (8)
−∞ 0 6 18 12 0 9
The variance of Y is Var[Y ] = E[Y 2 ] − (E[Y ])2 = 23/81.
(c) The correlation of X and Y is
''
E [XY ] = xyfX,Y (x, y) dx dy (9)
' 1' 2
x+y
= xy dy dx (10)
3
0
' 1 !0
$y=2 "
x2 y 2 xy 3 $$
= + dx (11)
0 6 9 $y=0
' 1 2 $1
2x 8x 2x3 4x2 $$ 2
= + dx = + $ = (12)
0 3 9 9 9 0 3
The covariance is Cov[X, Y ] = E[XY ] − E[X]E[Y ] = −1/81.
154
(d) The expected value of X and Y is
E [X + Y ] = E [X] + E [Y ] = 5/9 + 11/9 = 16/9 (13)

13 23 2 55
Var[X + Y ] = Var[X] + Var[Y ] + 2 Cov [X, Y ] = + − = (14)
162 81 81 162
(a) The first moment of X is

' 1' 1 ' 1
2 2
E [X] = 4x y dy dx = 2x2 dx = (1)
0 0 0 3
The second moment of X is

' 1' 1 '
1
1
E X2 = 3
4x y dy dx = 2x3 dx = (2)
0 0 0 2
The variance of X is Var[X] = E[X 2 ] − (E[X])2 = 1/2 − (2/3)2 = 1/18.
(b) The mean of Y is

' 1' 1 ' 1
2 4x 2
E [Y ] = 4xy dy dx = dx = (3)
0 0 0 3 3
' 1' 1 '
1
1
E Y2 = 3
4xy dy dx = x dx = (4)
0 0 0 2
The variance of Y is Var[Y ] = E[Y 2 ] − (E[Y ])2 = 1/2 − (2/3)2 = 1/18.
(c) To find the covariance, we first find the correlation

' 1' 1 ' 1
2 2 4x2 4
E [XY ] = 4x y dy dx = dx = (5)
0 0 0 3 9
The covariance is thus

2
4 2
Cov [X, Y ] = E [XY ] − E [X] E [Y ] = − =0 (6)
9 3
(d) E[X + Y ] = E[X] + E[Y ] = 2/3 + 2/3 = 4/3.
(e) By Theorem 4.15, the variance of X + Y is
Var[X] + Var[Y ] + 2 Cov [X, Y ] = 1/18 + 1/18 + 0 = 1/9 (7)
155
The joint PDF of X and Y and the region of nonzero probability are
Y
1
5x2 /2 −1 ≤ x ≤ 1, 0 ≤ y ≤ x2
fX,Y (x, y) = (1)
0 otherwise
X
-1 1
(a) The first moment of X is

' ' x2 ' $1
1
5x2 1
5x5 5x6 $$
E [X] = x dy dx = dx = =0 (2)
−1 0 2 −1 2 12 $−1
Since E[X] = 0, the variance of X and the second moment are both
' 1 ' x2 $1
2 2 5x
2 5x7 $$ 10
Var[X] = E X = x dy dx = $ = (3)
−1 0 2 14 −1 14
(b) The first and second moments of Y are

' 1 ' x2
5x2 5
E [Y ] = y dy dx = (4)
−1 0 2 14
' 1'
5x2 5
E Y2 = x2 y 2 dy dx = (5)
−1 0 2 26
Therefore, Var[Y ] = 5/26 − (5/14)2 = .0576.
(c) Since E[X] = 0, Cov[X, Y ] = E[XY ] − E[X]E[Y ] = E[XY ]. Thus,
' 1 ' x2 ' 1
5x2 5x7
Cov [X, Y ] = E [XY ] = xy dy dx = dx = 0 (6)
1 0 2 −1 4
(d) The expected value of the sum X + Y is

5
E [X + Y ] = E [X] + E [Y ] = (7)
14
(e) By Theorem 4.15, the variance of X + Y is
Var[X + Y ] = Var[X] + Var[Y ] + 2 Cov [X, Y ] = 5/7 + 0.0576 = 0.7719 (8)

Y
1

2 0≤y≤x≤1
fX,Y (x, y) = (1)
0 otherwise
X
1
156
Before finding moments, it is helpful to first find the marginal PDFs. For 0 ≤ x ≤ 1,
' ∞ ' x
fX (x) = fX,Y (x, y) dy = 2 dy = 2x (2)
−∞ 0
Note that fX (x) = 0 for x < 0 or x > 1. For 0 ≤ y ≤ 1,

' ∞ ' 1
fY (y) = fX,Y (x, y) dx = 2 dx = 2(1 − y) (3)
−∞ y
Also, for y < 0 or y > 1, fY (y) = 0. Complete expressions for the marginal PDFs are

2x 0 ≤ x ≤ 1 2(1 − y) 0 ≤ y ≤ 1
fX (x) = fY (y) = (4)
(a) The first two moments of X are

' ∞ ' 1
E [X] = xfX (x) dx = 2x2 dx = 2/3 (5)
−∞ 0
' ∞ '
1
E X2 = 2
x fX (x) dx = 2x3 dx = 1/2 (6)
−∞ 0
The variance of X is Var[X] = E[X 2 ] − (E[X])2 = 1/2 − 4/9 = 1/18.
(b) The expected value and second moment of Y are

' ' $1
∞ 1
2y 3 $$
E [Y ] = yfY (y) dy = 2y(1 − y) dy = y − = 1/3 2
(7)
−∞ 0 3 $0
' ∞ ' 1 $1
2 2y 3 y 4 $$
E Y = 2
y fY (y) dy = 2y (1 − y) dy =
2
− $ = 1/6 (8)
−∞ 0 3 2 0
The variance of Y is Var[Y ] = E[Y 2 ] − (E[Y ])2 = 1/6 − 1/9 = 1/18.
(c) Before finding the covariance, we find the correlation

' 1' x ' 1
E [XY ] = 2xy dy dx = x3 dx = 1/4 (9)
0 0 0
The covariance is
Cov [X, Y ] = E [XY ] − E [X] E [Y ] = 1/36. (10)
(d) E[X + Y ] = E[X] + E[Y ] = 2/3 + 1/3 = 1
Var[X + Y ] = Var[X] + Var[Y ] + 2 Cov [X, Y ] = 1/6. (11)
157
Y Random variables X and Y have joint PDF

1/2 −1 ≤ x ≤ y ≤ 1
fX,Y (x, y) = (1)
X
0 otherwise
1
The region of possible pairs (x, y) is shown with the joint PDF. The rest of
-1
this problem is just calculus.
' ' ' $1
1 1
xy 1 1
x2 x4 $$
E [XY ] = dy dx = x(1 − x ) dx =
2
− $ =0 (2)
−1 x 2 4 −1 8 16 −1
' 1' 1
1 x y
E eX+Y = e e dy dx (3)
−1 x 2
' 1
1
= ex (e1 − ex ) dx (4)
2 −1
$
1 1+x 1 2x $$1 e2 e−2 1
= e − e $ = + − (5)
2 4 −1 4 4 2

For this problem, calculating the marginal PMF of K is not easy. However, the marginal PMF of
N is easy to find. For n = 1, 2, . . .,

n
(1 − p)n−1 p
PN (n) = = (1 − p)n−1 p (1)
n
k=1
That is, N has a geometric PMF. From Appendix A, we note that

1 1−p
E [N ] = Var[N ] = (2)
p p2
We can use these facts to find the second moment of N .
2−p
E N 2 = Var[N ] + (E [N ])2 = (3)
p2
Now we can calculate the moments of K.
∞
∞
(1 − p)n−1 p (1 − p)n−1 p
n n
E [K] = k = k (4)
n n
n=1 k=1 n=1 k=1

n
Since k=1 k = n(n + 1)/2,
∞

n+1 N +1 1 1
E [K] = (1 − p) n−1
p=E = + (5)
2 2 2p 2
n=1
We now can calculate the sum of the moments.

3 1
E [N + K] = E [N ] + E [K] = + (6)
2p 2
158
The second moment of K is
∞ ∞
n
(1 − p)n−1 p (1 − p)n−1 p 2
n
E K2 = k2 = k (7)
n n
n=1 k=1 n=1 k=1

n 2
Using the identity k=1 k = n(n + 1)(2n + 1)/6, we obtain
∞
2 (n + 1)(2n + 1) (N + 1)(2N + 1)
E K = (1 − p)n−1
p=E (8)
6 6
n=1
Applying the values of E[N ] and E[N 2 ] found above, we find that

2 E N 2 E [N ] 1 2 1 1
E K = + + = 2+ + (9)
3 2 6 3p 6p 6
Thus, we can calculate the variance of K.
5 1 5
Var[K] = E K 2 − (E [K])2 = − + (10)
12p2 3p 12
To find the correlation of N and K,
∞
∞
(1 − p)n−1 p
n n
E [N K] = nk = (1 − p)n−1 p k (11)
n
n=1 k=1 n=1 k=1

n
Since k=1 k = n(n + 1)/2,
∞

n(n + 1) N (N + 1) 1
E [N K] = (1 − p) n−1
p=E = 2 (12)
2 2 p
n=1
Finally, the covariance is

1 1
Cov [N, K] = E [N K] − E [N ] E [K] = 2
− (13)
2p 2p

The event A occurs iff X > 5 and Y > 5 and has probability

10
10
P [A] = P [X > 5, Y > 5] = 0.01 = 0.25 (1)
x=6 y=6
From Theorem 4.19,

PX,Y (x,y)
P [A] (x, y) ∈ A
PX,Y |A (x, y) = (2)
0 otherwise

0.04 x = 6, . . . , 10; y = 6, . . . , 20
= (3)
0 otherwise
159
The event B occurs iff X ≤ 5 and Y ≤ 5 and has probability

5
5
P [B] = P [X ≤ 5, Y ≤ 5] = 0.01 = 0.25 (1)
x=1 y=1
From Theorem 4.19,

PX,Y (x,y)
P [B] (x, y) ∈ A
PX,Y |B (x, y) = (2)
0 otherwise

0.04 x = 1, . . . , 5; y = 1, . . . , 5
= (3)
0 otherwise

Given the event A = {X + Y ≤ 1}, we wish to find fX,Y |A (x, y). First we find
' 1 ' 1−x
P [A] = 6e−(2x+3y) dy dx = 1 − 3e−2 + 2e−3 (1)
0 0
So then
6e−(2x+3y)
1−3e−2 +2e−3
x + y ≤ 1, x ≥ 0, y ≥ 0
fX,Y |A (x, y) = (2)
0 otherwise

First we observe that for n = 1, 2, . . ., the marginal PMF of N satisfies

n
n
1
PN (n) = PN,K (n, k) = (1 − p)n−1 p = (1 − p)n−1 p (1)
n
k=1 k=1
Thus, the event B has probability

∞

P [B] = PN (n) = (1 − p)9 p[1 + (1 − p) + (1 − p)2 + · · · ] = (1 − p)9 (2)
n=10
From Theorem 4.19,

PN,K (n,k)
P [B] n, k ∈ B
PN,K|B (n, k) = (3)
0 otherwise

(1 − p)n−10 p/n n = 10, 11, . . . ; k = 1, . . . , n
= (4)
0 otherwise
The conditional PMF PN |B (n|b) could be found directly from PN (n) using Theorem 2.17. However,
we can also find it just by summing the conditional joint PMF.

n
(1 − p)n−10 p n = 10, 11, . . .
PN |B (n) = PN,K|B (n, k) = (5)
0 otherwise
k=1
160
From the conditional PMF PN |B (n), we can calculate directly the conditional moments of N given
B. Instead, however, we observe that given B, N = N − 9 has a geometric PMF with mean 1/p.
That is, for n = 1, 2, . . .,
PN |B (n) = P [N = n + 9|B] = PN |B (n + 9) = (1 − p)n−1 p (6)
Hence, given B, N = N + 9 and we can calculate the conditional expectations

E [N |B] = E N + 9|B = E N |B + 9 = 1/p + 9 (7)
Var[N |B] = Var[N + 9|B] = Var[N |B] = (1 − p)/p2 (8)
Note that further along in the problem we will need E[N 2 |B] which we now calculate.

E N 2 |B = Var[N |B] + (E [N |B])2 (9)
2 17
= 2+ + 81 (10)
p p
For the conditional moments of K, we work directly with the conditional PMF PN,K|B (n, k).
∞
∞
(1 − p)n−10 p
n n
(1 − p)n−10 p
E [K|B] = k = k (11)
n n
n=10 k=1 n=10 k=1

n
Since k=1 k = n(n + 1)/2,
∞
n+1 1 1
E [K|B] = (1 − p)n−1 p = E [N + 1|B] = +5 (12)
2 2 2p
n=1
We now can calculate the conditional expectation of the sum.

3
E [N + K|B] = E [N |B] + E [K|B] = 1/p + 9 + 1/(2p) + 5 = + 14 (13)
2p
The conditional second moment of K is
∞
∞

(1 − p)n−10 p 2
n n
2 (1 − p)n−10 p
E K |B = 2
k = k (14)
n n
n=10 k=1 n=10 k=1

n 2
Using the identity k=1 k = n(n + 1)(2n + 1)/6, we obtain
∞

(n + 1)(2n + 1) 1
E K |B =2
(1 − p)n−10 p = E [(N + 1)(2N + 1)|B] (15)
6 6
n=10
Applying the values of E[N |B] and E[N 2 |B] found above, we find that

2 E N 2 |B E [N |B] 1 2 37 2
E K |B = + + = 2+ + 31 (16)
3 2 6 3p 6p 3
Thus, we can calculate the conditional variance of K.
5 7 2
Var[K|B] = E K 2 |B − (E [K|B])2 = 2
− +6 (17)
12p 6p 3
161
To find the conditional correlation of N and K,
∞
n ∞

n
(1 − p)n−10 p
E [N K|B] = nk = (1 − p)n−1
p k (18)
n
n=10 k=1 n=10 k=1

n
Since k=1 k = n(n + 1)/2,
∞
n(n + 1) 1 1 9
E [N K|B] = (1 − p)n−10 p = E [N (N + 1)|B] = 2 + + 45 (19)
2 2 p p
n=10


(x + y)/3 0 ≤ x ≤ 1, 0 ≤ y ≤ 2
fX,Y (x, y) = (1)
0 otherwise
(a) The probability that Y ≤ 1 is
''
P [A] = P [Y ≤ 1] = fX,Y (x, y) dx dy (2)
Y
y≤1
2 ' 1' 1
x+y
= dy dx (3)
Y 1
3
' 1! $y=1 "
1 0 0
xy y 2 $$
= + $ dx (4)
X 0 3 6 y=0
1 ' $1
1
2x + 1 x2 x $$ 1
= dx = + $ = (5)
0 6 6 6 0 3
(b) By Definition 4.10, the conditional joint PDF of X and Y given A is

fX,Y (x,y)
P [A] (x, y) ∈ A x + y 0 ≤ x ≤ 1, 0 ≤ y ≤ 1
fX,Y |A (x, y) = = (6)
From fX,Y |A (x, y), we find the conditional marginal PDF fX|A (x). For 0 ≤ x ≤ 1,
' ∞ ' 1 $y=1
y2 $ 1
fX|A (x) = fX,Y |A (x, y) dy = (x + y) dy = xy + $$ =x+ (7)
−∞ 0 2 y=0 2
The complete expression is

x + 1/2 0 ≤ x ≤ 1
fX|A (x) = (8)
0 otherwise
For 0 ≤ y ≤ 1, the conditional marginal PDF of Y is
' ∞ ' 1 $x=1
x2 $
fY |A (y) = fX,Y |A (x, y) dx = (x + y) dx = + xy $$ = y + 1/2 (9)
−∞ 0 2 x=0

y + 1/2 0 ≤ y ≤ 1
fY |A (y) = (10)
0 otherwise
162

(4x + 2y)/3 0 ≤ x ≤ 1, 0 ≤ y ≤ 1
fX,Y (x, y) = (1)
0 otherwise
(a) The probability of event A = {Y ≤ 1/2} is

'' ' 1 ' 1/2
4x + 2y
P [A] = fX,Y (x, y) dy dx = dy dx. (2)
y≤1/2 0 0 3
With some calculus,

' $y=1/2 ' 1 $1
1
4xy + y 2 $$ 2x + 1/4 x2 x $$ 5
P [A] = $ dx = dx = + $ = 12 . (3)
0 3 y=0 0 3 3 12 0
(b) The conditional joint PDF of X and Y given A is

fX,Y (x,y)
P [A] (x, y) ∈ A
fX,Y |A (x, y) = (4)
0 otherwise

8(2x + y)/5 0 ≤ x ≤ 1, 0 ≤ y ≤ 1/2
= (5)
0 otherwise
For 0 ≤ x ≤ 1, the PDF of X given A is

' ∞ ' 1/2
8
fX|A (x) = fX,Y |A (x, y) dy = (2x + y) dy (6)
−∞ 5 0
$y=1/2
8 y 2 $$ 8x + 1
= 2xy + $ = (7)
5 2 y=0 5

(8x + 1)/5 0 ≤ x ≤ 1
fX|A (x) = (8)
0 otherwise
For 0 ≤ y ≤ 1/2, the conditional marginal PDF of Y given A is

' ∞ ' 1
8
fY |A (y) = fX,Y |A (x, y) dx = (2x + y) dx (9)
−∞ 5 0
$x=1
2
8x + 8xy $$ 8y + 8
= $ = (10)
5 x=0 5

(8y + 8)/5 0 ≤ y ≤ 1/2
fY |A (y) = (11)
0 otherwise
163
(a) The event A = {Y ≤ 1/4} has probability

' 1/2 ' x2
5x2
Y P [A] = 2 dy dx (1)
0 0 2
' 1 ' 1/4
1 Y<1/4
5x2
+2 dy dx
1/2 0 2
' 1/2 ' 1
¼ 4 5x2
= 5x dx + dx (2)
X 0 1/2 4
$1/2 $1
-1 -½ ½ 1 = x5 $0 + 5x3 /12$1/2 (3)
= 19/48 (4)
This implies

fX,Y (x, y) /P [A] (x, y) ∈ A
fX,Y |A (x, y) = (5)
0 otherwise

120x2 /19 −1 ≤ x ≤ 1, 0 ≤ y ≤ x2 , y ≤ 1/4
= (6)
0 otherwise
(b)
' ∞ '
1
120x2 80
19 (1 − y 3/2 ) 0 ≤ y ≤ 1/4
fY |A (y) = fX,Y |A (x, y) dx = 2 √
dx = (7)
−∞ y 19 0 otherwise
(c) The conditional expectation of Y given A is

' ! "$1/4
1/4
y 2 2y 7/2 $
80 80 $ 65
E [Y |A] = y (1 − y 3/2 ) dy = − $ = (8)
0 19 19 2 7 $ 532
0
(∞
(d) To find fX|A (x), we can write fX|A (x) = −∞ fX,Y |A (x, y) dy. However, when we substitute
fX,Y |A (x, y), the limits will depend on the value of x. When |x| ≤ 1/2,
' x2
120x2 120x4
fX|A (x) = dy = (9)
0 19 19
When −1 ≤ x ≤ −1/2 or 1/2 ≤ x ≤ 1,
' 1/4
120x2 30x2
fX|A (x) = dy = (10)
0 19 19
The complete expression for the conditional PDF of X given A is
⎧
⎪
⎪ 30x2 /19 −1 ≤ x ≤ −1/2
⎨
120x4 /19 −1/2 ≤ x ≤ 1/2
fX|A (x) = (11)
⎪
⎪ 30x2 /19 1/2 ≤ x ≤ 1
⎩
0 otherwise
164
(e) The conditional mean of X given A is
' −1/2 ' 1/2 ' 1
30x3 120x5 30x3
E [X|A] = dx + dx + dx = 0 (12)
−1 19 −1/2 19 1/2 19

The main part of this problem is just interpreting the problem statement. No calculations are
necessary. Since a trip is equally likely to last 2, 3 or 4 days,

1/3 d = 2, 3, 4
PD (d) = (1)
0 otherwise
Given a trip lasts d days, the weight change is equally likely to be any value between −d and d
pounds. Thus,
1/(2d + 1) w = −d, −d + 1, . . . , d
PW |D (w|d) = (2)
0 otherwise
The joint PMF is simply
PD,W (d, w) = PW |D (w|d) PD (d) (3)

1/(6d + 3) d = 2, 3, 4; w = −d, . . . , d
= (4)
0 otherwise

We can make a table of the possible outcomes and the corresponding values of W and Y
outcome P [·] W Y
hh p2 0 2
ht p(1 − p) 1 1 (1)
th p(1 − p) −1 1
tt (1 − p)2 0 0
In the following table, we write the joint PMF PW,Y (w, y) along with the marginal PMFs PY (y)
and PW (w).
PW,Y (w, y) w = −1 w=0 w=1 PY (y)
y=0 0 (1 − p)2 0 (1 − p)2
y=1 p(1 − p) 0 p(1 − p) 2p(1 − p) (2)
y=2 0 p2 0 p2
PW (w) p(1 − p) 1 − 2p + 2p2 p(1 − p)
Using the definition PW |Y (w|y) = PW,Y (w, y)/PY (y), we can find the conditional PMFs of W given
Y.

1 w=0 1/2 w = −1, 1
PW |Y (w|0) = PW |Y (w|1) = (3)

1 w=0
PW |Y (w|2) = (4)
0 otherwise
165
Similarly, the conditional PMFs of Y given W are
⎧ (1−p)2
⎪
⎨ 1−2p+2p2
y=0
1 y=1 p2
PY |W (y| − 1) = PY |W (y|0) = y=2 (5)
0 otherwise ⎪
⎩ 1−2p+2p2
0 otherwise

1 y=1
PY |W (y|1) = (6)
0 otherwise

(x + y) 0 ≤ x, y ≤ 1
fX,Y (x, y) = (1)
0 otherwise
(a) The conditional PDF fX|Y (x|y) is defined for all y such that 0 ≤ y ≤ 1. For 0 ≤ y ≤ 1,

(x+y)
fX,Y (x, y) (x + y) 0≤x≤1
fX|Y (x) = = (1 = x+1/2 (2)
fX (x) 0 otherwise
0 (x + y) dy
(b) The conditional PDF fY |X (y|x) is defined for all values of x in the interval [0, 1]. For 0 ≤
x ≤ 1,

(x+y)
y+1/2 0 ≤ y ≤ 1
fX,Y (x, y) (x + y)
fY |X (y) = = (1 = (3)
fY (y) 0 otherwise
0 (x + y) dx

Y
1

2 0≤y≤x≤1
fX,Y (x, y) = (1)
0 otherwise
X
1
For 0 ≤ y ≤ 1,
' ∞ ' 1
fY (y) = fX,Y (x, y) dx = 2 dx = 2(1 − y) (2)
−∞ y
Also, for y < 0 or y > 1, fY (y) = 0. The complete expression for the marginal PDF is

2(1 − y) 0 ≤ y ≤ 1
fY (y) = (3)
0 otherwise
By Theorem 4.24, the conditional PDF of X given Y is

1
fX,Y (x, y) 1−y y≤x≤1
fX|Y (x|y) = = (4)
fY (y) 0 otherwise
166
That is, since Y ≤ X ≤ 1, X is uniform over [y, 1] when Y = y. The conditional expectation of X
given Y = y can be calculated as
' ∞
E [X|Y = y] = xfX|Y (x|y) dx (5)
−∞
' $1
1
x x2 $$ 1+y
= dx = = (6)
y 1−y 2(1 − y) $y 2
In fact, since we know that the conditional PDF of X is uniform over [y, 1] when Y = y, it wasn’t
really necessary to perform the calculation.

Y
1

2 0≤y≤x≤1
fX,Y (x, y) = (1)
0 otherwise
X
1
For 0 ≤ x ≤ 1, the marginal PDF for X satisfies
' ∞ ' x
fX (x) = fX,Y (x, y) dy = 2 dy = 2x (2)
−∞ 0
Note that fX (x) = 0 for x < 0 or x > 1. Hence the complete expression for the marginal PDF of
X is
2x 0 ≤ x ≤ 1
fX (x) = (3)
0 otherwise
The conditional PDF of Y given X = x is

fX,Y (x, y) 1/x 0 ≤ y ≤ x
fY |X (y|x) = = (4)
fX (x) 0 otherwise
Given X = x, Y has a uniform PDF over [0, x] and thus (has conditional expected value E[Y |X = x] =
∞
x/2. Another way to obtain this result is to calculate −∞ yfY |X (y|x) dy.

We are told in the problem statement that if we know r, the number of feet a student sits from the
blackboard, then we also know that that student’s grade is a Gaussian random variable with mean
80 − r and standard deviation r. This is exactly
1 2
/2r 2
fX|R (x|r) = √ e−(x−[80−r]) (1)
2πr2
167
(a) First we observe that A takes on the values SA = {−1, 1} while B takes on values from
SB = {0, 1}. To construct a table describing PA,B (a, b) we build a table for all possible values
of pairs (A, B). The general form of the entries is
PA,B (a, b) b=0 b=1

a = −1 PB|A (0| − 1) PA (−1) PB|A (1| − 1) PA (−1) (1)
a=1 PB|A (0|1) PA (1) PB|A (1|1) PA (1)
Now we fill in the entries using the conditional PMFs PB|A (b|a) and the marginal PMF PA (a).
This yields
PA,B (a, b) b=0 b=1 PA,B (a, b) b = 0 b = 1

a = −1 (1/3)(1/3) (2/3)(1/3) which simplifies to a = −1 1/9 2/9 (2)
a=1 (1/2)(2/3) (1/2)(2/3) a=1 1/3 1/3
(b) Since PA (1) = PA,B (1, 0) + PA,B (1, 1) = 2/3,

PA,B (1, b) 1/2 b = 0, 1,
PB|A (b|1) = = (3)
PA (1) 0 otherwise.
If A = 1, the conditional expectation of B is

1
E [B|A = 1] = bPB|A (b|1) = PB|A (1|1) = 1/2. (4)
b=0
(c) Before finding the conditional PMF PA|B (a|1), we first sum the columns of the joint PMF
table to find
4/9 b = 0
PB (b) = (5)
5/9 b = 1
The conditional PMF of A given B = 1 is

PA,B (a, 1) 2/5 a = −1
PA|B (a|1) = = (6)
PB (1) 3/5 a = 1
(d) Now that we have the conditional PMF PA|B (a|1), calculating conditional expectations is
easy.

E [A|B = 1] = aPA|B (a|1) = −1(2/5) + (3/5) = 1/5 (7)
a=−1,1

E A2 |B = 1 = a2 PA|B (a|1) = 2/5 + 3/5 = 1 (8)
a=−1,1
The conditional variance is then

Var[A|B = 1] = E A2 |B = 1 − (E [A|B = 1])2 = 1 − (1/5)2 = 24/25 (9)
168
(e) To calculate the covariance, we need

E [A] = aPA (a) = −1(1/3) + 1(2/3) = 1/3 (10)
a=−1,1

1
E [B] = bPB (b) = 0(4/9) + 1(5/9) = 5/9 (11)
b=0

1
E [AB] = abPA,B (a, b) (12)
a=−1,1 b=0
= −1(0)(1/9) + −1(1)(2/9) + 1(0)(1/3) + 1(1)(1/3) = 1/9 (13)
The covariance is just
Cov [A, B] = E [AB] − E [A] E [B] = 1/9 − (1/3)(5/9) = −2/27 (14)

First we need to find the conditional expectations

1
E [B|A = −1] = bPB|A (b| − 1) = 0(1/3) + 1(2/3) = 2/3 (1)
b=0
1
E [B|A = 1] = bPB|A (b|1) = 0(1/2) + 1(1/2) = 1/2 (2)
b=0
Keep in mind that E[B|A] is a random variable that is a function of A. that is we can write

2/3 A = −1
E [B|A] = g(A) = (3)
1/2 A = 1
We see that the range of U is SU = {1/2, 2/3}. In particular,
PU (1/2) = PA (1) = 2/3 PU (2/3) = PA (−1) = 1/3 (4)
The complete PMF of U is

2/3 u = 1/2
PU (u) = (5)
1/3 u = 2/3
Note that
E [E [B|A]] = E [U ] = uPU (u) = (1/2)(2/3) + (2/3)(1/3) = 5/9 (6)
u
You can check that E[U ] = E[B].

Random variables N and K have the joint PMF
⎧
⎨ 100n e−100 k = 0, 1, . . . , n;
PN,K (n, k) = (n+1)! n = 0, 1, . . . (1)
⎩
0 otherwise
169
We can find the marginal PMF for N by summing over all possible K. For n ≥ 0,

n
100n e−100 100n e−100
PN (n) = = (2)
(n + 1)! n!
k=0
We see that N has a Poisson PMF with expected value 100. For n ≥ 0, the conditional PMF of K
given N = n is
PN,K (n, k) 1/(n + 1) k = 0, 1, . . . , n
PK|N (k|n) = = (3)
PN (n) 0 otherwise
That is, given N = n, K has a discrete uniform PMF over {0, 1, . . . , n}. Thus,

n
E [K|N = n] = k/(n + 1) = n/2 (4)
k=0
We can conclude that E[K|N ] = N/2. Thus, by Theorem 4.25,
E [K] = E [E [K|N ]] = E [N/2] = 50. (5)

This problem is fairly easy when we use conditional PMF’s. In particular, given that N = n pizzas
were sold before noon, each of those pizzas has mushrooms with probability 1/3. The conditional
PMF of M given N is the binomial distribution
n m n−m
PM |N (m|n) = m (1/3) (2/3) m = 0, 1, . . . , n
(1)
0 otherwise
The other fact we know is that for each of the 100 pizzas sold, the pizza is sold before noon with
probability 1/2. Hence, N has the binomial PMF
100 n 100−n n = 0, 1, . . . , 100
PN (n) = n (1/2) (1/2) (2)
0 otherwise
The joint PMF of N and M is for integers m, n,
PM,N (m, n) = PM |N (m|n) PN (n) (3)

n 100 m n−m (1/2)100 0 ≤ m ≤ n ≤ 100
= m n (1/3) (2/3) (4)
0 otherwise

Y

1/2 −1 ≤ x ≤ y ≤ 1
X fX,Y (x, y) = (1)
0 otherwise
1
-1
170
(a) For −1 ≤ y ≤ 1, the marginal PDF of Y is
' ∞ '
1 y
fY (y) = fX,Y (x, y) dx = dx = (y + 1)/2 (2)
−∞ 2 −1
The complete expression for the marginal PDF of Y is

(y + 1)/2 −1 ≤ y ≤ 1
fY (y) = (3)
0 otherwise
(b) The conditional PDF of X given Y is

fX,Y (x, y) 1
1+y −1 ≤ x ≤ y
fX|Y (x|y) = = (4)
fY (y) 0 otherwise
(c) Given Y = y, the conditional PDF of X is uniform over [−1, y]. Hence the conditional
expected value is E[X|Y = y] = (y − 1)/2.

We are given that the joint PDF of X and Y is

1/(πr2 ) 0 ≤ x2 + y 2 ≤ r2
fX,Y (x, y) = (1)
0 otherwise

' √ √
r 2 −x2 2 r 2 −x2
1
πr 2
−r ≤ x ≤ r
fX (x) = 2 dy = (2)
0 πr2 0 otherwise
The conditional PDF of Y given X is

√
fX,Y (x, y) 1/(2 r2 − x2 ) y 2 ≤ r2 − x2
fY |X (y|x) = = (3)
fX (x) 0 otherwise
√ √
(b) Given X = x, we observe that over the interval [− r2 − x2 , r2 − x2 ], Y has a uniform PDF.
Since the conditional PDF fY |X (y|x) is symmetric about y = 0,
E [Y |X = x] = 0 (4)

The key to solving this problem is to find the joint PMF of M and N . Note that N ≥ M . For
n > m, the joint event {M = m, N = n} has probability
m−1 n−m−1
calls calls

P [M = m, N = n] = P [dd · · · d v dd · · · d v] (1)
= (1 − p) m−1
p(1 − p) n−m−1
p (2)
= (1 − p)n−2 p2 (3)
171
A complete expression for the joint PMF of M and N is

(1 − p)n−2 p2 m = 1, 2, . . . , n − 1; n = m + 1, m + 2, . . .
PM,N (m, n) = (4)
0 otherwise
The marginal PMF of N satisfies

n−1
PN (n) = (1 − p)n−2 p2 = (n − 1)(1 − p)n−2 p2 , n = 2, 3, . . . (5)
m=1
Similarly, for m = 1, 2, . . ., the marginal PMF of M satisfies

∞

PM (m) = (1 − p)n−2 p2 (6)
n=m+1
= p [(1 − p)m−1 + (1 − p)m + · · · ]
2
(7)
= (1 − p) m−1
p (8)
The complete expressions for the marginal PMF’s are

(1 − p)m−1 p m = 1, 2, . . .
PM (m) = (9)
0 otherwise

(n − 1)(1 − p)n−2 p2 n = 2, 3, . . .
PN (n) = (10)
0 otherwise
Not surprisingly, if we view each voice call as a successful Bernoulli trial, M has a geometric PMF
since it is the number of trials up to and including the first success. Also, N has a Pascal PMF
since it is the number of trials required to see 2 successes. The conditional PMF’s are now easy to
find.

PM,N (m, n) (1 − p)n−m−1 p n = m + 1, m + 2, . . .
PN |M (n|m) = = (11)
PM (m) 0 otherwise
The interpretation of the conditional PMF of N given M is that given M = m, N = m + N where

N has a geometric PMF with mean 1/p. The conditional PMF of M given N is

PM,N (m, n) 1/(n − 1) m = 1, . . . , n − 1
PM |N (m|n) = = (12)
PN (n) 0 otherwise
Given that call N = n was the second voice call, the first voice call is equally likely to occur in any
of the previous n − 1 calls.
(a) The number of buses, N , must be greater than zero. Also, the number of minutes that
pass cannot be less than the number of buses. Thus, P [N = n, T = t] > 0 for integers n, t
satisfying 1 ≤ n ≤ t.
172
(b) First, we find the joint PMF of N and T by carefully considering the possible sample paths.
In particular, PN,T (n, t) = P [ABC] = P [A]P [B]P [C] where the events A, B and C are
A = {n − 1 buses arrive in the first t − 1 minutes} (1)
B = {none of the first n − 1 buses are boarded} (2)
C = {at time t a bus arrives and is boarded} (3)
These events are independent since each trial to board a bus is independent of when the buses
arrive. These events have probabilities

t − 1 n−1
P [A] = p (1 − p)t−1−(n−1) (4)
n−1
P [B] = (1 − q)n−1 (5)
P [C] = pq (6)
Consequently, the joint PMF of N and T is
t−1 n−1
n−1 p (1 − p)t−n (1 − q)n−1 pq n ≥ 1, t ≥ n
PN,T (n, t) = (7)
0 otherwise
(c) It is possible to find the marginal PMF’s by summing the joint PMF. However, it is much
easier to obtain the marginal PMFs by consideration of the experiment. Specifically, when a
bus arrives, it is boarded with probability q. Moreover, the experiment ends when a bus is
boarded. By viewing whether each arriving bus is boarded as an independent trial, N is the
number of trials until the first success. Thus, N has the geometric PMF

(1 − q)n−1 q n = 1, 2, . . .
PN (n) = (8)
0 otherwise
To find the PMF of T , suppose we regard each minute as an independent trial in which a
success occurs if a bus arrives and that bus is boarded. In this case, the success probability
is pq and T is the number of minutes up to and including the first success. The PMF of T is
also geometric.
(1 − pq)t−1 pq t = 1, 2, . . .
PT (t) = (9)
0 otherwise
(d) Once we have the marginal PMFs, the conditional PMFs are easy to find.
n−1 t−1−(n−1)
PN,T (n, t) t−1 p(1−q) 1−p
PN |T (n|t) = = n−1 1−pq 1−pq n = 1, 2, . . . , t (10)
PT (t) 0 otherwise
That is, given you depart at time T = t, the number of buses that arrive during minutes
1, . . . , t−1 has a binomial PMF since in each minute a bus arrives with probability p. Similarly,
the conditional PMF of T given N is
t−1 n
n−1 p (1 − p)
PN,T (n, t) t−n t = n, n + 1, . . .
PT |N (t|n) = = (11)
PN (n) 0 otherwise
This result can be explained. Given that you board bus N = n, the time T when you leave
is the time for n buses to arrive. If we view each bus arrival as a success of an independent
trial, the time for n buses to arrive has the above Pascal PMF.
173
If you construct a tree describing what type of call (if any) that arrived in any 1 millisecond period,
it will be apparent that a fax call arrives with probability α = pqr or no fax arrives with probability
1 − α. That is, whether a fax message arrives each millisecond is a Bernoulli trial with success
probability α. Thus, the time required for the first success has the geometric PMF

(1 − α)t−1 α t = 1, 2, . . .
PT (t) = (1)
0 otherwise
Note that N is the number of trials required to observe 100 successes. Moreover, the number of
trials needed to observe 100 successes is N = T + N where N is the number of trials needed
to observe successes 2 through 100. Since N is just the number of trials needed to observe 99
successes, it has the Pascal (k = 99, p) PMF

n − 1 99
PN (n) = α (1 − α)n−99 . (2)
98
Since the trials needed to generate successes 2 though 100 are independent of the trials that yield
the first success, N and T are independent. Hence
PN |T (n|t) = PN |T (n − t|t) = PN (n − t) . (3)
Applying the PMF of N found above, we have

n − t − 1 99
PN |T (n|t) = α (1 − α)n−t−99 . (4)
98
Finally the joint PMF of N and T is
PN,T (n, t) = PN |T (n|t) PT (t) (5)
n−t−1 100
98 α (1 − α)n−100 t = 1, 2, . . . ; n = 99 + t, 100 + t, . . .
= (6)
0 otherwise
This solution can also be found a consideration of the sample sequence of Bernoulli trials in which
we either observe or do not observe a fax message.
To find the conditional PMF PT |N (t|n), we first must recognize that N is simply the number
of trials needed to observe 100 successes and thus has the Pascal PMF

n − 1 100
PN (n) = α (1 − α)n−100 (7)
99
Hence for any integer n ≥ 100, the conditional PMF is
⎧ n−t−1
PN,T (n, t) ⎨ ( n−198 )
t = 1, 2, . . . , n − 99
PT |N (t|n) = = ( 99 ) (8)
PN (n) ⎩ 0 otherwise.

Flip a fair coin 100 times and let X be the number of heads in the first 75 flips and Y be the
number of heads in the last 25 flips. We know that X and Y are independent and can find their
PMFs easily.

75 75 25
PX (x) = (1/2) PY (y) = (1/2)25 (1)
x y
174
The joint PMF of X and N can be expressed as the product of the marginal PMFs because we
know that X and Y are independent.

75 25
PX,Y (x, y) = (1/2)100 (2)
x y

Using the following probability model
⎧
⎨ 3/4 k = 0
PX (k) = PY (k) = 1/4 k = 20 (1)
⎩
0 otherwise
We can calculate the requested moments.
E [X] = 3/4 · 0 + 1/4 · 20 = 5 (2)
Var[X] = 3/4 · (0 − 5)2 + 1/4 · (20 − 5)2 = 75 (3)
E [X + Y ] = E [X] + E [X] = 2E [X] = 10 (4)
Since X and Y are independent, Theorem 4.27 yields
Var[X + Y ] = Var[X] + Var[Y ] = 2 Var[X] = 150 (5)
Since X and Y are independent, PX,Y (x, y) = PX (x)PY (y) and

E XY 2XY = XY 2XY PX,Y (x, y) = (20)(20)220(20) PX (20) PY (20) (6)
x=0,20 y=0,20
= 2.75 × 1012 (7)
(a) Normally, checking independence requires the marginal PMFs. However, in this problem, the
zeroes in the table of the joint PMF PX,Y (x, y) allows us to verify very quickly that X and
Y are dependent. In particular, PX (−1) = 1/4 and PY (1) = 14/48 but
PX,Y (−1, 1) = 0 = PX (−1) PY (1) (1)
(b) To fill in the tree diagram, we need the marginal PMF PX (x) and the conditional PMFs
PY |X (y|x). By summing the rows on the table for the joint PMF, we obtain
PX,Y (x, y) y = −1 y = 0 y = 1 PX (x)
x = −1 3/16 1/16 0 1/4
(2)
x=0 1/6 1/6 1/6 1/2
x=1 0 1/8 1/8 1/4
Now we use the conditional PMF PY |X (y|x) = PX,Y (x, y)/PX (x) to write
⎧
⎨ 3/4 y = −1
1/3 y = −1, 0, 1
PY |X (y| − 1) = 1/4 y = 0 PY |X (y|0) = (3)
⎩ 0 otherwise
0 otherwise

1/2 y = 0, 1
PY |X (y|1) = (4)
0 otherwise
175
Now we can us these probabilities to label the tree. The generic solution and the specific
solution with the exact values are
Y =−1
PY |X (−1|−1) Y =−1
3/4
X=−1
Y =0 X=−1
Y =0
PY |X (0|−1) 1/4
PX (−1) 1/4
Y =−1
PY |X (−1|0) Y =−1
1/3

@ PX (0)
X=0 H
PY |X (0|0)
HH
Y =0
@ 1/2
X=0 HH 1/3 Y =0
@ H @ HH
PY |X (1|0) H
H 1/3 H
H
@ Y =1 @ Y =1
@
PX (1)
@
1/4
@
@ X=1
P
XYX
|X (0|1)
Y =0
@
@ X=1 XXX1/2 Y =0
XX XX
X
X
PY |X (1|1) Y =1 1/2 X Y =1

In the solution to Problem 4.9.10, we found that the conditional PMF of M given N is
n m n−m m = 0, 1, . . . , n
PM |N (m|n) = m (1/3) (2/3) (1)
0 otherwise
Since PM |N (m|n) depends on the event N = n, we see that M and N are dependent.

We can solve this problem for the general case when the probability of heads is p. For the fair coin,
p = 1/2. Viewing each flip as a Bernoulli trial in which heads is a success, the number of flips until
heads is the number of trials needed for the first success which has the geometric PMF

(1 − p)x−1 p x = 1, 2, . . .
PX1 (x) = (1)
0 otherwise
Similarly, no matter how large X1 may be, the number of additional flips for the second heads
is the same experiment as the number of flips needed for the first occurrence of heads. That
is, PX2 (x) = PX1 (x). Moreover, the flips needed to generate the second occurrence of heads are
independent of the flips that yield the first heads. Hence, it should be apparent that X1 and X2
are independent and

(1 − p)x1 +x2 −2 p2 x1 = 1, 2, . . . ; x2 = 1, 2, . . .
PX1 ,X2 (x1 , x2 ) = PX1 (x1 ) PX2 (x2 ) = (2)
0 otherwise
However, if this independence is not obvious, it can be derived by examination of the sample path.
When x1 ≥ 1 and x2 ≥ 1, the event {X1 = x1 , X2 = x2 } occurs iff we observe the sample sequence
tt · · · t h tt
· · · t h
(3)
x1 − 1 times x2 − 1 times
The above sample sequence has probability (1−p)x1 −1 p(1−p)x2 −1 p which in fact equals PX1 ,X2 (x1 , x2 )
given earlier.
176
We will solve this problem when the probability of heads is p. For the fair coin, p = 1/2. The
number X1 of flips until the first heads and the number X2 of additional flips for the second heads
both have the geometric PMF

(1 − p)x−1 p x = 1, 2, . . .
PX1 (x) = PX2 (x) = (1)
0 otherwise
Thus, E[Xi ] = 1/p and Var[Xi ] = (1 − p)/p2 . By Theorem 4.14,
E [Y ] = E [X1 ] − E [X2 ] = 0 (2)
Since X1 and X2 are independent, Theorem 4.27 says
2(1 − p)
Var[Y ] = Var[X1 ] + Var[−X2 ] = Var[X1 ] + Var[X2 ] = (3)
p2

X and Y are independent random variables with PDFs
1 −x/3 1 −y/2
3 e x ≥ 0 2e y≥0
fX (x) = fY (y) = (1)
(a) To calculate P [X > Y ], we use the joint PDF fX,Y (x, y) = fX (x)fY (y).
''
P [X > Y ] = fX (x) fY (y) dx dy (2)
x>y
' ∞ '
1 −y/2 ∞ 1 −x/3
= e e dx dy (3)
0 2 y 3
' ∞
1 −y/2 −y/3
= e e dy (4)
2
' ∞
0
1 −(1/2+1/3)y 1/2 3
= e dy = = (5)
0 2 1/2 + 2/3 7
(b) Since X and Y are exponential random variables with parameters λX = 1/3 and λY = 1/2,
Appendix A tells us that E[X] = 1/λX = 3 and E[Y ] = 1/λY = 2. Since X and Y are
independent, the correlation is E[XY ] = E[X]E[Y ] = 6.
(c) Since X and Y are independent, Cov[X, Y ] = 0.
(a) Since E[−X2 ] = −E[X2 ], we can use Theorem 4.13 to write
E [X1 − X2 ] = E [X1 + (−X2 )] = E [X1 ] + E [−X2 ] (1)

= E [X1 ] − E [X2 ] (2)
=0 (3)
177
(b) By Theorem 3.5(f), Var[−X2 ] = (−1)2 Var[X2 ] = Var[X2 ]. Since X1 and X2 are independent,
Theorem 4.27(a) says that
Var[X1 − X2 ] = Var[X1 + (−X2 )] (4)

= Var[X1 ] + Var[−X2 ] (5)
= 2 Var[X] (6)

Since X and Y are take on only integer values, W = X + Y is integer valued as well. Thus for an
integer w,
PW (w) = P [W = w] = P [X + Y = w] . (1)
Suppose X = k, then W = w if and only if Y = w − k. To find all ways that X + Y = w, we must

consider each possible integer k such that X = k. Thus
∞
∞

PW (w) = P [X = k, Y = w − k] = PX,Y (k, w − k) . (2)
k=−∞ k=−∞
Since X and Y are independent, PX,Y (k, w − k) = PX (k)PY (w − k). It follows that for any integer
w,
∞

PW (w) = PX (k) PY (w − k) . (3)
k=−∞

The key to this problem is understanding that “short order” and “long order” are synonyms for
N = 1 and N = 2. Similarly, “vanilla”, “chocolate”, and “strawberry” correspond to the events
D = 20, D = 100 and D = 300.
(a) The following table is given in the problem statement.
vanilla choc. strawberry

short
0.2 0.2 0.2
order
long
0.1 0.2 0.1
order
This table can be translated directly into the joint PMF of N and D.
PN,D (n, d) d = 20 d = 100 d = 300

n=1 0.2 0.2 0.2 (1)
n=2 0.1 0.2 0.1
178
(b) We find the marginal PMF PD (d) by summing the columns of the joint PMF. This yields
⎧
⎪
⎪ 0.3 d = 20,
⎨
0.4 d = 100,
PD (d) = (2)
⎪
⎪ 0.3 d = 300,
⎩
0 otherwise.
(c) To find the conditional PMF PD|N (d|2), we first need to find the probability of the condition-
ing event
PN (2) = PN,D (2, 20) + PN,D (2, 100) + PN,D (2, 300) = 0.4 (3)
The conditional PMF of N D given N = 2 is
⎧
⎪
⎪ 1/4 d = 20
PN,D (2, d) ⎨ 1/2 d = 100
PD|N (d|2) = = (4)
PN (2) ⎪
⎪ 1/4 d = 300
⎩
0 otherwise
(d) The conditional expectation of D given N = 2 is

E [D|N = 2] = dPD|N (d|2) = 20(1/4) + 100(1/2) + 300(1/4) = 130 (5)
d
(e) To check independence, we could calculate the marginal PMFs of N and D. In this case,
however, it is simpler to observe that PD (d) = PD|N (d|2). Hence N and D are dependent.
(f) In terms of N and D, the cost (in cents) of a fax is C = N D. The expected value of C is

E [C] = ndPN,D (n, d) (6)
n,d
= 1(20)(0.2) + 1(100)(0.2) + 1(300)(0.2) (7)

+ 2(20)(0.3) + 2(100)(0.4) + 2(300)(0.3) = 356 (8)

The key to this problem is understanding that “Factory Q” and “Factory R” are synonyms for
M = 60 and M = 180. Similarly, “small”, “medium”, and “large” orders correspond to the events
B = 1, B = 2 and B = 3.
(a) The following table given in the problem statement
Factory Q Factory R
small order 0.3 0.2
medium order 0.1 0.2
large order 0.1 0.1
can be translated into the following joint PMF for B and M .

PB,M (b, m) m = 60 m = 180
b=1 0.3 0.2
(1)
b=2 0.1 0.2
b=3 0.1 0.1
179
(b) Before we find E[B], it will prove helpful to find the marginal PMFs PB (b) and PM (m). These
can be found from the row and column sums of the table of the joint PMF
PB,M (b, m) m = 60 m = 180 PB (b)

b=1 0.3 0.2 0.5
b=2 0.1 0.2 0.3 (2)
b=3 0.1 0.1 0.2
PM (m) 0.5 0.5
The expected number of boxes is

E [B] = bPB (b) = 1(0.5) + 2(0.3) + 3(0.2) = 1.7 (3)
b
(c) From the marginal PMF of B, we know that PB (2) = 0.3. The conditional PMF of M given
B = 2 is ⎧
PB,M (2, m) ⎨
1/3 m = 60
PM |B (m|2) = = 2/3 m = 180 (4)
PB (2) ⎩
0 otherwise
(d) The conditional expectation of M given B = 2 is

E [M |B = 2] = mPM |B (m|2) = 60(1/3) + 180(2/3) = 140 (5)
m
(e) From the marginal PMFs we calculated in the table of part (b), we can conclude that B and
M are not independent. since PB,M (1, 60) = PB (1)PM (m)60.
(f) In terms of M and B, the cost (in cents) of sending a shipment is C = BM . The expected
value of C is

E [C] = bmPB,M (b, m) (6)
b,m
= 1(60)(0.3) + 2(60)(0.1) + 3(60)(0.1) (7)

+ 1(180)(0.2) + 2(180)(0.2) + 3(180)(0.1) = 210 (8)

Random variables X1 and X2 are iiid with PDF

x/2 0 ≤ x ≤ 2,
fX (x) = (1)
0 otherwise.
(a) Since X1 and X2 are identically distributed they will share the same CDF FX (x).
⎧
' x x≤0
⎨ 0 2
FX (x) = fX x dx = x /4 0 ≤ x ≤ 2 (2)
⎩
0 1 x≥2
180
(b) Since X1 and X2 are independent, we can say that
1
P [X1 ≤ 1, X2 ≤ 1] = P [X1 ≤ 1] P [X2 ≤ 1] = FX1 (1) FX2 (1) = [FX (1)]2 = (3)
16
(c) For W = max(X1 , X2 ),
FW (1) = P [max(X1 , X2 ) ≤ 1] = P [X1 ≤ 1, X2 ≤ 1] (4)
Since X1 and X2 are independent,
FW (1) = P [X1 ≤ 1] P [X2 ≤ 1] = [FX (1)]2 = 1/16 (5)
(d)
FW (w) = P [max(X1 , X2 ) ≤ w] = P [X1 ≤ w, X2 ≤ w] (6)
Since X1 and X2 are independent,

⎧
⎨ 0 w≤0
FW (w) = P [X1 ≤ w] P [X2 ≤ w] = [FX (w)] =
2
w /16 0 ≤ w ≤ 2
4 (7)
⎩
1 w≥2

X and Y are independent random variables with PDFs
2
2x 0 ≤ x ≤ 1 3y 0 ≤ y ≤ 1
fX (x) = fY (y) = (1)
For the event A = {X > Y }, this problem asks us to calculate the conditional expectations E[X|A]
and E[Y |A]. We will do this using the conditional joint PDF fX,Y |A (x, y). Since X and Y are
independent, it is tempting to argue that the event X > Y does not alter the probability model
for X and Y . Unfortunately, this is not the case. When we learn that X > Y , it increases the
probability that X is large and Y is small. We will see this when we compare the conditional
expectations E[X|A] and E[Y |A] to E[X] and E[Y ].
(a) We can calculate the unconditional expectations, E[X] and E[Y ], using the marginal PDFs
fX (x) and fY (y).
' ∞ ' 1
E [X] = fX (x) dx = 2x2 dx = 2/3 (2)
−∞ 0
' ∞ ' 1
E [Y ] = fY (y) dy = 3y 3 dy = 3/4 (3)
−∞ 0
(b) First, we need to calculate the conditional joint PDF ipdf X, Y |Ax, y. The first step is to
write down the joint PDF of X and Y :

6xy 2 0 ≤ x ≤ 1, 0 ≤ y ≤ 1
fX,Y (x, y) = fX (x) fY (y) = (4)
0 otherwise
181
The event A has probability
Y
''
1 P [A] = fX,Y (x, y) dy dx (5)
x>y
' 1' x
X>Y
= 6xy 2 dy dx (6)
0 0
X ' 1
1 = 2x4 dx = 2/5 (7)
0
Y
The conditional joint PDF of X and Y given A is
1
fX,Y (x,y)
P [A] (x, y) ∈ A
fX,Y |A (x, y) = (8)
0 otherwise

15xy 2 0 ≤ y ≤ x ≤ 1
X = (9)
1 0 otherwise
The triangular region of nonzero probability is a signal that given A, X and Y are no longer
independent. The conditional expected value of X given A is
' ∞' ∞
E [X|A] = xfX,Y |A (x, y|a) x, y dy dx (10)
−∞ −∞
' 1 ' x
2
= 15 x y 2 dy dx (11)
0 0
' 1
=5 x5 dx = 5/6. (12)
0
The conditional expected value of Y given A is

' ∞' ∞
E [Y |A] = yfX,Y |A (x, y) dy dx (13)
−∞ −∞
' 1 ' x
= 15 x y 3 dy dx (14)
0 0
' 1
15
= x5 dx = 5/8. (15)
4 0
We see that E[X|A] > E[X] while E[Y |A] < E[Y ]. That is, learning X > Y gives us a clue
that X may be larger than usual while Y may be smaller than usual.

This problem is quite straightforward. From Theorem 4.4, we can find the joint PDF of X and Y
is
∂ 2 [FX (x) FY (y)] ∂[fX (x) FY (y)]
fX,Y (x, y) = = = fX (x) fY (y) (1)
∂x ∂y ∂y
Hence, FX,Y (x, y) = FX (x)FY (y) implies that X and Y are independent.
If X and Y are independent, then
fX,Y (x, y) = fX (x) fY (y) (2)
182
By Definition 4.3,
' x ' y
−∞
' x
−∞
' y
= fX (u) du fY (v) dv (4)
−∞ −∞
= FX (x) FX (x) (5)


λ2 e−λy 0 ≤ x ≤ y
fX,Y (x, y) = (1)
0 otherwise
For W = Y − X we can find fW (w) by integrating over the region indicated in the figure below to
get FW (w) then taking the derivative with respect to w. Since Y ≥ X, W = Y − X is nonnegative.
Hence FW (w) = 0 for w < 0. For w ≥ 0,
Y
FW (w) = 1 − P [W > w] = 1 − P [Y > X + w] (2)

' ∞' ∞
=1− λ2 e−λy dy dx (3)
w 0 x+w
X<Y<X+w
−λw
X =1−e (4)
The complete expressions for the joint CDF and corresponding joint PDF are

0 w<0 0 w<0
FW (w) = fW (w) = (5)
1 − e−λw w ≥ 0 λe−λw w ≥ 0
(a) To find if W and X are independent, we must be able to factor the joint density function
fX,W (x, w) into the product fX (x)fW (w) of marginal density functions. To verify this, we
must find the joint PDF of X and W . First we find the joint CDF.
FX,W (x, w) = P [X ≤ x, W ≤ w] (1)
= P [X ≤ x, Y − X ≤ w] = P [X ≤ x, Y ≤ X + w] (2)
Since Y ≥ X, the CDF of W satisfies FX,W (x, w) = P [X ≤ x, X ≤ Y ≤ X + w]. Thus, for
x ≥ 0 and w ≥ 0,
' x ' x +w
Y FX,W (x, w) = λ2 e−λy dy dx (3)
x
{X<x}∩{X<Y<X+w} '0 x $x +w
−λy $
= −λe $ dx (4)
x
' x
0

w = −λe−λ(x +w) + λe−λx dx (5)
X
0

$
$x
x = e−λ(x +w) − e−λx $ (6)
0
= (1 − e−λx )(1 − e −λw
) (7)
183
We see that FX,W (x, w) = FX (x)FW (w). Moreover, by applying Theorem 4.4,
∂ 2 FX,W (x, w)
fX,W (x, w) = = λe−λx λe−λw = fX (x) fW (w) . (8)
∂x ∂w
Since we have our desired factorization, W and X are independent.
(b) Following the same procedure, we find the joint CDF of Y and W .
FW,Y (w, y) = P [W ≤ w, Y ≤ y] = P [Y − X ≤ w, Y ≤ y] (9)
= P [Y ≤ X + w, Y ≤ y] . (10)
The region of integration corresponding to the event {Y ≤ x + w, Y ≤ y} depends on whether
y < w or y ≥ w. Keep in mind that although W = Y − X ≤ Y , the dummy arguments y and
w of fW,Y (w, y) need not obey the same constraints. In any case, we must consider each case
separately.
For y > w, the integration is
' y−w ' u+w
Y
FW,Y (w, y) = λ2 e−λv dv du
{Y<y}Ç{Y<X+w} 0 u
' y ' y
y + λ2 e−λv dv du (11)
y−w u
w ' y−w
X =λ e−λu − e−λ(u+w) du
y-w y 0
' y
+λ e−λu − e−λy du (12)
y−w
It follows that
$y−w $y
$ $
FW,Y (w, y) = −e−λu + e−λ(u+w) $ + −e−λu − uλe−λy $ (13)
0 y−w
= 1 − e−λw − λwe−λy . (14)
For y ≤ w,
Y ' y ' y
FW,Y (w, y) = λ2 e−λv dv du (15)
{Y<y} '0
y
u
w = −λe−λy + λe−λu du (16)
y 0
$y
$
X = −λue−λy − e−λu $ (17)
0
= 1 − (1 + λy)e−λy (18)
The complete expression for the joint CDF is
⎧
⎨ 1 − e−λw − λwe−λy 0 ≤ w ≤ y
FW,Y (w, y) = 1 − (1 + λy)e−λy 0≤y≤w (19)
⎩
0 otherwise
Applying Theorem 4.4 yields

∂ 2 FW,Y (w, y) 2λ2 e−λy 0 ≤ w ≤ y
fW,Y (w, y) = = (20)
∂w ∂y 0 otherwise
The joint PDF fW,Y (w, y) doesn’t factor and thus W and Y are dependent.
184
We need to define the events A = {U ≤ u} and B = {V ≤ v}. In this case,
FU,V (u, v) = P [AB] = P [B] − P [Ac B] = P [V ≤ v] − P [U > u, V ≤ v] (1)
Note that U = min(X, Y ) > u if and only if X > u and Y > u. In the same way, since V =
max(X, Y ), V ≤ v if and only if X ≤ v and Y ≤ v. Thus
P [U > u, V ≤ v] = P [X > u, Y > u, X ≤ v, Y ≤ v] (2)
= P [u < X ≤ v, u < Y ≤ v] (3)
Thus, the joint CDF of U and V satisfies
FU,V (u, v) = P [V ≤ v] − P [U > u, V ≤ v] (4)
= P [X ≤ v, Y ≤ v] − P [u < X ≤ v, u < X ≤ v] (5)
Since X and Y are independent random variables,
FU,V (u, v) = P [X ≤ v] P [Y ≤ v] − P [u < X ≤ v] P [u < X ≤ v] (6)
= FX (v) FY (v) − (FX (v) − FX (u)) (FY (v) − FY (u)) (7)
= FX (v) FY (u) + FX (u) FY (v) − FX (u) FY (u) (8)
The joint PDF is
∂ 2 FU,V (u, v)
fU,V (u, v) = (9)
∂u∂v
∂
= [fX (v) FY (u) + FX (u) fY (v)] (10)
∂u
= fX (u) fY (v) + fX (v) fY (v) (11)

2 /8)−(y 2 /18)
fX,Y (x, y) = ce−(x (1)
The omission of any limits for the PDF indicates that it is defined over all x and y. We know that
fX,Y (x, y) is in the form of the bivariate Gaussian distribution so we look to Definition 4.17 and
attempt to find values for σY , σX , E[X], E[Y ] and ρ. First, we know that the constant is
1
c= (2)
2πσX σY 1 − ρ2
Because the exponent of fX,Y (x, y) doesn’t contain any cross terms we know that ρ must be zero,
and we are left to solve the following for E[X], E[Y ], σX , and σY :

x − E [X] 2 x2 y − E [Y ] 2 y2
= = (3)
σX 8 σY 18
From which we can conclude that
E [X] = E [Y ] = 0 (4)
√
σX = 8 (5)
√
σY = 18 (6)
1
Putting all the pieces together, we find that c = 24π . Since ρ = 0, we also find that X and Y are
independent.
185
For the joint PDF
2 −4xy+4y 2 )
fX,Y (x, y) = ce−(2x , (1)
we proceed as in Problem 4.11.1 to find values for σY , σX , E[X], E[Y ] and ρ.
(a) First, we try to solve the following equations

x − E [X] 2
= 4(1 − ρ2 )x2 (2)
σX

y − E [Y ] 2
= 8(1 − ρ2 )y 2 (3)
σY
2ρ
= 8(1 − ρ2 ) (4)
σX σY
The first two equations yield E[X] = E[Y ] = 0
(b) To find the correlation coefficient ρ, we observe that

σX = 1/ 4(1 − ρ2 ) σY = 1/ 8(1 − ρ2 ) (5)
√
Using σX and σY in the third equation yields ρ = 1/ 2.
√
(c) Since ρ = 1/ 2, now we can solve for σX and σY .
√
σX = 1/ 2 σY = 1/2 (6)
(d) From here we can solve for c.

1 2
c= = (7)
2πσX σY 1− ρ2 π
(e) X and Y are dependent because ρ = 0.

From the problem statement, we learn that
2
µX = µY = 0 σX = σY2 = 1 (1)
From Theorem 4.30, the conditional expectation of Y given X is

σY
E [Y |X] = µ̃Y (X) = µY + ρ (X − µX ) = ρX (2)
σX
In the problem statement, we learn that E[Y |X] = X/2. Hence ρ = 1/2. From Definition 4.17, the
joint PDF is
1 2 2
fX,Y (x, y) = √ e−2(x −xy+y )/3 (3)
3π 2
186
The event B is the set of outcomes satisfying X 2 + Y 2 ≤ 22 . Of ocurse, the calculation of P [B]
depends on the probability model for X and Y .
(a) In this instance, X and Y have the same PDF

0.01 −50 ≤ x ≤ 50
fX (x) = fY (x) = (1)
0 otherwise
Since X and Y are independent, their joint PDF is
−4
10 −50 ≤ x ≤ 50, −50 ≤ y ≤ 50
fX,Y (x, y) = fX (x) fY (y) = (2)
0 otherwise
Because X and Y have a uniform PDF over the bullseye area, P [B] is just the value of the
joint PDF over the area times the area of the bullseye.

P [B] = P X 2 + Y 2 ≤ 22 = 10−4 · π22 = 4π · 10−4 ≈ 0.0013 (3)
(b) In this case, the joint PDF of X and Y is inversely proportional to the area of the target.

1/[π502 ] x2 + y 2 ≤ 502
fX,Y (x, y) = (4)
0 otherwise
The probability of a bullseye is
2
π22 1
P [B] = P X 2 + Y 2 ≤ 22 = = ≈ 0.0016. (5)
π502 25
(c) In this instance, X and Y have the identical Gaussian (0, σ) PDF with σ 2 = 100; i.e.,
1 2 /2σ 2
fX (x) = fY (x) = √ e−x (6)
2πσ 2
Since X and Y are independent, their joint PDF is
1 −(x2 +y2 )/2σ2
fX,Y (x, y) = fX (x) fY (y) = e (7)
2πσ 2
To find P [B], we write
''

P [B] = P X + Y ≤ 2
2 2 2
= fX,Y (x, y) dx dy (8)
x2 +y 2 ≤22
''
1 2 +y 2 )/2σ 2
= e−(x dx dy (9)
2πσ 2 x2 +y 2 ≤22
This integral is easy using polar coordinates. With the substitutions x2 + y 2 = r2 , and
dx dy = r dr dθ,
' 2 ' 2π
1 2 2
P [B] = 2
e−r /2σ r dr dθ (10)
2πσ 0 0
' 2
1 2 2
= 2 re−r /2σ dr (11)
σ 0
2
$
2 $2
= −e−r /2σ $ = 1 − e−4/200 ≈ 0.0198. (12)
0
187
(a) The person’s temperature is high with probability
p = P [T > 38] = P [T − 37 > 38 − 37] = 1 − Φ(1) = 0.159. (1)
Given that the temperature is high, then W is measured. Since ρ = 0, W and T are inde-
pendent and

W −7 10 − 7
q = P [W > 10] = P > = 1 − Φ(1.5) = 0.067. (2)
2 2
q W >10

p T >38 W ≤10

1−q

T ≤38
1−p
The probability the person is ill is
P [I] = P [T > 38, W > 10] = P [T > 38] P [W > 10] = pq = 0.0107. (3)
(b) The general form of the bivariate Gaussian PDF is

⎡ 2 2 ⎤
2ρ(w−µ1 )(t−µ2 )
⎢
w−µ1
σ1 − σ1 σ2 + t−µ2
σ2 ⎥
exp ⎣− ⎦
2(1 − ρ )2
fW,T (w, t) = (4)

2πσ1 σ2 1 − ρ2
√
With µ1 = E[W ] = 7, σ1 = σW = 2, µ2 = E[T ] = 37 and σ2 = σT = 1 and ρ = 1/ 2, we
have 0 √ 1
1 (w − 7)2 2(w − 7)(t − 37)
fW,T (w, t) = √ exp − − + (t − 37)2
(5)
2π 2 4 2
To find the conditional probability P [I|T = t], we need to find the conditional PDF of W
given T = t. The direct way is simply to use algebra to find
fW,T (w, t)
fW |T (w|t) = (6)
fT (t)
The required algebra is essentially the same as that needed to prove Theorem 4.29. Its easier
just to apply Theorem 4.29 which says that given T = t, the conditional distribution of W is
Gaussian with
σW
E [W |T = t] = E [W ] + ρ (t − E [T ]) (7)
σT
Var[W |T = t] = σW2
(1 − ρ2 ) (8)
188
Plugging in the various parameters gives
√
E [W |T = t] = 7 + 2(t − 37) and Var [W |T = t] = 2 (9)
Using this conditional mean and variance, we obtain the conditional Gaussian PDF
1 √ 2
fW |T (w|t) = √ e−(w−(7+ 2(t−37))) /4 . (10)
4π
Given T = t, the conditional probability the person is declared ill is
P [I|T = t] = P [W > 10|T = t] (11)

0 √ √ 1
W − (7 + 2(t − 37)) 10 − (7 + 2(t − 37))
=P √ > √ (12)
2 2
0 √ 1 ! √ "
3 − 2(t − 37) 3 2
=P Z> √ =Q − (t − 37) . (13)
2 2

The given joint PDF is
2 x2 +bxy+c2 y 2 )
fX,Y (x, y) = de−(a (1)
In order to be an example of the bivariate Gaussian PDF given in Definition 4.17, we must have
1 1
a2 = c2 =
2σX −
2 (1 ρ2 ) − ρ2 )
2σY2 (1
−ρ 1
b= d=
σX σY (1 − ρ2 ) 2πσX σY 1 − ρ2
We can solve for σX and σY , yielding

1 1
σX = σY = (2)
a 2(1 − ρ2 ) c 2(1 − ρ2 )
Plugging these values into the equation for b, it follows that b = −2acρ, or, equivalently, ρ = −b/2ac.
This implies
1
d2 = 2 2 2 = (1 − ρ2 )a2 c2 = a2 c2 − b2 /4 (3)
4π σX σY (1 − ρ2 )
Since |ρ| ≤ 1,
we see that |b| ≤ 2ac. Further, for any choice of a, b and c that meets this constraint,
choosing d = a2 c2 − b2 /4 yields a valid PDF.

From Equation (4.146), we can write the bivariate Gaussian PDF as
1 2 2 1 2 2
fX,Y (x, y) = √ e−(x−µX ) /2σX
√ e−(y−µ̃Y (x)) /2σ̃Y
(1)
σX 2π σ̃Y 2π

where µ̃Y (x) = µY + ρ σσX
Y
(x − µX ) and σ̃Y = σY 1 − ρ2 . However, the definitions of µ̃Y (x) and
σ̃Y are not particularly important for this exercise. When we integrate the joint PDF over all x
189
and y, we obtain
' ∞' ∞ ' ∞ ' ∞
1 −(x−µX )2 /2σX
2 1 2 2
fX,Y (x, y) dx dy = √ e √ e−(y−µ̃Y (x)) /2σ̃Y
dy dx (2)
−∞ −∞ −∞ σX 2π −∞ σ̃Y 2π

1
' ∞
1 2 2
= √ e−(x−µX ) /2σX
dx (3)
−∞ σX 2π
The marked integral equals 1 because for each value of x, it is the integral of a Gaussian PDF of
one variable over all possible values. In fact, it is the integral of the conditional PDF fY |X (y|x)
over all possible y. To complete the proof, we see that
' ∞' ∞ ' ∞
1 2 2
fX,Y (x, y) dx dy = √ e−(x−µX ) /2σX dx = 1 (4)
−∞ −∞ −∞ σX 2π
since the remaining integral is the integral of the marginal Gaussian PDF fX (x) over all possible
x.

In this problem, X1 and X2 are jointly Gaussian random variables with E[Xi ] = µi , Var[Xi ] = σi2 ,
and correlation coefficient ρ12 = ρ. The goal is to show that Y = X1 X2 has variance
Var[Y ] = (1 + ρ2 )σ12 σ22 + µ21 σ22 + µ22 σ12 + 2ρµ1 µ2 σ1 σ2 . (1)
Since Var[Y ] = E[Y 2 ] − (E[Y ])2 , we will find the moments of Y . The first moment is
E [Y ] = E [X1 X2 ] = Cov [X1 , X2 ] + E [X1 ] E [X2 ] = ρσ1 σ2 + µ1 µ2 . (2)
For the second moment of Y , we follow the problem hint and use the iterated expectation

E Y 2 = E X12 X22 = E E X12 X22 |X2 = E X22 E X12 |X2 . (3)
Given X2 = x2 , we observe from Theorem 4.30 that X1 is is Gaussian with

σ1
E [X1 |X2 = x2 ] = µ1 + ρ (x2 − µ2 ), Var[X1 |X2 = x2 ] = σ12 (1 − ρ2 ). (4)
σ2
Thus, the conditional second moment of X1 is

E X12 |X2 = (E [X1 |X2 ])2 + Var[X1 |X2 ] (5)
2
σ1
= µ1 + ρ (X2 − µ2 ) + σ12 (1 − ρ2 ) (6)
σ2
σ1 σ2
= [µ21 + σ12 (1 − ρ2 )] + 2ρµ1 (X2 − µ2 ) + ρ2 12 (X2 − µ2 )2 . (7)
σ2 σ2
It follows that

E X12 X22 = E X22 E X12 |X22 (8)
2

σ1 2 σ1
= E [µ1 + σ1 (1 − ρ )]X2 + 2ρµ1 (X2 − µ2 )X2 + ρ 2 (X2 − µ2 ) X2 .
2 2 2 2 2 2 2
(9)
σ2 σ2
190
Since E[X22 ] = σ22 + µ22 ,

E X12 X22 = µ21 + σ12 (1 − ρ2 ) (σ22 + µ22 )
σ1 σ2
+ 2ρµ1 E (X2 − µ2 )X22 + ρ2 12 E (X2 − µ2 )2 X22 . (10)
σ2 σ2
We observe that

E (X2 − µ2 )X22 = E (X2 − µ2 )(X2 − µ2 + µ2 )2 (11)

= E (X2 − µ2 ) (X2 − µ2 )2 + 2µ2 (X2 − µ2 ) + µ22 (12)

= E (X2 − µ2 )3 + 2µ2 E (X2 − µ2 )2 + µ2 E [(X2 − µ2 )] (13)
We recall that E[X2 − µ2 ] = 0 and that E[(X2 − µ2 )2 ] = σ22 . We now look ahead to Problem 6.3.4
to learn that
E (X2 − µ2 )3 = 0, E (X2 − µ2 )4 = 3σ24 . (14)
This implies

E (X2 − µ2 )X22 = 2µ2 σ22 . (15)
Following this same approach, we write

E (X2 − µ2 )2 X22 = E (X2 − µ2 )2 (X2 − µ2 + µ2 )2 (16)

= E (X2 − µ2 )2 (X2 − µ2 )2 + 2µ2 (X2 − µ2 ) + µ22 (17)

= E (X2 − µ2 )2 (X2 − µ2 )2 + 2µ2 (X2 − µ2 ) + µ22 (18)

= E (X2 − µ2 )4 + 2µ2 E X2 − µ2 )3 + µ22 E (X2 − µ2 )2 . (19)
It follows that

E (X2 − µ2 )2 X22 = 3σ24 + µ22 σ22 . (20)
Combining the above results, we can conclude that

σ1 σ2
E X12 X22 = µ21 + σ12 (1 − ρ2 ) (σ22 + µ22 ) + 2ρµ1 (2µ2 σ22 ) + ρ2 12 (3σ24 + µ22 σ22 ) (21)
σ2 σ2
= (1 + 2ρ2 )σ12 σ22 + µ22 σ12 + µ21 σ22 + µ21 µ22 + 4ρµ1 µ2 σ1 σ2 . (22)
Finally, combining Equations (2) and (22) yields

Var[Y ] = E X12 X22 − (E [X1 X2 ])2 (23)
= (1 + ρ2 )σ12 σ22 + µ21 σ22 + µ22 σ12 + 2ρµ1 µ2 σ1 σ2 . (24)

The script imagepmf in Example 4.27 generates the grid variables SX, SY, and PXY. Recall that
for each entry in the grid, SX. SY and PXY are the corresponding values of x, y and PX,Y (x, y).
Displaying them as adjacent column vectors forms the list of all possible pairs x, y and the proba-
bilities PX,Y (x, y). Since any Matlab vector or matrix x is reduced to a column vector with the
command x(:), the following simple commands will generate the list:
191
>> format rat;
>> imagepmf;
>> [SX(:) SY(:) PXY(:)]
ans =
800 400 1/5
1200 400 1/20
1600 400 0
800 800 1/20
1200 800 1/5
1600 800 1/10
800 1200 1/10
1200 1200 1/10
1600 1200 1/5
>>
Note that the command format rat wasn’t necessary; it just formats the output as rational num-
bers, i.e., ratios of integers, which you may or may not find esthetically pleasing.

In this problem, we need to calculate E[X], E[Y ], the correlation E[XY ], and the covariance
Cov[X, Y ] for random variables X and Y in Example 4.27. In this case, we can use the script
imagepmf.m in Example 4.27 to generate the grid variables SX, SY and PXY that describe the joint
PMF PX,Y (x, y).
However, for the rest of the problem, a general solution is better than a specific solution. The
general problem is that given a pair of finite random variables described by the grid variables SX,
SY and PXY, we want Matlab to calculate an expected value E[g(X, Y )]. This problem is solved
in a few simple steps. First we write a function that calculates the expected value of any finite
random variable.
function ex=finiteexp(sx,px);
%Usage: ex=finiteexp(sx,px)
%returns the expected value E[X]
%of finite random variable X described
%by samples sx and probabilities px
ex=sum((sx(:)).*(px(:)));
Note that finiteexp performs its calculations on the sample values sx and probabilities px using
the column vectors sx(:) and px(:). As a result, we can use the same finiteexp function when
the random variable is represented by grid variables. For example, we can calculate the correlation
r = E[XY ] as
r=finiteexp(SX.*SY,PXY)
It is also convenient to define a function that returns the covariance:
function covxy=finitecov(SX,SY,PXY);
%Usage: cxy=finitecov(SX,SY,PXY)
%returns the covariance of
%finite random variables X and Y
%given by grids SX, SY, and PXY
ex=finiteexp(SX,PXY);
ey=finiteexp(SY,PXY);
R=finiteexp(SX.*SY,PXY);
covxy=R-ex*ey;
192
The following script calculates the desired quantities:
%imageavg.m >> imageavg
%Solution for Problem 4.12.2 ex =
imagepmf; %defines SX, SY, PXY 1180
ex=finiteexp(SX,PXY) ey =
ey=finiteexp(SY,PXY) 860
rxy=finiteexp(SX.*SY,PXY) rxy =
cxy=finitecov(SX,SY,PXY) 1064000
cxy =
49200
>>
The careful reader will observe that imagepmf is inefficiently coded in that the correlation E[XY ]
is calculated twice, once directly and once inside of finitecov. For more complex problems, it
would be worthwhile to avoid this duplication.

The script is just a Matlab calculation of FX,Y (x, y) in Equation (4.29).
%trianglecdfplot.m
[X,Y]=meshgrid(0:0.05:1.5);
R=(0<=Y).*(Y<=X).*(X<=1).*(2*(X.*Y)-(Y.^2));
R=R+((0<=X).*(X<Y).*(X<=1).*(X.^2));
R=R+((0<=Y).*(Y<=1).*(1<X).*((2*Y)-(Y.^2)));
R=R+((X>1).*(Y>1));
mesh(X,Y,R);
xlabel(’\it x’);
ylabel(’\it y’);
For functions like FX,Y (x, y) that have multiple cases, we calculate the function for each case and
multiply by the corresponding boolean condition so as to have a zero contribution when that case
doesn’t apply. Using this technique, its important to define the boundary conditions carefully to
make sure that no point is included in two different boundary conditions.

By following the formulation of Problem 4.2.6, the code to set up the sample grid is reasonably
straightforward:
function [SX,SY,PXY]=circuits(n,p);
%Usage: [SX,SY,PXY]=circuits(n,p);
% (See Problem 4.12.4)
[SX,SY]=ndgrid(0:n,0:n);
PXY=0*SX;
PXY(find((SX==n) & (SY==n)))=p^n;
for y=0:(n-1),
I=find((SY==y) &(SX>=SY) &(SX<n));
PXY(I)=(p^y)*(1-p)* ...
binomialpmf(n-y-1,p,SX(I)-y);
end;
The only catch is that for a given value of y, we need to calculate the binomial probability of x − y
successes in (n − y − 1) trials. We can do this using the function call
binomialpmf(n-y-1,p,x-y)
193
However, this function expects the argument n-y-1 to be a scalar. As a result, we must perform a
separate call to binomialpmf for each value of y.
An alternate solution is direct calculation of the PMF PX,Y (x, y) in Problem 4.2.6. In this
case, we calculate m! using the Matlab function gamma(m+1). Because, gamma(x) function will
calculate the gamma function for each element in a vector x, we can calculate the PMF without
any loops:
function [SX,SY,PXY]=circuits2(n,p);
%Usage: [SX,SY,PXY]=circuits2(n,p);
% (See Problem 4.12.4)
[SX,SY]=ndgrid(0:n,0:n);
PXY=0*SX;
PXY(find((SX==n) & (SY==n)))=p^n;
I=find((SY<=SX) &(SX<n));
PXY(I)=(gamma(n-SY(I))./(gamma(SX(I)-SY(I)+1)...
.*gamma(n-SX(I)))).*(p.^SX(I)).*((1-p).^(n-SX(I)));
Some experimentation with cputime will show that circuits2(n,p) runs much faster than circuits(n,p).
As is typical, the for loop in circuit results in time wasted running the Matlab interpretor and
in regenerating the binomial PMF in each cycle.
To finish the problem, we need to calculate the correlation coefficient
Cov [X, Y ]
ρX,Y = . (1)
σX σY
In fact, this is one of those problems where a general solution is better than a specific solution.
The general problem is that given a pair of finite random variables described by the grid variables
SX, SY and PMF PXY, we wish to calculate the correlation coefficient
This problem is solved in a few simple steps. First we write a function that calculates the
expected value of a finite random variable.
function ex=finiteexp(sx,px);
%Usage: ex=finiteexp(sx,px)
%returns the expected value E[X]
%of finite random variable X described
%by samples sx and probabilities px
ex=sum((sx(:)).*(px(:)));
Note that finiteexp performs its calculations on the sample values sx and probabilities px using
the column vectors sx(:) and px(:). As a result, we can use the same finiteexp function when
the random variable is represented by grid variables. We can build on finiteexp to calculate the
variance using finitevar:
function v=finitevar(sx,px);
%Usage: ex=finitevar(sx,px)
% returns the variance Var[X]
% of finite random variables X described by
% samples sx and probabilities px
ex2=finiteexp(sx.^2,px);
ex=finiteexp(sx,px);
v=ex2-(ex^2);
Putting these pieces together, we can calculate the correlation coefficient.
194
function rho=finitecoeff(SX,SY,PXY);
%Usage: rho=finitecoeff(SX,SY,PXY)
%Calculate the correlation coefficient rho of
%finite random variables X and Y
ex=finiteexp(SX,PXY); vx=finitevar(SX,PXY);
ey=finiteexp(SY,PXY); vy=finitevar(SY,PXY);
R=finiteexp(SX.*SY,PXY);
rho=(R-ex*ey)/sqrt(vx*vy);
Calculating the correlation coefficient of X and Y , is now a two line exercise..

>> [SX,SY,PXY]=circuits2(50,0.9);
>> rho=finitecoeff(SX,SY,PXY)
rho =
0.4451
>>

In the first approach X is an exponential (λ) random variable, Y is an independent exponential
(µ) random variable, and W = Y /X. we implement this approach in the function wrv1.m shown
below.
In the second approach, we use Theorem 3.22 and generate samples of a uniform (0, 1) random
−1
variable U and calculate W = FW (U ). In this problem,
λ/µ
FW (w) = 1 − . (1)
λ/µ + w
Setting u = FW (w) and solving for w yields

−1 λ u
w= FW (u) = (2)
µ 1−u
We implement this solution in the function wrv2. Here are the two solutions:
function w=wrv1(lambda,mu,m) function w=wrv2(lambda,mu,m)
%Usage: w=wrv1(lambda,mu,m) %Usage: w=wrv1(lambda,mu,m)
%Return m samples of W=Y/X %Return m samples of W=Y/X
%X is exponential (lambda) %X is exponential (lambda)
%Y is exponential (mu) %Y is exponential (mu)
%Uses CDF of F_W(w)
x=exponentialrv(lambda,m);
y=exponentialrv(mu,m); u=rand(m,1);
w=y./x; w=(lambda/mu)*u./(1-u);
We would expect that wrv2 would be faster simply because it does less work. In fact, its
instructive to account for the work each program does.
• wrv1 Each exponential random sample requires the generation of a uniform random variable,
and the calculation of a logarithm. Thus, we generate 2m uniform random variables, calculate
2m logarithms, and perform m floating point divisions.
• wrv2 Generate m uniform random variables and perform m floating points divisions.
This quickie analysis indicates that wrv1 executes roughly 5m operations while wrv2 executes about
2m operations. We might guess that wrv2 would be faster by a factor of 2.5. Experimentally, we
calculated the execution time associated with generating a million samples:
195
>> t2=cputime;w2=wrv2(1,1,1000000);t2=cputime-t2
t2 =
0.2500
>> t1=cputime;w1=wrv1(1,1,1000000);t1=cputime-t1
t1 =
0.7610
>>
We see in our simple experiments that wrv2 is faster by a rough factor of 3. (Note that repeating
such trials yielded qualitatively similar results.)
196

The repair of each laptop can be viewed as an independent trial with four possible outcomes
corresponding to the four types of needed repairs.
(a) Since the four types of repairs are mutually exclusive choices and since 4 laptops are returned
for repair, the joint distribution of N1 , . . . , N4 is the multinomial PMF

4
PN1 ,...,N4 (n1 , . . . , n4 ) = pn1 pn2 pn3 pn4 (1)
n 1 , n2 , n3 , n4 1 2 3 4
8 n1 4 n2 2 n3 1 n4
n 1 !n2
4!
!n3 !n4 ! 15 15 15 15 n1 + · · · + n4 = 4; ni ≥ 0
=
0 otherwise
(2)
(b) Let L2 denote the event that exactly two laptops need LCD repairs. Thus P [L2 ] = PN1 (2).
Since each laptop requires an LCD repair with probability p1 = 8/15, the number of LCD
repairs, N1 , is a binomial (4, 8/15) random variable with PMF

4
PN1 (n1 ) = (8/15)n1 (7/15)4−n1 (3)
n1
The probability that two laptops need LCD repairs is

4
PN1 (2) = (8/15)2 (7/15)2 = 0.3717 (4)
2
(c) A repair is type (2) with probability p2 = 4/15. A repair is type (3) with probability p3 =
2/15; otherwise a repair is type “other” with probability po = 9/15. Define X as the number
of “other” repairs needed. The joint PMF of X, N2 , N3 is the multinomial PMF
n2 n3 x
4 4 2 9
PN2 ,N3 ,X (n2 , n3 , x) = (5)
n 2 , n3 , x 15 15 15
However, Since X + 4 − N2 − N3 , we observe that
PN2 ,N3 (n2 , n3 ) = PN2 ,N3 ,X (n2 , n3 , 4 − n2 − n3 ) (6)

n2 n3 4−n2 −n3
4 4 2 9
= (7)
n 2 , n3 , 4 − n 2 − n 3 15 15 15
4 n2 n3
9 4 4 2
= (8)
15 n 2 , n3 , 4 − n 2 − n 3 9 9
Similarly, since each repair is a motherboard repair with probability p2 = 4/15, the number
of motherboard repairs has binomial PMF
n2 4−n2
4 4 11
PN2 (n2 ) n2 = (9)
n2 15 15
197
Finally, the probability that more laptops require motherboard repairs than keyboard repairs
is
P [N2 > N3 ] = PN2 ,N3 (1, 0) + PN2 ,N3 (2, 0) + PN2 ,N3 (2, 1) + PN2 (3) + PN2 (4) (10)
where we use the fact that if N2 = 3 or N2 = 4, then we must have N2 > N3 . Inserting the
various probabilities, we obtain
P [N2 > N3 ] = PN2 ,N3 (1, 0) + PN2 ,N3 (2, 0) + PN2 ,N3 (2, 1) + PN2 (3) + PN2 (4) (11)
Plugging in the various probabilities yields P [N2 > N3 ] = 8,656/16,875 ≈ 0.5129.

Whether a computer has feature i is a Bernoulli trial with success probability pi = 2−i . Given that
n computers were sold, the number of computers sold with feature i has the binomial PMF
n ni
ni pi (1 − pi )
ni n = 0, 1, . . . , n
i
PNi (ni ) = (1)
0 otherwise
Since a computer has feature i with probability pi independent of whether any other feature is
on the computer, the number Ni of computers with feature i is independent of the number of
computers with any other features. That is, N1 , . . . , N4 are mutually independent and have joint
PMF
PN1 ,...,N4 (n1 , . . . , n4 ) = PN1 (n1 ) PN2 (n2 ) PN3 (n3 ) PN4 (n4 ) (2)
(a) In terms of the joint PDF, we can write joint CDF as

' x1 ' xn
FX1 ,...,Xn (x1 , . . . , xn ) = ··· fX1 ,...,Xn (y1 , . . . , yn ) dy1 · · · dyn (1)
−∞ −∞
However, simplifying the above integral depends on the values of each xi . In particular,
fX1 ,...,Xn (y1 , . . . , yn ) = 1 if and only if 0 ≤ yi ≤ 1 for each i. Since FX1 ,...,Xn (x1 , . . . , xn ) = 0
if any xi < 0, we limit, for the moment, our attention to the case where xi ≥ 0 for all i. In
this case, some thought will show that we can write the limits in the following way:
' max(1,x1 ) ' min(1,xn )
FX1 ,...,Xn (x1 , . . . , xn ) = ··· dy1 · · · dyn (2)
0 0
= min(1, x1 ) min(1, x2 ) · · · min(1, xn ) (3)
A complete expression for the CDF of X1 , . . . , Xn is

2n
FX1 ,...,Xn (x1 , . . . , xn ) = i=1 min(1, xi ) 0 ≤ xi , i = 1, 2, . . . , n (4)
0 otherwise
198
(b) For n = 3,

1 − P min Xi ≤ 3/4 = P min Xi > 3/4 (5)
i i
= P [X1 > 3/4, X2 > 3/4, X3 > 3/4] (6)

' 1 ' 1 ' 1
= dx1 dx2 dx3 (7)
3/4 3/4 3/4
= (1 − 3/4)3 = 1/64 (8)
Thus P [mini Xi ≤ 3/4] = 63/64.

This problem is very simple. In terms of the vector X, the PDF is

1 0≤x≤1
fX (x) = (1)
0 otherwise
However, just keep in mind that the inequalities 0 ≤ x and x ≤ 1 are vector inequalities that must
hold for every component xi .

In this problem, we find the constant c from ( ∞ the (requirement that that the integral of the vector
∞
PDF

n over all possible values is 1. That is, −∞ · · · −∞ Xf (x) dx1 · · · dxn = 1. Since fX (x) = ca x =
c i=1 ai xi , we have that
' ∞ ' ∞ ' 1 ' 1 ! n
"
··· fX (x) dx1 · · · dxn = c ··· ai xi dx1 · · · dxn (1)
−∞ −∞ 0 0 i=1
n ' 1
' 1
=c ··· ai xi dx1 · · · dxn (2)
i=1 0 0
n ' 1 ' 1 ' 1

=c ai dx1 · · · xi dxi · · · dxn (3)
i=1 0 0 0
! $1 "
n
x2i $$
n
ai
=c ai $ = c (4)
2 0 2
i=1 i=1
The requirement that the PDF integrate to unity thus implies

2
c =
n (5)
i=1 ai

Here we solve the following problem:1
Given fX (x) with c = 2/3 and a1 = a2 = a3 = 1 in Problem 5.2.2, find the marginal
PDF fX3 (x3 ).
1
The wrong problem statement appears in the first printing.
199
Filling in the parameters in Problem 5.2.2, we obtain the vector PDF
2
fX (x) = 3 (x1 + x2 + x3 ) 0 ≤ x1 , x2 , x3 ≤ 1 (1)
0 otherwise
In this case, for 0 ≤ x3 ≤ 1, the marginal PDF of X3 is

' '
2 1 1
fX3 (x3 ) = (x1 + x2 + x3 ) dx1 dx2 (2)
3 0 0
' $x1 =1
2 1 x21 $
= + x2 x1 + x3 x1 $$ dx2 (3)
3 0 2 x1 =0
'
2 1 1
= + x2 + x3 dx2 (4)
3 0 2
$x2 =1
2 x2 x22 $ 2 1 1
= + + x3 x2 $$ = + + x3 (5)
3 2 2 x2 =0 3 2 2
The complete expresion for the marginal PDF of X3 is

2(1 + x3 )/3 0 ≤ x3 ≤ 1,
fX3 (x3 ) = (6)
0 otherwise.

Since J1 , J2 and J3 are independent, we can write
PK (k) = PJ1 (k1 ) PJ2 (k2 − k1 ) PJ3 (k3 − k2 ) (1)
Since PJi (j) > 0 only for integers j > 0, we have that PK (k) > 0 only for 0 < k1 < k2 < k3 ;
otherwise PK (k) = 0. Finally, for 0 < k1 < k2 < k3 ,
PK (k) = (1 − p)k1 −1 p(1 − p)k2 −k1 −1 p(1 − p)k3 −k2 −1 p (2)

k3 −3 3
= (1 − p) p (3)

The joint PMF is

p3 (1 − p)k3 −3 1 ≤ k1 < k2 < k3
PK (k) = PK1 ,K2 ,K3 (k1 , k2 , k3 ) = (1)
0 otherwise
(a) We start by finding PK1 ,K2 (k1 , k2 ). For 1 ≤ k1 < k2 ,

∞

PK1 ,K2 (k1 , k2 ) = PK1 ,K2 ,K3 (k1 , k2 , k3 ) (2)
k3 =−∞
∞
= p3 (1 − p)k3 −3 (3)
k3 =k2 +1

= p (1 − p)k2 −2 1 + (1 − p) + (1 − p)2 + · · ·
3
(4)
= p2 (1 − p)k2 −2 (5)
200

p2 (1 − p)k2 −2 1 ≤ k1 < k2
PK1 ,K2 (k1 , k2 ) = (6)
0 otherwise
Next we find PK1 ,K3 (k1 , k3 ). For k1 ≥ 1 and k3 ≥ k1 + 2, we have
∞
3 −1
k
PK1 ,K3 (k1 , k3 ) = PK1 ,K2 ,K3 (k1 , k2 , k3 ) = p3 (1 − p)k3 −3 (7)
k2 =−∞ k2 =k1 +1
= (k3 − k1 − 1)p3 (1 − p)k3 −3 (8)

The complete expression of the PMF of K1 and K3 is

(k3 − k1 − 1)p3 (1 − p)k3 −3 1 ≤ k1 , k1 + 2 ≤ k3 ,
PK1 ,K3 (k1 , k3 ) = (9)
0 otherwise.
The next marginal PMF is
∞
2 −1
k
PK2 ,K3 (k2 , k3 ) = PK1 ,K2 ,K3 (k1 , k2 , k3 ) = p3 (1 − p)k3 −3 (10)
k1 =−∞ k1 =1
= (k2 − 1)p3 (1 − p)k3 −3 (11)

The complete expression of the PMF of K2 and K3 is

(k2 − 1)p3 (1 − p)k3 −3 1 ≤ k2 < k3 ,
PK2 ,K3 (k2 , k3 ) = (12)
0 otherwise.
(b) Going back to first principles, we note that Kn is the number of trials up to and including
the nth success. Thus K1 is a geometric (p) random variable, K2 is an Pascal (2, p) random
variable, and K3 is an Pascal (3, p) random variable. We could write down the respective
marginal PMFs of K1 , K2 and K3 just by looking up the Pascal (n, p) PMF. Nevertheless, it
is instructive to derive these PMFs from the joint PMF PK1 ,K2 ,K3 (k1 , k2 , k3 ).
For k1 ≥ 1, we can find PK1 (k1 ) via
∞
∞

PK1 (k1 ) = PK1 ,K2 (k1 , k2 ) = p2 (1 − p)k2 −2 (13)
k2 =−∞ k2 =k1 +1
= p (1 − p)k1 −1 [1 + (1 − p) + (1 − p)2 + · · · ]
2
(14)
k1 −1
= p(1 − p) (15)
The complete expression for the PMF of K1 is the usual geometric PMF

p(1 − p)k1 −1 k1 = 1, 2, . . . ,
PK1 (k1 ) = (16)
0 otherwise.
Following the same procedure, the marginal PMF of K2 is
∞
2 −1
k
PK2 (k2 ) = PK1 ,K2 (k1 , k2 ) = p2 (1 − p)k2 −2 (17)
k1 =−∞ k1 =1
= (k2 − 1)p2 (1 − p)k2 −2 (18)
201
Since PK2 (k2 ) = 0 for k2 < 2, the complete PMF is the Pascal (2, p) PMF

k2 − 1 2
PK2 (k2 ) = p (1 − p)k2 −2 (19)
1
Finally, for k3 ≥ 3, the PMF of K3 is

∞
3 −1
k
PK3 (k3 ) = PK2 ,K3 (k2 , k3 ) = (k2 − 1)p3 (1 − p)k3 −3 (20)
k2 =−∞ k2 =2
= [1 + 2 + · · · + (k3 − 2)]p3 (1 − p)k3 −3 (21)

(k3 − 2)(k3 − 1) 3
= p (1 − p)k3 −3 (22)
2
Since PK3 (k3 ) = 0 for k3 < 3, the complete expression for PK3 (k3 ) is the Pascal (3, p) PMF

k3 − 1 3
PK3 (k3 ) = p (1 − p)k3 −3 . (23)
2

For 0 ≤ y1 ≤ y4 ≤ 1, the marginal PDF of Y1 and Y4 satisfies
''
fY1 ,Y4 (y1 , y4 ) = fY (y) dy2 dy3 (1)
' y4 ' y4
= 24 dy3 dy2 (2)
y1 y2
' y4
= 24(y4 − y2 ) dy2 (3)
y1
$y =y
= −12(y4 − y2 )2 $y22 =y41 = 12(y4 − y1 )2 (4)
The complete expression for the joint PDF of Y1 and Y4 is

12(y4 − y1 )2 0 ≤ y1 ≤ y4 ≤ 1
fY1 ,Y4 (y1 , y4 ) = (5)
0 otherwise
For 0 ≤ y1 ≤ y2 ≤ 1, the marginal PDF of Y1 and Y2 is

''
fY1 ,Y2 (y1 , y2 ) = fY (y) dy3 dy4 (6)
' 1 ' 1
= 24 dy4 dy3 (7)
y2 y3
' 1
= 24(1 − y3 ) dy3 = 12(1 − y2 )2 (8)
y2
The complete expression for the joint PDF of Y1 and Y2 is

12(1 − y2 )2 0 ≤ y1 ≤ y2 ≤ 1
fY1 ,Y2 (y1 , y2 ) = (9)
0 otherwise
202
For 0 ≤ y1 ≤ 1, the marginal PDF of Y1 can be found from
' ∞ ' 1
fY1 (y1 ) = fY1 ,Y2 (y1 , y2 ) dy2 = 12(1 − y2 )2 dy2 = 4(1 − y1 )3 (10)
−∞ y1
The complete expression of the PDF of Y1 is

4(1 − y1 )3 0 ≤ y1 ≤ 1
fY1 (y1 ) = (11)
0 otherwise
(∞
Note that the integral fY1 (y1 ) = −∞ fY1 ,Y4 (y1 , y4 ) dy4 would have yielded the same result. This is
a good way to check our derivations of fY1 ,Y4 (y1 , y4 ) and fY1 ,Y2 (y1 , y2 ).

The value of each byte is an independent experiment with 255 possible outcomes. Each byte takes
on the value bi with probability pi = p = 1/255. The joint PMF of N0 , . . . , N255 is the multinomial
PMF
10000!
PN0 ,...,N255 (n0 , . . . , n255 ) = pn0 pn1 · · · pn255 n0 + · · · + n255 = 10000 (1)
n0 !n1 ! · · · n255 !
10000!
= (1/255)10000 n0 + · · · + n255 = 10000 (2)
n0 !n1 ! · · · n255 !
To evaluate the joint PMF of N0 and N1 , we define a new experiment with three categories: b0 ,
b1 and “other.” Let N̂ denote the number of bytes that are “other.” In this case, a byte is in the
“other” category with probability p̂ = 253/255. The joint PMF of N0 , N1 , and N̂ is
n0 n1 n̂
10000! 1 1 253
PN0 ,N1 ,N̂ (n0 , n1 , n̂) = n0 + n1 + n̂ = 10000 (3)
n0 !n1 !n̂! 255 255 255
Now we note that the following events are one in the same:
3 4
{N0 = n0 , N1 = n1 } = N0 = n0 , N1 = n1 , N̂ = 10000 − n0 − n1 (4)
Hence, for non-negative integers n0 and n1 satisfying n0 + n1 ≤ 10000,
PN0 ,N1 (n0 , n1 ) = PN0 ,N1 ,N̂ (n0 , n1 , 10000 − n0 − n1 ) (5)

10000! 1 n0 +n1 253 10000−n0 −n1
= (6)
n0 !n1 !(10000 − n0 − n1 )! 255 255

In Example 5.1, random variables N1 , . . . , Nr have the multinomial distribution

n
PN1 ,...,Nr (n1 , . . . , nr ) = pn1 · · · pnr r (1)
n1 , . . . , nr 1
where n > r > 2.
203
(a) To evaluate the joint PMF of N1 and N2 , we define a new experiment with mutually exclusive
events: s1 , s2 and “other” Let N̂ denote the number of trial outcomes that are “other”. In
this case, a trial is in the “other” category with probability p̂ = 1 − p1 − p2 . The joint PMF
of N1 , N2 , and N̂ is
n!
PN1 ,N2 ,N̂ (n1 , n2 , n̂) = pn1 pn2 (1 − p1 − p2 )n̂ n1 + n2 + n̂ = n (2)
n1 !n2 !n̂! 1 2
Now we note that the following events are one in the same:
3 4
{N1 = n1 , N2 = n2 } = N1 = n1 , N2 = n2 , N̂ = n − n1 − n2 (3)
Hence, for non-negative integers n1 and n2 satisfying n1 + n2 ≤ n,
PN1 ,N2 (n1 , n2 ) = PN1 ,N2 ,N̂ (n1 , n2 , n − n1 − n2 ) (4)

n!
= pn1 pn2 (1 − p1 − p2 )n−n1 −n2 (5)
n1 !n2 !(n − n1 − n2 )! 1 2
(b) We could find the PMF of Ti by summing the joint PMF PN1 ,...,Nr (n1 , . . . , nr ). However, it
is easier to start from first principles. Suppose we say a success occurs if the outcome of
the trial is in the set {s1 , s2 , . . . , si } and otherwise a failure occurs. In this case, the success
probability is qi = p1 + · · · + pi and Ti is the number of successes in n trials. Thus, Ti has the
binomial PMF n t
t qi (1 − qi )
n−t t = 0, 1, . . . , n
PTi (t) = (6)
0 otherwise
(c) The joint PMF of T1 and T2 satisfies
PT1 ,T2 (t1 , t2 ) = P [N1 = t1 , N1 + N2 = t2 ] (7)

= P [N1 = t1 , N2 = t2 − t1 ] (8)
= PN1 ,N2 (t1 , t2 − t1 ) (9)
By the result of part (a),
n!
PT1 ,T2 (t1 , t2 ) = pt1 pt2 −t1 (1 − p1 − p2 )n−t2 0 ≤ t 1 ≤ t2 ≤ n (10)
t1 !(t2 − t1 )!(n − t2 )! 1 2
(a) Note that Z is the number of three-page faxes. In principle, we can sum the joint PMF
PX,Y,Z (x, y, z) over all x, y to find PZ (z). However, it is better to realize that each fax has 3
pages with probability 1/6, independent of any other fax. Thus, Z has the binomial PMF
5 z 5−z z = 0, 1, . . . , 5
PZ (z) = z (1/6) (5/6) (1)
0 otherwise
(b) From the properties of the binomial distribution given in Appendix A, we know that E[Z] =
5(1/6).
204
(c) We want to find the conditional PMF of the number X of 1-page faxes and number Y of
2-page faxes given Z = 2 3-page faxes. Note that given Z = 2, X + Y = 3. Hence for
non-negative integers x, y satisfying x + y = 3,
5! x y 2
PX,Y,Z (x, y, 2) x!y!2! (1/3) (1/2) (1/6)
PX,Y |Z (x, y|2) = = 5 (2)
PZ (2) 2
2 (1/6) (5/6)
3
With some algebra, the complete expression of the conditional PMF is

3! x y x + y = 3, x ≥ 0, y ≥ 0; x, y integer
PX,Y |Z (x, y|2) = x!y! (2/5) (3/5) (3)
0 otherwise
In the above expression, we note that if Z = 2, then Y = 3 − X and

3 x 3−x x = 0, 1, 2, 3
PX|Z (x|2) = PX,Y |Z (x, 3 − x|2) = x (2/5) (3/5) (4)
0 otherwise
That is, given Z = 2, there are 3 faxes left, each of which independently could be a 1-page fax.
The conditonal PMF of the number of 1-page faxes is binomial where 2/5 is the conditional
probability that a fax has 1 page given that it either has 1 page or 2 pages. Moreover given
X = x and Z = 2 we must have Y = 3 − x.
(d) Given Z = 2, the conditional PMF of X is binomial for 3 trials and success probability 2/5.
The conditional expectation of X given Z = 2 is E[X|Z = 2] = 3(2/5) = 6/5.
(e) There are several ways to solve this problem. The most straightforward approach is to
realize that for integers 0 ≤ x ≤ 5 and 0 ≤ y ≤ 5, the event {X = x, Y = y} occurs iff
{X = x, Y = y, Z = 5 − (x + y)}. For the rest of this problem, we assume x and y are non-
negative integers so that
PX,Y (x, y) = PX,Y,Z (x, y, 5 − (x + y)) (5)

1 x 1 y 1 5−x−y
5!
x!y!(5−x−y)! 3 2 6 0 ≤ x + y ≤ 5, x ≥ 0, y ≥ 0
= (6)
0 otherwise
The above expression may seem unwieldy and it isn’t even clear that it will sum to 1. To
simplify the expression, we observe that
PX,Y (x, y) = PX,Y,Z (x, y, 5 − x − y) = PX,Y |Z (x, y|5 − x + y) PZ (5 − x − y) (7)
Using PZ (z) found in part (c), we can calculate PX,Y |Z (x, y|5 − x − y) for 0 ≤ x + y ≤ 5,
integer valued.
PX,Y,Z (x, y, 5 − x − y)
PX,Y |Z (x, y|5 − x + y) = (8)
PZ (5 − x − y)
x y
x+y 1/3 1/2
= (9)
x 1/2 + 1/3 1/2 + 1/3
x (x+y)−x
x+y 2 3
= (10)
x 5 5
In the above expression, it is wise to think of x + y as some fixed value. In that case, we see
that given x + y is a fixed value, X and Y have a joint PMF given by a binomial distribution
205
in x. This should not be surprising since it is just a generalization of the case when Z = 2.
That is, given that there were a fixed number of faxes that had either one or two pages, each
of those faxes is a one page fax with probability (1/3)/(1/2 + 1/3) and so the number of one
page faxes should have a binomial distribution, Moreover, given the number X of one page
faxes, the number Y of two page faxes is completely specified.
Finally, by rewriting PX,Y (x, y) given above, the complete expression for the joint PMF of X
and Y is
1 5−x−y 5 x+y x+y 2 x 3 y
5
5−x−y 6 6 x 5 5 x, y ≥ 0
PX,Y (x, y) = (11)
0 otherwise

In Problem 5.3.2, we found that the joint PMF of K = K1 K2 K3 is

p3 (1 − p)k3 −3 k1 < k2 < k3
PK (k) = (1)
0 otherwise
In this problem, we generalize the result to n messages.
(a) For k1 < k2 < · · · < kn , the joint event
{K1 = k1 , K2 = k2 , · · · , Kn = kn } (2)
occurs if and only if all of the following events occur
A1 k1 − 1 failures, followed by a successful transmission

A2 (k2 − 1) − k1 failures followed by a successful transmission
A3 (k3 − 1) − k2 failures followed by a successful transmission
..
.
An (kn − 1) − kn−1 failures followed by a successful transmission
Note that the events A1 , A2 , . . . , An are independent and
P [Aj ] = (1 − p)kj −kj−1 −1 p. (3)
Thus
PK1 ,...,Kn (k1 , . . . , kn ) = P [A1 ] P [A2 ] · · · P [An ] (4)

(k1 −1)+(k2 −k1 −1)+(k3 −k2 −1)+···+(kn −kn−1 −1)
= p (1 − p)
n
(5)
= pn (1 − p)kn −n (6)

To clarify subsequent results, it is better to rename K as Kn = K1 K2 · · · Kn . We see
that n
p (1 − p)kn −n 1 ≤ k1 < k2 < · · · < kn ,
PKn (kn ) = (7)
0 otherwise.
206
(b) For j < n,
PK1 ,K2 ,...,Kj (k1 , k2 , . . . , kj ) = PKj (kj ) . (8)
Since Kj is just Kn with n = j, we have
j
p (1 − p)kj −j 1 ≤ k1 < k2 < · · · < kj ,
PKj (kj ) = (9)
0 otherwise.
(c) Rather than try to deduce PKi (ki ) from the joint PMF PKn (kn ), it is simpler to return to
first principles. In particular, Ki is the number of trials up to and including the ith success
and has the Pascal (i, p) PMF

ki − 1 i
PKi (ki ) = p (1 − p)ki −i . (10)
i−1

For i = j, Xi and Xj are independent and E[Xi Xj ] = E[Xi ]E[Xj ] = 0 since E[Xi ] = 0. Thus the
i, jth entry in the covariance matrix CX is
2
σi i = j,
CX (i, j) = E [Xi Xj ] = (1)
0 otherwise.

Thus for random vector X = X1 X2 · · · Xn , all the off-diagonal entries in the covariance
matrix are zero and the covariance matrix is
⎡ 2 ⎤
σ1
⎢ σ22 ⎥
⎢ ⎥
CX = ⎢ .. ⎥. (2)
⎣ . ⎦
σn2

The random variables N1 , N2 , N3 and N4 are dependent. To see this we observe that PNi (4) = p4i .
However,
PN1 ,N2 ,N3 ,N4 (4, 4, 4, 4) = 0 = p41 p42 p43 p44 = PN1 (4) PN2 (4) PN3 (4) PN4 (4) . (1)

We will use the PDF
1 0 ≤ xi ≤ 1, i = 1, 2, 3, 4
fX (x) = (1)
0 otherwise.
to find the marginal PDFs fXi (xi ). In particular, for 0 ≤ x1 ≤ 1,
' 1' 1' 1
fX1 (x1 ) = fX (x) dx2 dx3 dx4 (2)
0 0 0
' 1 ' 1 ' 1
= dx2 dx3 dx4 = 1. (3)
0 0 0
207
Thus,
1 0 ≤ x ≤ 1,
fX1 (x1 ) = (4)
0 otherwise.
Following similar steps, one can show that

1 0 ≤ x ≤ 1,
fX1 (x) = fX2 (x) = fX3 (x) = fX4 (x) = (5)
0 otherwise.
Thus
fX (x) = fX1 (x) fX2 (x) fX3 (x) fX4 (x) . (6)
We conclude that X1 , X2 , X3 and X4 are independent.

We will use the PDF

6e−(x1 +2x2 +3x3 ) x1 ≥ 0, x2 ≥ 0, x3 ≥ 0
fX (x) = (1)
0 otherwise.
to find the marginal PDFs fXi (xi ). In particular, for x1 ≥ 0,
' ∞' ∞
fX1 (x1 ) = fX (x) dx2 dx3 (2)
0
' ∞
0
' ∞
−x1 −2x2
= 6e e dx2 e−3x3 dx3 (3)
0 $∞ 0 $
−x1 1 $
−2x2 $ 1 −3x3 $$∞
= 6e − e $ − e $ = e−x1 . (4)
2 0 3 0
Thus,
e−x1 x1 ≥ 0,
fX1 (x1 ) = (5)
0 otherwise.
Following similar steps, one can show that
' ∞' ∞ −2x
2 2 x2 ≥ 0,
fX2 (x2 ) = fX (x) dx1 dx3 = (6)
0 0 0 otherwise.
' ∞' ∞ −3x
3 3 x3 ≥ 0,
fX3 (x3 ) = fX (x) dx1 dx2 = (7)
0 0 0 otherwise.
Thus
fX (x) = fX1 (x1 ) fX2 (x2 ) fX3 (x3 ) . (8)
We conclude that X1 , X2 , and X3 are independent.

This problem can be solved without any real math. Some thought should convince you that for any
xi > 0, fXi (xi ) > 0. Thus, fX1 (10) > 0, fX2 (9) > 0, and fX3 (8) > 0. Thus fX1 (10)fX2 (9)fX3 (8) >
0. However, from the definition of the joint PDF
fX1 ,X2 ,X3 (10, 9, 8) = 0 = fX1 (10) fX2 (9) fX3 (8) . (1)
It follows that X1 , X2 and X3 are dependent. Readers who find this quick answer dissatisfying
are invited to confirm this conclusions by solving Problem 5.4.6 for the exact expressions for the
marginal PDFs fX1 (x1 ), fX2 (x2 ), and fX3 (x3 ).
208
We find the marginal PDFs using Theorem 5.5. First we note that for x < 0, fXi (x) = 0. For
x1 ≥ 0,
' ∞ ' ∞ ' ∞
−x3
fX1 (x1 ) = e dx3 dx2 = e−x2 dx2 = e−x1 (1)
x1 x2 x1
Similarly, for x2 ≥ 0, X2 has marginal PDF

' x 2 ' ∞ ' x2
−x3
fX2 (x2 ) = e dx3 dx1 = e−x2 dx1 = x2 e−x2 (2)
0 x2 0
Lastly,
' x3 ' x3 ' x3
−x3
fX3 (x3 ) = e dx2 dx1 = (x3 − x1 )e−x3 dx1 (3)
0 x1 0
$x1 =x3
1 $
2 −x3 $ 1
= − (x3 − x1 ) e $ = x23 e−x3 (4)
2 x1 =0 2
The complete expressions for the three marginal PDFs are

−x
e 1 x1 ≥ 0
fX1 (x1 ) = (5)
0 otherwise
−x
x2 e 2 x2 ≥ 0
fX2 (x2 ) = (6)
0 otherwise
−x
(1/2)x3 e 3 x3 ≥ 0
2
fX3 (x3 ) = (7)
0 otherwise
In fact, each Xi is an Erlang (n, λ) = (i, 1) random variable.

Since U1 , . . . , Un are iid uniform (0, 1) random variables,

1/T n 0 ≤ ui ≤ 1; i = 1, 2, . . . , n
fU1 ,...,Un (u1 , . . . , un ) = (1)
0 otherwise
Since U1 , . . . , Un are continuous, P [Ui = Uj ] = 0 for all i = j. For the same reason, P [Xi = Xj ] = 0
for i = j. Thus we need only to consider the case when x1 < x2 < · · · < xn .
To understand the claim, it is instructive to start with the n = 2 case. In this case, (X1 , X2 ) =
(x1 , x2 ) (with x1 < x2 ) if either (U1 , U2 ) = (x1 , x2 ) or (U1 , U2 ) = (x2 , x1 ). For infinitesimal ∆,
fX1 ,X2 (x1 , x2 ) ∆2 = P [x1 < X1 ≤ x1 + ∆, x2 < X2 ≤ x2 + ∆] (2)

= P [x1 < U1 ≤ x1 + ∆, x2 < U2 ≤ x2 + ∆]
+ P [x2 < U1 ≤ x2 + ∆, x1 < U2 ≤ x1 + ∆] (3)
= fU1 ,U2 (x1 , x2 ) ∆2 + fU1 ,U2 (x2 , x1 ) ∆2 (4)
We see that for 0 ≤ x1 < x2 ≤ 1 that
fX1 ,X2 (x1 , x2 ) = 2/T n . (5)
209

For the general case of n uniform random variables, we define π = π(1) . . . π(n) as a permu-
tation vector of the integers 1, 2, . . . , n and Π as the set of n! possible permutation vectors. In this
case, the event {X1 = x1 , X2 = x2 , . . . , Xn = xn } occurs if
U1 = xπ(1) , U2 = xπ(2) , . . . , Un = xπ(n) (6)
for any permutation π ∈ Π. Thus, for 0 ≤ x1 < x2 < · · · < xn ≤ 1,

fX1 ,...,Xn (x1 , . . . , xn ) ∆n = fU1 ,...,Un xπ(1) , . . . , xπ(n) ∆n . (7)
π∈Π
Since there are n! permutations and fU1 ,...,Un (xπ(1) , . . . , xπ(n) ) = 1/T n for each permutation π, we
can conclude that
fX1 ,...,Xn (x1 , . . . , xn ) = n!/T n . (8)
Since the order statistics are necessarily ordered, fX1 ,...,Xn (x1 , . . . , xn ) = 0 unless x1 < · · · < xn .

For discrete random vectors, it is true in general that
PY (y) = P [Y = y] = P [AX + b = y] = P [AX = y − b] . (1)
For an arbitrary matrix A, the system of equations Ax = y − b may have no solutions (if the
columns of A do not span the vector space), multiple solutions (if the columns of A are linearly
dependent), or, when A is invertible, exactly one solution. In the invertible case,

PY (y) = P [AX = y − b] = P X = A−1 (y − b) = PX A−1 (y − b) . (2)
As an aside, we note that when Ax = y − b has multiple solutions, we would need to do some
bookkeeping to add up the probabilities PX (x) for all vectors x satisfying Ax = y − b. This can
get disagreeably complicated.

The random variable Jn is the number of times that message n is transmitted. Since each trans-
mission is a success with probability p, independent of any other transmission, the number of
transmissions of message n is independent of the number of transmissions of message m. That
is, for m = n, Jm and Jn are independent random variables. Moreover, because each message is
transmitted over and over until it is transmitted succesfully, each Jm is a geometric (p) random
variable with PMF
(1 − p)j−1 p j = 1, 2, . . .
PJm (j) = (1)
0 otherwise.

Thus the PMF of J = J1 J2 J3 is
⎧ 3
⎪
⎪ p (1 − p)j1 +j2 +j3 −3 ji = 1, 2, . . . ;
⎨
i = 1, 2, 3
PJ (j) = PJ1 (j1 ) PJ2 (j2 ) PJ3 (j3 ) = (2)
⎪
⎪
⎩
0 otherwise.
210
The response time Xi of the ith truck has PDF fXi (xi ) and CDF FXi (xi ) given by
1 −x/2
2 e x ≥ 0, 1 − e−x/2 x ≥ 0
fXi (xi ) = FXi (xi ) = FX (xi ) = (1)
0 otherwise, 0 otherwise.
Let R = max(X1 , X2 , . . . , X6 ) denote the maximum response time. From Theorem 5.7, R has PDF
FR (r) = (FX (r))6 . (2)
(a) The probability that all six responses arrive within five seconds is
P [R ≤ 5] = FR (5) = (FX (5))6 = (1 − e−5/2 )6 = 0.5982. (3)
(b) This question is worded in a somewhat confusing way. The “expected response time” refers
to E[Xi ], the response time of an individual truck, rather than E[R]. If the expected response
time of a truck is τ , then each Xi has CDF

1 − e−x/τ x ≥ 0
FXi (x) = FX (x) = (4)
0 otherwise.
The goal of this problem is to find the maximum permissible value of τ . When each truck
has expected response time τ , the CDF of R is

6 (1 − e−r/τ )6 r ≥ 0,
FR (r) = (FX (x) r) = (5)
0 otherwise.
We need to find τ such that
P [R ≤ 3] = (1 − e−3/τ )6 = 0.9. (6)
This implies
−3
τ= = 0.7406 s. (7)
ln 1 − (0.9)1/6

Let Xi denote the finishing time of boat i. Since finishing times of all boats are iid Gaussian random
variables with expected value 35 minutes and standard deviation 5 minutes, we know that each Xi
has CDF
Xi − 35 x − 35 x − 35
FXi (x) = P [Xi ≤ x] = P ≤ =Φ (1)
5 5 5
(a) The time of the winning boat is
W = min(X1 , X2 , . . . , X10 ) (2)
To find the probability that W ≤ 25, we will find the CDF FW (w) since this will also be
useful for part (c).
FW (w) = P [min(X1 , X2 , . . . , X10 ) ≤ w] (3)

= 1 − P [min(X1 , X2 , . . . , X10 ) > w] (4)
= 1 − P [X1 > w, X2 > w, . . . , X10 > w] (5)
211
Since the Xi are iid,

10
FW (w) = 1 − P [Xi > w] = 1 − (1 − FXi (w))10 (6)
i=1
10
w − 35
=1− 1−Φ (7)
5
Thus,
P [W ≤ 25] = FW (25) = 1 − (1 − Φ(−2))10 (8)

= 1 − [Φ(2)]
10
= 0.2056. (9)
(b) The finishing time of the last boat is L = max(X1 , . . . , X10 ). The probability that the last
boat finishes in more than 50 minutes is
P [L > 50] = 1 − P [L ≤ 50] (10)

= 1 − P [X1 ≤ 50, X2 ≤ 50, . . . , X10 ≤ 50] (11)
Once again, since the Xi are iid Gaussian (35, 5) random variables,

10
P [L > 50] = 1 − P [Xi ≤ 50] = 1 − (FXi (50))10 (12)
i=1
= 1 − (Φ([50 − 35]/5))10 (13)
= 1 − (Φ(3)) 10
= 0.0134 (14)
(c) A boat will finish in negative time if and only iff the winning boat finishes in negative time,
which has probability
FW (0) = 1 − (1 − Φ(−35/5))10 = 1 − (1 − Φ(−7))10 = 1 − (Φ(7))10 . (15)
Unfortunately, the tables in the text have neither Φ(7) nor Q(7). However, those with access
to Matlab, or a programmable calculator, can find out that Q(7) = 1 − Φ(7) = 1.28 × 10−12 .
This implies that a boat finishes in negative time with probability
FW (0) = 1 − (1 − 1.28 × 10−12 )10 = 1.28 × 10−11 . (16)

Since 50 cents of each dollar ticket is added to the jackpot,
Ni
Ji−1 = Ji + (1)
2
Given Ji = j, Ni has a Poisson distribution with mean j. It follows that E[Ni |Ji = j] = j and that
Var[Ni |Ji = j] = j. This implies

E Ni2 |Ji = j = Var[Ni |Ji = j] + (E [Ni |Ji = j])2 = j + j 2 (2)
212
In terms of the conditional expectations given Ji , these facts can be written as

E [Ni |Ji ] = Ji E Ni2 |Ji = Ji + Ji2 (3)
This permits us to evaluate the moments of Ji−1 in terms of the moments of Ji . Specifically,
1 Ji 3Ji
E [Ji−1 |Ji ] = E [Ji |Ji ] + E [Ni |Ji ] = Ji + = (4)
2 2 2
This implies
3
E [Ji−1 ] = E [Ji ] (5)
2
We can use this the calculate E[Ji ] for all i. Since the jackpot starts at 1 million dollars, J6 = 106
and E[J6 ] = 106 . This implies
E [Ji ] = (3/2)6−i 106 (6)
Now we will find the second moment E[Ji2 ]. Since Ji−1
2 = J 2 + N J + N 2 /4, we have
i i i i
2 2 2
E Ji−1 |Ji = E Ji |Ji + E [Ni Ji |Ji ] + E Ni |Ji /4 (7)
= Ji2 + Ji E [Ni |Ji ] + (Ji + Ji2 )/4 (8)
= (3/2)2 Ji2 + Ji /4 (9)
By taking the expectation over Ji we have
2
E Ji−1 = (3/2)2 E Ji2 + E [Ji ] /4 (10)
This recursion allows us to calculate E[Ji2 ] for i = 6, 5, . . . , 0. Since J6 = 106 , E[J62 ] = 1012 . From
the recursion, we obtain
1
E J52 = (3/2)2 E J62 + E [J6 ] /4 = (3/2)2 1012 + 106 (11)
4
1
E J42 = (3/2)2 E J52 + E [J5 ] /4 = (3/2)4 1012 + (3/2)2 + (3/2) 106 (12)
4
1
E J32 = (3/2)2 E J42 + E [J4 ] /4 = (3/2)6 1012 + (3/2)4 + (3/2)3 + (3/2)2 106 (13)
4
The same recursion will also allow us to show that
1
E J22 = (3/2)8 1012 + (3/2)6 + (3/2)5 + (3/2)4 + (3/2)3 106 (14)
4
1
E J12 = (3/2)10 1012 + (3/2)8 + (3/2)7 + (3/2)6 + (3/2)5 + (3/2)4 106 (15)
4
1
E J02 = (3/2)12 1012 + (3/2)10 + (3/2)9 + · · · + (3/2)5 106 (16)
4
Finally, day 0 is the same as any other day in that J = J0 + N0 /2 where N0 is a Poisson random
variable with mean J0 . By the same argument that we used to develop recursions for E[Ji ] and
E[Ji2 ], we can show
E [J] = (3/2)E [J0 ] = (3/2)7 106 ≈ 17 × 106 (17)
and

E J 2 = (3/2)2 E J02 + E [J0 ] /4 (18)
1
= (3/2)14 1012 + (3/2)12 + (3/2)11 + · · · + (3/2)6 106 (19)
4
106
= (3/2)14 1012 + (3/2)6 [(3/2)7 − 1] (20)
2
213
Finally, the variance of J is
106
Var[J] = E J 2 − (E [J])2 = (3/2)6 [(3/2)7 − 1] (21)
2
Since the variance is hard to interpret, we note that the standard deviation of J is σJ ≈ 9572.
Although the expected jackpot grows rapidly, the standard deviation of the jackpot is fairly small.

Let A denote the event Xn = max(X1 , . . . , Xn ). We can find P [A] by conditioning on the value of
Xn .
P [A] = P [X1 ≤ Xn , X2 ≤ Xn , · · · , Xn1 ≤ Xn ] (1)

' ∞
= P [X1 < Xn , X2 < Xn , · · · , Xn−1 < Xn |Xn = x] fXn (x) dx (2)
−∞
' ∞
= P [X1 < x, X2 < x, · · · , Xn−1 < x|Xn = x] fX (x) dx (3)
−∞
Since X1 , . . . , Xn−1 are independent of Xn ,

' ∞
P [A] = P [X1 < x, X2 < x, · · · , Xn−1 < x] fX (x) dx. (4)
−∞
Since X1 , . . . , Xn−1 are iid,

' ∞
P [A] = P [X1 ≤ x] P [X2 ≤ x] · · · P [Xn−1 ≤ x] fX (x) dx (5)
−∞
' ∞ $∞
1 $
n$ 1
= [FX (x)]n−1
fX (x) dx = [FX (x)] $ = (1 − 0) (6)
−∞ n −∞ n
Not surprisingly, since the Xi are identical, symmetry would suggest that Xn is as likely as any of
the other Xi to be the largest. Hence P [A] = 1/n should not be surprising.

(a) The coavariance matrix of X = X1 X2 is

Var[X1 ] Cov [X1 , X2 ] 4 3
CX = = . (1)
Cov [X1 , X2 ] Var[X2 ] 3 9
(b) From the problem statement,

Y1 1 −2
Y= = X = AX. (2)
Y2 3 4
By Theorem 5.13, Y has covariance matrix

1 −2 4 3 1 3 28 −66
CY = ACX A = = . (3)
3 4 3 9 −2 4 −66 252
214
The mean value of a sum of random variables is always the sum of their individual means.

n
E [Y ] = E [Xi ] = 0 (1)
i=1
The variance of any sum of random variables can be expressed in terms of the individual variances
and co-variances. Since the E[Y ] is zero, Var[Y ] = E[Y 2 ]. Thus,
⎡! "2 ⎤ ⎡ ⎤
n n n n
n
Var[Y ] = E ⎣ Xi ⎦ = E ⎣ Xi Xj ⎦ = E Xi2 + E [Xi Xj ] (2)
i=1 i=1 j=1 i=1 i=1 j
=i
Since E[Xi ] = 0, E[Xi2 ] = Var[Xi ] = 1 and for i = j,
E [Xi Xj ] = Cov [Xi , Xj ] = ρ (3)
Thus, Var[Y ] = n + n(n − 1)ρ.

Since X and Y are independent and E[Yj ] = 0 for all components Yj , we observe that E[Xi Yj ] =
E[Xi ]E[Yj ] = 0. This implies that the cross-covariance matrix is

E XY = E [X] E Y = 0. (1)

Inspection of the vector PDF fX (x) will show that X1 , X2 , X3 , and X4 are iid uniform (0, 1)
random variables. That is,
fX (x) = fX1 (x1 ) fX2 (x2 ) fX3 (x3 ) fX4 (x4 ) (1)
where each Xi has the uniform (0, 1) PDF

1 0≤x≤1
fXi (x) = (2)
0 otherwise
It follows that for each i, E[Xi ] = 1/2, E[Xi2 ] = 1/3 and Var[Xi ] = 1/12. In addition, Xi and Xj
have correlation
E [Xi Xj ] = E [Xi ] E [Xj ] = 1/4. (3)
and covariance Cov[Xi , Xj ] = 0 for i = j since independent random variables always have zero
covariance.
(a) The expected value vector is

E [X] = E [X1 ] E [X2 ] E [X3 ] E [X4 ] = 1/2 1/2 1/2 1/2 . (4)
215
(b) The correlation matrix is

⎡ ⎤
E X12 E [X
1 X2 ] E [X1 X3 ] E [X1 X4 ]
⎢E [X2 X1 ] E X22 E [X ⎥
RX = E XX = ⎢ 2X 3 ] E [X2 X4 ]⎥ (5)
⎣E [X3 X1 ] E [X3 X2 ] E X3 2 E [X ⎦
3X
2
4 ]
E [X4 X1 ] E [X4 X2 ] E [X4 X3 ] E X4
⎡ ⎤
1/3 1/4 1/4 1/4
⎢1/4 1/3 1/4 1/4⎥
=⎢⎣1/4 1/4
⎥ (6)
1/3 1/4⎦
1/4 1/4 1/4 1/3
(c) The covariance matrix for X is the diagonal matrix

⎡ ⎤
Var[X1 ] Cov [X1 , X2 ] Cov [X1 , X3 ] Cov [X1 , X4 ]
⎢Cov [X2 , X1 ] Var[X2 ] Cov [X2 , X3 ] Cov [X2 , X4 ]⎥
CX = ⎢⎣Cov [X3 , X1 ] Cov [X3 , X2 ]
⎥ (7)
Var[X3 ] Cov [X3 , X4 ]⎦
Cov [X4 , X1 ] Cov [X4 , X2 ] Cov [X4 , X3 ] Var[X4 ]
⎡ ⎤
1/12 0 0 0
⎢ 0 1/12 0 0 ⎥
=⎢⎣ 0
⎥ (8)
0 1/12 0 ⎦
0 0 0 1/12
Note that its easy to verify that CX = RX − µX µX .

The random variable Jm is the number of times that message m is transmitted. Since each trans-
mission is a success with probability p, independent of any other transmission, J1 , J2 and J3 are
iid geometric (p) random variables with
1 1−p
E [Jm ] = , Var[Jm ] = . (1)
p p2

Thus the vector J = J1 J2 J3 has expected value

E [J] = E [J1 ] E [J2 ] EJ3 = 1/p 1/p 1/p . (2)
For m = n, the correlation matrix RJ has m, nth entry
RJ (m, n) = E [Jm Jn ] = E [Jm ] Jn = 1/p2 (3)
For m = n,
2 2 2 1−p 1 2−p
RJ (m, m) = E Jm = Var[Jm ] + (E Jm ) = 2
+ 2 = . (4)
p p p2
Thus ⎡ ⎤
2−p 1 1
1 ⎣
RJ = 2 1 2−p 1 ⎦. (5)
p
1 1 2−p
Because Jm and Jn are independent, off-diagonal terms in the covariance matrix are
CJ (m, n) = Cov [Jm , Jn ] = 0 (6)
216
Since CJ (m, m) = Var[Jm ], we have that
⎡ ⎤
1 0 0
1−p 1−p⎣
CJ = 2
I= 0 1 0⎦ . (7)
p p2
0 0 1

This problem is quite difficult
unless oneuses the observation that the vector K can be expressed
in terms of the vector J = J1 J2 J3 where Ji is the number of transmissions of message i.
Note that we can write ⎡ ⎤
1 0 0
K = AJ = ⎣1 1 0⎦ J (1)
1 1 1
We also observe that since each transmission is an independent Bernoulli trial with success prob-
ability p, the components of J are iid geometric (p) random variables. Thus E[Ji ] = 1/p and
Var[Ji ] = (1 − p)/p2 . Thus J has expected value

E [J] = µJ = E [J1 ] E [J2 ] E [J3 ] = 1/p 1/p 1/p . (2)
Since the components of J are independent, it has the diagonal covariance matrix
⎡ ⎤
Var[J1 ] 0 0
1−p
CJ = ⎣ 0 Var[J2 ] 0 ⎦= I (3)
p2
0 0 Var[J3 ]
Given these properties of J, finding the same properties of K = AJ is simple.
(a) The expected value of K is

⎡ ⎤⎡ ⎤ ⎡ ⎤
1 0 0 1/p 1/p
E [K] = AµJ = ⎣1 1 0⎦ ⎣1/p⎦ = ⎣2/p⎦ (4)
1 1 1 1/p 3/p
(b) From Theorem 5.13, the covariance matrix of K is
CK = ACJ A (5)
1−p
= AIA (6)
p2
⎡ ⎤⎡ ⎤ ⎡ ⎤
1 0 0 1 1 1 1 1 1
1−p⎣ 1−p⎣
= 2
1 1 0⎦ ⎣0 1 1⎦ = 1 2 2⎦ (7)
p p2
1 1 1 0 0 1 1 2 3
(c) Given the expected value vector µK and the covariance matrix CK , we can use Theorem 5.12
217
to find the correlation matrix
RK = CK + µK µK (8)
⎡ ⎤ ⎡ ⎤
1 1 1 1/p
1−p⎣ ⎦ + ⎣2/p⎦ 1/p 2/p 3/p
= 1 2 2 (9)
p2
1 2 3 3/p
⎡ ⎤ ⎡ ⎤
1 1 1 1 2 3
1−p⎣ 1
= 2
1 2 2⎦ + 2 ⎣2 4 6⎦ (10)
p p
1 2 3 3 6 9
⎡ ⎤
2−p 3−p 4−p
1
= 2 ⎣3 − p 6 − 2p 8 − 2p ⎦ (11)
p
4 − p 8 − 2p 12 − 3p

The preliminary work for this problem appears in a few different places. In Example 5.5, we found
the marginal PDF of Y3 and in Example 5.6, we found the marginal PDFs of Y1 , Y2 , and Y4 . We
summarize these results here:

2(1 − y) 0 ≤ y ≤ 1,
fY1 (y) = fY3 (y) = (1)
0 otherwise,

2y 0 ≤ y ≤ 1,
fY2 (y) = fY4 (y) = (2)
0 otherwise.
This implies
' 1
E [Y1 ] = E [Y3 ] = 2y(1 − y) dy = 1/3 (3)
0
' 1
E [Y2 ] = E [Y4 ] = 2y 2 dy = 2/3 (4)
0

Thus Y has expected value E[Y] = 1/3 2/3 1/3 2/3 . The second part of the problem is to
find the correlation matrix RY . In fact, we need to find RY (i, j) = E[Yi Yj ] for each i, j pair. We
will see that these are seriously tedious calculations. For i = j, the second moments are
' 1

E Y12 = E Y32 = 2y 2 (1 − y) dy = 1/6, (5)
0
' 1
2 2
E Y2 = E Y4 = 2y 3 dy = 1/2. (6)
0
In terms of the correlation matrix,

RY (1, 1) = RY (3, 3) = 1/6, RY (2, 2) = RY (4, 4) = 1/2. (7)
To find the off diagonal terms RY (i, j) = E[Yi Yj ], we need to find the marginal PDFs fYi ,Yj (yi , yj ).
Example 5.5 showed that

4(1 − y1 )y4 0 ≤ y1 ≤ 1, 0 ≤ y4 ≤ 1,
fY1 ,Y4 (y1 , y4 ) = (8)
0 otherwise.

4y2 (1 − y3 ) 0 ≤ y2 ≤ 1, 0 ≤ y3 ≤ 1,
fY2 ,Y3 (y2 , y3 ) = (9)
0 otherwise.
218
Inspection will show that Y1 and Y4 are independent since fY1 ,Y4 (y1 , y4 ) = fY1 (y1 )fY4 (y4 ). Similarly,
Y2 and Y4 are independent since fY2 ,Y3 (y2 , y3 ) = fY2 (y2 )fY3 (y3 ). This implies
RY (1, 4) = E [Y1 Y4 ] = E [Y1 ] E [Y4 ] = 2/9 (10)

RY (2, 3) = E [Y2 Y3 ] = E [Y2 ] E [Y3 ] = 2/9 (11)
We also need to calculate fY1 ,Y2 (y1 , y2 ), fY3 ,Y4 (y3 , y4 ), fY1 ,Y3 (y1 , y3 ) and fY2 ,Y4 (y2 , y4 ). To start, for
0 ≤ y1 ≤ y2 ≤ 1,
' ∞' ∞
fY1 ,Y2 (y1 , y2 ) = fY1 ,Y2 ,Y3 ,Y4 (y1 , y2 , y3 , y4 ) dy3 dy4 (12)
−∞ −∞
' 1 ' y4 ' 1
= 4 dy3 dy4 = 4y4 dy4 = 2. (13)
0 0 0
Similarly, for 0 ≤ y3 ≤ y4 ≤ 1,
' ∞ ' ∞
−∞ −∞
' 1 ' y2 ' 1
= 4 dy1 dy2 = 4y2 dy2 = 2. (15)
0 0 0
In fact, these PDFs are the same in that

2 0 ≤ x ≤ y ≤ 1,
fY1 ,Y2 (x, y) = fY3 ,Y4 (x, y) = (16)
0 otherwise.
This implies RY (1, 2) = RY (3, 4) = E[Y3 Y4 ] and that

' 1' y ' 1 '
2 $y 1
1
E [Y3 Y4 ] = 2xy dx dy = $
yx 0 dy = y 3 dy = . (17)
0 0 0 0 4
Continuing in the same way, we see for 0 ≤ y1 ≤ 1 and 0 ≤ y3 ≤ 1 that
' ∞' ∞
−∞ −∞
' 1 ' 1
=4 dy2 dy4 (19)
y1 y3
= 4(1 − y1 )(1 − y3 ). (20)
We observe that Y1 and Y3 are independent since fY1 ,Y3 (y1 , y3 ) = fY1 (y1 )fY3 (y3 ). It follows that
RY (1, 3) = E [Y1 Y3 ] = E [Y1 ] E [Y3 ] = 1/9. (21)
Finally, we need to calculate

' ∞ ' ∞
−∞ −∞
' y2 ' y4
=4 dy1 dy3 (23)
0 0
= 4y2 y4 . (24)
219
We observe that Y2 and Y4 are independent since fY2 ,Y4 (y2 , y4 ) = fY2 (y2 )fY4 (y4 ). It follows that
RY (2, 4) = E[Y2 Y4 ] = E[Y2 ]E[Y4 ] = 4/9. The above results give RY (i, j) for i ≤ j. Since RY is a
symmetric matrix, ⎡ ⎤
1/6 1/4 1/9 2/9
⎢1/4 1/2 2/9 4/9⎥
RY = ⎢ ⎣1/9 2/9 1/6 1/4⎦ .
⎥ (25)
2/9 4/9 1/4 1/2

Since µX = 1/3 2/3 1/3 2/3 , the covariance matrix is
CY = RY − µX µX (26)
⎡ ⎤
⎡ ⎤ 1/3
1/6 1/4 1/9 2/9 ⎢2/3⎥
= ⎣1/4 1/2 2/9 4/9⎦ − ⎢ ⎥
⎣1/3⎦ 1/3 2/3 1/3 2/3 (27)
2/9 4/9 1/4 1/2
2/3
⎡ ⎤
1/18 1/36 0 0
⎢1/36 1/18 0 0 ⎥
=⎢ ⎣ 0
⎥ . (28)
0 1/18 1/36⎦
0 0 1/36 1/18

The off-diagonal zero blocks are a consequence of Y1 Y2 being independent of Y3 Y4 . Along
the diagonal, the two identical sub-blocks occur because fY1 ,Y2 (x, y) = fY3 ,Y4 (x, y). In short, the

matrix structure is the result of Y1 Y2 and Y3 Y4 being iid random vectors.

The 2-dimensional random vector Y has PDF

2 y ≥ 0, 1 1 y ≤ 1,
fY (y) = (1)
0 otherwise.
Rewritten in terms of the variables y1 and y2 ,

2 y1 ≥ 0, y2 ≥ 0, y1 + y2 ≤ 1,
fY1 ,Y2 (y1 , y2 ) = (2)
0 otherwise.
In this problem, the PDF is simple enough that we can compute E[Yin ] for arbitrary integers n ≥ 0.
' ∞' ∞ ' 1 ' 1−y2
n n
E [Y1 ] = y1 fY1 ,Y2 (y1 , y2 ) dy1 dy2 = 2y1n dy1 dy2 . (3)
−∞ −∞ 0 0
A little calculus yields

' 1! $1−y2 " ' 1
2 $
n+1 $ 2 2
n
E [Y1 ] = y dy2 = (1 − y2 )n+1 dy2 = . (4)
0 n + 1 1 $0 n+1 0 (n + 1)(n + 2)
Symmetry of the joint PDF fY1 ,2 (y1 ,2 ) implies that E[Y2n ] = E[Y1n ]. Thus, E[Y1 ] = E[Y2 ] = 1/3
and
E [Y] = µY = 1/3 1/3 . (5)
In addition,

RY (1, 1) = E Y12 = 1/6, RY (2, 2) = E Y22 = 1/6. (6)
220
To complete the correlation matrix, we find
' ∞' ∞ ' 1 ' 1−y2
RY (1, 2) = E [Y1 Y2 ] = y1 y2 fY1 ,Y2 (y1 , y2 ) dy1 dy2 = 2y1 y2 dy1 dy2 . (7)
−∞ −∞ 0 0
Following through on the calculus, we obtain

' 1 ' 1 $
$
2 $1−y−2 1 2 2 3 1 4 $$1 1
RY (1, 2) = y1 0 y2 dy2 = y2 (1 − y2 ) dy2 = y2 − y2 + y2 $ = .
2
(8)
0 0 2 3 4 0 12
Thus we have found that

E Y12 E [Y Y
=
1 2 ] 1/6 1/12
RY = . (9)
E [Y2 Y1 ] E Y22 1/12 1/6
Lastly, Y has covariance matrix

1/6 1/12 1/3
CY = RY − µY µY = − 1/3 1/3 (10)
1/12 1/6 1/3

1/9 −1/36
= . (11)
−1/36 1/9

Given an arbitrary random vector X, we can define Y = X − µX so that

CX = E (X − µX )(X − µX ) = E YY = RY . (1)
It follows that the covariance matrix CX is positive semi-definite if and only if the correlation
matrix RY is positive semi-definite. Thus, it is sufficient to show that every correlation matrix,
whether it is denoted RY or RX , is positive semi-definite.
To show a correlation matrix RX is positive semi-definite, we write

a RX a = a E XX a = E a XX a = E (a X)(X a) = E (a X)2 . (2)
We note that W = a X is a random variable. Since E[W 2 ] ≥ 0 for any random variable W ,

a RX a = E W 2 ≥ 0. (3)
(a) From Theorem 5.12, the correlation matrix of X is
RX = CX + µX µX (1)
⎡ ⎤ ⎡ ⎤
4 −2 1 4
= ⎣−2 4 −2⎦ + ⎣8⎦ 4 8 6 (2)
1 −2 4 6
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
4 −2 1 16 32 24 20 30 25
= ⎣−2 4 −2⎦ + ⎣32 64 48⎦ = ⎣30 68 46⎦ (3)
1 −2 4 24 48 36 25 46 40
221

(b) Let Y = X1 X2 . Since Y is a subset of the components of X, it is a Gaussian random
vector with expected velue vector

µY = E [X1 ] E [X2 ] = 4 8 . (4)
and covariance matrix

Var[X1 ] Cov [X1 , X2 ] 4 −2
CY = = (5)
CX1 X2 Var[X2 ] −2 4
We note that det(CY ) = 12 and that

1 4 2 1/3 1/6
C−1
Y = = . (6)
12 2 4 1/6 1/3
This implies that

1/3 1/6 y1 − 4
(y − µY )
C−1
Y (y − µY ) = y1 − 4 y2 − 8 (7)
1/6 1/3 y2 − 8

y1 /3 + y2 /6 − 8/3
= y1 − 4 y2 − 8 (8)
y1 /6 + y2 /3 − 10/3
y12 y1 y2 16y1 20y2 y22 112
= + − − + + (9)
3 3 3 3 3 3
The PDF of Y is
1 −1
fY (y) = √ e−(y−µY ) CY (y−µY )/2
(10)
2π 12
1 2 2
=√ e−(y1 +y1 y2 −16y1 −20y2 +y2 +112)/6 (11)
48π 2

Since Y = X1 , X2 , the PDF of X1 and X2 is simply
1 2 2
fX1 ,X2 (x1 , x2 ) = fY1 ,Y2 (x1 , x2 ) = √ e−(x1 +x1 x2 −16x1 −20x2 +x2 +112)/6 (12)
48π 2
(c) We can observe directly from µX and CX that X1 is a Gaussian (4, 2) random variable. Thus,

X1 − 4 8−4
P [X1 > 8] = P > = Q(2) = 0.0228 (13)
2 2

We are given that X is a Gaussian random vector with
⎡ ⎤ ⎡ ⎤
4 4 −2 1
µX = ⎣8⎦ CX = ⎣−2 4 −2⎦ . (1)
6 1 −2 4
We are also given that Y = AX + b where

1 1/2 2/3 −4
A= b= . (2)
1 −1/2 2/3 −4
Since the two rows of A are linearly independent row vectors, A has rank 2. By Theorem 5.16,
Y is a Gaussian random vector. Given these facts, the various parts of this problem are just
straightforward calculations using Theorem 5.16.
222
(a) The expected value of Y is
⎡ ⎤
4
1 1/2 2/3 ⎣ ⎦ −4 8
µY = AµX + b = 8 + = . (3)
1 −1/2 2/3 −4 0
6
(b) The covariance matrix of Y is
CY = ACX A (4)
⎡ ⎤⎡ ⎤
4 −2 1 1 1
1 1/2 2/3 ⎣ 1 43 55
= −2 4 −2⎦ ⎣1/2 −1/2⎦ = . (5)
1 −1/2 2/3 9 55 103
1 −2 4 2/3 2/3
(c) Y has correlation matrix

1 43 55 8 1 619 55
RY = CY + µY µY = + 8 0 = (6)
9 55 103 0 9 55 103
(d) From µY , we see that E[Y2 ] = 0. From the covariance matrix CY , we learn that Y2 has
variance σ22 = CY (2, 2) = 103/9. Since Y2 is a Gaussian random variable,

1 Y2 1
P [−1 ≤ Y2 ≤ 1] = P − ≤ ≤ (7)
σ2 σ2 σ2

1 −1
=Φ −Φ (8)
σ2 σ2

1
= 2Φ −1 (9)
σ2

3
= 2Φ √ − 1 = 0.2325. (10)
103

This problem is just a special case of Theorem 5.16 with the matrix A replaced by the row vector
a and a 1 element vector b = b = 0. In this case, the vector Y becomes the scalar Y . The expected
value vector µY = [µY ] and the covariance “matrix” of Y is just the 1 × 1 matrix [σY2 ]. Directly
from Theorem 5.16, we can conclude that Y is a length 1 Gaussian random vector, which is just a
Gaussian random variable. In addition, µY = a µX and
Var[Y ] = CY = a CX a. (1)

From Definition 5.17, the n = 2 dimensional Gaussian vector X has PDF

1 1 −1
fX (x) = exp − (x − µX ) CX (x − µX ) (1)
2π[det (CX )]1/2 2
where CX has determinant
det (CX ) = σ12 σ22 − ρ2 σ12 σ22 = σ12 σ22 (1 − ρ2 ). (2)
223
Thus,
1 1
= . (3)
2π[det (CX )]1/2
2πσ1 σ2 1 − ρ2
Using the 2 × 2 matrix inverse formula
−1
a b 1 d −b
= , (4)
c d ad − bc −c a
we obtain 0 1 −ρ 1
1 σ22 −ρσ1 σ2 1 σ12
C−1
σ1 σ2
X = 2 2 = −ρ . (5)
σ1 σ2 (1 − ρ2 ) −ρσ1 σ2 2
σ1 1 − ρ2 σ1 σ2 1
σ22
Thus
0 −ρ 1
1
σ12 σ1 σ2 x1 − µ1
x1 − µ1 x2 − µ2 −ρ
1 σ1 σ2
1
σ22
x2 − µ2
− (x − µX ) C−1
X (x − µX ) = − (6)
2 2(1 − ρ2 )
⎡ ⎤

x1 −µ1
σ12
− ρ(xσ21−µ 2)
x2 − µ2 ⎣ ⎦
σ2
x1 − µ1
− ρ(xσ11−µ
σ2
1)
+ x2σ−µ
2
2
2
=− (7)
2(1 − ρ2 )
(x1 −µ1 )2 2ρ(x1 −µ1 )(x2 −µ2 ) (x2 −µ2 )2
σ12
− σ1 σ2 + σ22
=− . (8)
2(1 − ρ2 )
Combining Equations (1), (3), and (8), we see that

⎡ (x −µ )2 2ρ(x1 −µ1 )(x2 −µ2 ) (x2 −µ2 )2
⎤
1
1
σ12
1
− + σ22
exp ⎣− ⎦,
σ1 σ2
fX (x) = (9)
2πσ1 σ2 1 − ρ2 2(1 − ρ2 )
which is the bivariate Gaussian PDF in Definition 4.17.

Since
X I
W= = X = DX (1)
Y A
Suppose that X Gaussian (0, I) random vector. By Theorem 5.13, µW = 0 and CW = DD . The
matrix D is (m + n) × n and has rank n. That is, the rows of D are dependent and there exists
a vector y such that y D = 0. This implies y DD y = 0. Hence det(CW ) = 0 and C−1W does not
exist. Hence W is not a Gaussian random vector.
The point to keep in mind is that the definition of a Gaussian random vector does not permit
a component random variable to be a deterministic linear combination of other components.
224
(a) From Theorem 5.13, Y has covariance matrix
CY = QCX Q (1)

cos θ − sin θ σ12 0 cos θ sin θ
= (2)
sin θ cos θ 0 σ22 − sin θ cos θ
2
σ cos2 θ + σ22 sin2 θ (σ12 − σ22 ) sin θ cos θ
= 12 . (3)
(σ1 − σ22 ) sin θ cos θ σ12 sin2 θ + σ22 cos2 θ
We conclude that Y1 and Y2 have covariance
Cov [Y1 , Y2 ] = CY (1, 2) = (σ12 − σ22 ) sin θ cos θ. (4)
Since Y1 and Y2 are jointly Gaussian, they are independent if and only if Cov[Y1 , Y2 ] =
0. Thus, Y1 and Y2 are independent for all θ if and only if σ12 = σ22 . In this case, when
the joint PDF fX (x) is symmetric in x1 and x2 . In terms of polar coordinates, the PDF
fX (x) = fX1 ,X2 (x1 , x2 ) depends on r = x21 + x22 but for a given r, is constant for all
−1
φ = tan (x2 /x1 ). The transformation of X to Y is just a rotation of the coordinate system
by θ preserves this circular symmetry.
(b) If σ22 > σ12 , then Y1 and Y2 are independent if and only if sin θ cos θ = 0. This occurs in the
following cases:
• θ = 0: Y1 = X1 and Y2 = X2
• θ = π/2: Y1 = −X2 and Y2 = −X1
• θ = π: Y1 = −X1 and Y2 = −X2
• θ = −π/2: Y1 = X2 and Y2 = X1
In all four cases, Y1 and Y2 are just relabeled versions, possibly with sign changes, of X1 and
X2 . In these cases, Y1 and Y2 are independent because X1 and X2 are independent. For
other values of θ, each Yi is a linear combination of both X1 and X2 . This mixing results in
correlation between Y1 and Y2 .

The difficulty of this problem is overrated since its a pretty simple application of Problem 5.7.6. In
particular, $
cos θ − sin θ $$ 1 1 −1
Q= =√ . (1)
sin θ cos θ $θ=45◦ 2 1 1
Since X = QY, we know from Theorem 5.16 that X is Gaussian with covariance matrix
CX = QCY Q (2)

1 1 −1 1 + ρ 0 1 1 1
=√ √ (3)
2 1 1 0 1−ρ 2 −1 1

1 1 + ρ −(1 − ρ) 1 1
= (4)
2 1+ρ 1−ρ −1 1

1 ρ
= . (5)
ρ 1
225
As given in the
problem
statement, we define the m-dimensional vector X, the n-dimensional vector
X
Y and W = . Note that W has expected value
Y

X E [X] µX
µW = E [W] = E = = . (1)
Y E [Y] µY
The covariance matrix of W is

CW = E (W − µW )(W − µW ) (2)

X − µX

=E (X − µX ) (Y − µY ) (3)
Y − µY

E [(X − µX )(X − µX ) ] E [(X − µX )(Y − µY ) ]
= (4)
E [(Y − µY )(X − µX ) ] E [(Y − µY )(Y − µY ) ]

CX CXY
= . (5)
CYX CY
The assumption that X and Y are independent implies that

CXY = E (X − µX )(Y − µY ) = (E [(X − µX )] E (Y − µY ) = 0. (6)
This also implies CYX = CXY = 0 . Thus

CX 0
CW = . (7)
0 CY
(a) If you are familiar with the Gram-Schmidt procedure, the argument is that applying Gram-
Schmidt to the rows of A yields m orthogonal row vectors. It is then possible to augment
those vectors with an additional n − m orothogonal vectors. Those orthogonal vectors would
be the rows of Ã.
An alternate argument is that since A has rank m the nullspace of A, i.e., the set of all
vectors y such that Ay = 0 has dimension n − m. We can choose any n − m linearly
independent vectors y1 , y2 , . . . , yn−m in the nullspace A. We then define Ã to have columns
y1 , y2 , . . . , yn−m . It follows that AÃ = 0.
(b) To use Theorem 5.16 for the case m = n to show

Y A
Ȳ = = X. (1)
Ŷ Â
is a Gaussian random vector requires us to show that

A A
Ā = = (2)
Â ÃC−1
X
226
is a rank n matrix. To prove this fact, we will suppose there exists w such that Āw = 0,
and then show that w is a zero vector. Since A and Ã together have n linearly independent
rows, we can write the row vector w as a linear combination of the rows of A and Ã. That
is, for some v and ṽ,
w = vt A + ṽ Ã. (3)
The condition Āw = 0 implies
0
A
A v + Ã ṽ = . (4)
ÃC−1
X 0
This implies
AA v + AÃ ṽ = 0 (5)

ÃC−1 −1
X Av + ÃCX Ã ṽ = 0 (6)
Since AÃ = 0, Equation (5) implies that AA v = 0. Since A is rank m, AA is an m × m
rank m matrix. It follows that v = 0. We can then conclude from Equation (6) that
ÃC−1
X Ã ṽ = 0. (7)
This would imply that ṽ ÃC−1 −1

X Ã ṽ = 0. Since CX is invertible, this would imply that

Ã ṽ = 0. Since the rows of Ã are linearly independent, it must be that ṽ = 0. Thus Ā is
full rank and Ȳ is a Gaussian random vector.
(c) We note that By Theorem 5.16, the Gaussian vector Ȳ = ĀX has covariance matrix
C̄ = ĀCX Ā . (8)
Since (C−1 −1
X ) = CX ,
Ā = A (ÃC−1
X ) = A C−1
X Ã . (9)
Applying this result to Equation (8) yields

A −1
ACX −1
ACX A AÃ
C̄ = CX A CX Ã = A CX Ã = . (10)
ÃC−1
X Ã ÃA ÃC−1
X Ã
Since ÃA = 0,

ACX A 0 CY 0
C̄ = = . (11)
0 ÃC−1
X Ã 0 CŶ
We see that C̄ is block diagonal covariance matrix. From the claim of Problem 5.7.8, we can
conclude that Y and Ŷ are independent Gaussian random vectors.

We can use Theorem 5.16 since the scalar Y is also a 1-dimensional vector. To do so, we write

Y = 1/3 1/3 1/3 X = AX. (1)
By Theorem 5.16, Y is a Gaussian vector with expected value
E [Y ] = AµX = (E [X1 ] + E [X2 ] + E [X3 ])/3 = (4 + 8 + 6)/3 = 6 (2)
227
and covariance matrix
CY = Var[Y ] = ACX A (3)
⎡ ⎤⎡ ⎤
4 −2 1
1/3
⎣ ⎦ ⎣ 2
= 1/3 1/3 1/3 −2 4 −2 1/3⎦ = (4)
3
1 −2 4 1/3

Thus Y is a Gaussian (6, 2/3) random variable, implying
0 1
Y −6 4−6 √ √
P [Y > 4] = P > = 1 − Φ(− 6) = Φ( 6) = 0.9928 (5)
2/3 2/3
(a) The covariance matrix CX has Var[Xi ] = 25 for each diagonal entry. For i = j, the i, jth
entry of CX is
[CX ]ij = ρXi Xj Var[Xi ] Var[Xj ] = (0.8)(25) = 20 (1)
The covariance matrix of X is a 10 × 10 matrix of the form
⎡ ⎤
25 20 · · · 20
⎢ .. ⎥
⎢20 25 . . . .⎥
CX = ⎢ ⎢ .. . . ..
⎥.
⎥ (2)
⎣. . . 20⎦
20 · · · 20 25
(b) We observe that

Y = 1/10 1/10 · · · 1/10 X = AX (3)
Since Y is the average of 10 iid random variables, E[Y ] = E[Xi ] = 5. Since Y is a scalar, the
1 × 1 covariance matrix CY = Var[Y ]. By Theorem 5.13, the variance of Y is
Var[Y ] = CY = ACX A = 20.5 (4)
Since Y is Gaussian,

Y −5 25 − 20.5
P [Y ≤ 25] = P √ ≤ √ = Φ(0.9939) = 0.8399. (5)
20.5 20.5

Under the model of Quiz 5.8, the temperature on day i and on day j have covariance
36
Cov [Ti , Tj ] = CT [i − j] = (1)
1 + |i − j|

From this model, the vector T = T1 · · · T31 has covariance matrix
⎡ ⎤
CT [0] CT [1] ··· CT [30]
⎢ .. .. ⎥
⎢ CT [1] CT [0] . . ⎥
⎢
CT = ⎢ . ⎥. (2)
.. .. ⎥
⎣ .. . . CT [1] ⎦
CT [30] ··· CT [1] CT [0]
228
If you have read the solution to Quiz 5.8, you know that CT is a symmetric Toeplitz matrix and
that Matlab has a toeplitz function to generate Toeplitz matrices. Using the toeplitz function
to generate the covariance matrix, it is easy to use gaussvector to generate samples of the random
vector T. Here is the code for estimating P [A] using m samples.
function p=julytemp583(m); julytemp583(100000)
c=36./(1+(0:30)); ans =
CT=toeplitz(c); 0.0684
mu=80*ones(31,1); >> julytemp583(100000)
T=gaussvector(mu,CT,m); ans =
Y=sum(T)/31; 0.0706
Tmin=min(T); >> julytemp583(100000)
p=sum((Tmin>=72) & (Y <= 82))/m; ans =
0.0714
>> julytemp583(100000)
ans =
0.0701
We see from repeated experiments with m = 100,000 trials that P [A] ≈ 0.07.

The covariance matrix CX has Var[Xi ] = 25 for each diagonal entry. For i = j, the i, jth entry of
CX is
[CX ]ij = ρXi Xj Var[Xi ] Var[Xj ] = (0.8)(25) = 20 (1)
The covariance matrix of X is a 10 × 10 matrix of the form
⎡ ⎤
25 20 · · · 20
⎢ .⎥
⎢20 25 . . . .. ⎥
⎢
CX = ⎢ . . ⎥. (2)
. ⎥
⎣.. . . . . 20⎦
20 · · · 20 25
A program to estimate P [W ≤ 25] uses gaussvector to generate m sample vector of race times X.
In the program sailboats.m, X is an 10 × m matrix such that each column of X is a vector of race
times. In addition min(X) is a row vector indicating the fastest time in each race.
function p=sailboats(w,m) >> sailboats(25,10000)
%Usage: p=sailboats(f,m) ans =
%In Problem 5.8.4, W is the 0.0827
%winning time in a 10 boat race. >> sailboats(25,100000)
%We use m trials to estimate ans =
%P[W<=w] 0.0801
CX=(5*eye(10))+(20*ones(10,10)); >> sailboats(25,100000)
mu=35*ones(10,1); ans =
X=gaussvector(mu,CX,m); 0.0803
W=min(X); >> sailboats(25,100000)
p=sum(W<=w)/m; ans =
0.0798
We see from repeated experiments with m = 100,000 trials that P [W ≤ 25] ≈ 0.08.

When we built poissonrv.m, we went to some trouble to be able to generate m iid samples at
once. In this problem, each Poisson random variable that we generate has an expected value that
229
is different from that of any other Poisson random variables. Thus, we must generate the daily
jackpots sequentially. Here is a simple program for this purpose.
function jackpot=lottery1(jstart,M,D)
%Usage: function j=lottery1(jstart,M,D)
%Perform M trials of the D day lottery
%of Problem 5.5.5 and initial jackpot jstart
jackpot=zeros(M,1);
for m=1:M,
disp(’trm)
jackpot(m)=jstart;
for d=1:D,
jackpot(m)=jackpot(m)+(0.5*poissonrv(jackpot(m),1));
end
end
The main problem with lottery1 is that it will run very slowly. Each call to poissonrv generates
an entire Poisson PMF PX (x) for x = 0, 1, .
. . , xmax where xmax ≥ 2 · 106 . This is slow in several
ways. First, we repeating the calculation of xj=1 max
log j with each call to poissonrv. Second, each
call to poissonrv asks for a Poisson sample value with expected value α > 1 · 106 . In these cases,
for small values of x, PX (x) = αx e−αx /x! is so small that it is less than the smallest nonzero number
that Matlab can store!
To speed up the simulation, we have written a program bigpoissonrv which generates Poisson
(α) samples for large α. The program makes an approximation that for a Poisson (α) random vari-
√ √
able X, PX (x) ≈ 0 for |x − α| > 6 α. Since X has standard deviation α, we are assuming that X
cannot be more than six standard deviations away from its mean value. The error in this approxi-
mation is very small. In fact, for a Poisson (a) random variable, the program poissonsigma(a,k)
√
calculates the error P [|X − a| > k a]. Here is poissonsigma.m and some simple calculations:
function err=poissonsigma(a,k); >> poissonsigma(1,6)
xmin=max(0,floor(a-k*sqrt(a))); ans =
xmax=a+ceil(k*sqrt(a)); 1.0249e-005
sx=xmin:xmax; >> poissonsigma(10,6)
logfacts =cumsum([0,log(1:xmax)]); ans =
%logfacts includes 0 in case xmin=0 2.5100e-007
%Now we extract needed values: >> poissonsigma(100,6)
logfacts=logfacts(sx+1); ans =
%pmf(i,:) is a Poisson a(i) PMF 1.2620e-008
% from xmin to xmax >> poissonsigma(1000,6)
pmf=exp(-a+ (log(a)*sx)-(logfacts)); ans =
err=1-sum(pmf); 2.6777e-009
>> poissonsigma(10000,6)
ans =
1.8081e-009
>> poissonsigma(100000,6)
ans =
-1.6383e-010
The error reported by poissonsigma(a,k) should always be positive. In fact, we observe
negative errors for very large a. For large α and x, numerical calculation of PX (x) = αx e−α /x!
is tricky because we are taking ratios of very large numbers. In fact, for α = x = 1,000,000,
Matlab calculation of αx and x! will report infinity while e−α will evaluate as zero. Our method
230

x
of calculating the Poisson (α) PMF is to use the fact that ln x! = j=1 ln j to calculate
⎛ ⎞

x
exp (ln PX (x)) = exp ⎝x ln α − α − ln j ⎠ . (1)
j=1
This method works reasonably well except that the calculation of the logarithm has finite precision.
The consequence is that the calculated sum over the PMF can vary from 1 by a very small amount,
on the order of 10−7 in our experiments. In our problem, the error is inconsequential, however, one
should keep in mind that this may not be the case in other other experiments using large Poisson
random variables. In any case, we can conclude that within the accuracy of Matlab’s simulated
experiments, the approximations to be used by bigpoissonrv are not significant.
The other feature

of bigpoissonrv is that for a vector alpha corresponding to expected values
α1 · · · αm , bigpoissonrv returns a vector X such that X(i) is a Poisson alpha(i) sample.
The work of calculating the sum of logarithms is done only once for all calculated samples. The
result is a significant savings in cpu time as long as the values of alpha are reasonably close to each
other.
function x=bigpoissonrv(alpha)
%for vector alpha, returns a vector x such that
% x(i) is a Poisson (alpha(i)) rv
%set up Poisson CDF from xmin to xmax for each alpha(i)
alpha=alpha(:);
amin=min(alpha(:));
amax=max(alpha(:));
%Assume Poisson PMF is negligible +-6 sigma from the average
xmin=max(0,floor(amin-6*sqrt(amax)));
xmax=amax+ceil(6*sqrt(amax));%set max range
sx=xmin:xmax;
%Now we include the basic code of poissonpmf (but starting at xmin)
logfacts =cumsum([0,log(1:xmax)]); %include 0 in case xmin=0
logfacts=logfacts(sx+1); %extract needed values
%pmf(i,:) is a Poisson alpha(i) PMF from xmin to xmax
pmf=exp(-alpha*ones(size(sx))+ ...
(log(alpha)*sx)-(ones(size(alpha))*logfacts));
cdf=cumsum(pmf,2); %each row is a cdf
x=(xmin-1)+sum((rand(size(alpha))*ones(size(sx)))<=cdf,2);
Finally, given bigpoissonrv, we can write a short program lottery that simulates trials of the
jackpot experiment. Ideally, we would like to use lottery to perform m = 1,000 trials in a single
pass. In general, Matlab is more efficient when calculations are executed in parallel using vectors.
√
However, in bigpoissonrv, the matrix pmf will have m rows and at least 12 α = 12,000 columns.
For m more than several hundred, Matlab running on my laptop reported an “Out of Memory”
error. Thus, we wrote the program lottery to perform M trials at once and to repeat that N times.
The output is an M × N matrix where each i, j entry is a sample jackpot after seven days.
231
function jackpot=lottery(jstart,M,N,D)
%Usage: function j=lottery(jstart,M,N,D)
%Perform M trials of the D day lottery
%of Problem 5.5.5 and initial jackpot jstart
jackpot=zeros(M,N);
for n=1:N,
jackpot(:,n)=jstart*ones(M,1);
for d=1:D,
disp(d);
jackpot(:,n)=jackpot(:,n)+(0.5*bigpoissonrv(jackpot(:,n)));
end
end
Executing J=lottery(1e6,200,10,7) generates a matrix J of 2,000 sample jackpots. The com-

mand hist(J(:),50) generates a histogram of the values with 50 bins. An example is shown
here:
150
100
Frequency
50
0
1.7076 1.7078 1.708 1.7082 1.7084 1.7086 1.7088 1.709 1.7092 1.7094 1.7096
J 7
x 10
If you go back and solve Problem 5.5.5, you will see that the jackpot J has expected value
E[J] = (3/2)7 × 106 = 1.70859 × 107 dollars. Thus it is not surprising that the histogram is
centered around a jackpot of 1.708 × 107 dollars. If we did more trials, and used more histogram
bins, the histogram would appear to converge to the shape of a Gaussian PDF. This fact is explored
in Chapter 6.
232

The random variable X33 is a Bernoulli random variable that indicates the result of flip 33. The
PMF of X33 is ⎧
⎨ 1−p x=0
PX33 (x) = p x=1 (1)
⎩
0 otherwise
Note that each Xi has expected value E[X] = p and variance Var[X] = p(1 − p). The random
variable Y = X1 + · · · + X100 is the number of heads in 100 coin flips. Hence, Y has the binomial
PMF 100 y
y p (1 − p)
100−y y = 0, 1, . . . , 100
PY (y) = (2)
0 otherwise
Since the Xi are independent, by Theorems 6.1 and 6.3, the mean and variance of Y are
E [Y ] = 100E [X] = 100p Var[Y ] = 100 Var[X] = 100p(1 − p) (3)

Let Y = X1 − X2 .
(a) Since Y = X1 + (−X2 ), Theorem 6.1 says that the expected value of the difference is
E [Y ] = E [X1 ] + E [−X2 ] = E [X] − E [X] = 0 (1)
(b) By Theorem 6.2, the variance of the difference is

Var[Y ] = Var[X1 ] + Var[−X2 ] = 2 Var[X] (2)
(a) The PMF of N1 , the number of phone calls needed to obtain the correct answer, can be
determined by observing that if the correct answer is given on the nth call, then the previous
n − 1 calls must have given wrong answers so that

(3/4)n−1 (1/4) n = 1, 2, . . .
PN1 (n) = (1)
0 otherwise
(b) N1 is a geometric random variable with parameter p = 1/4. In Theorem 2.5, the mean of a
geometric random variable is found to be 1/p. For our case, E[N1 ] = 4.
(c) Using the same logic as in part (a) we recognize that in order for n to be the fourth correct
answer, that the previous n − 1 calls must have contained exactly 3 correct answers and that
the fourth correct answer arrived on the n-th call. This is described by a Pascal random
variable. n−1 n−4 (1/4)4 n = 4, 5, . . .
PN4 (n4 ) = 3 (3/4) (2)
0 otherwise
(d) Using the hint given in the problem statement we can find the mean of N4 by summing up
the means of the 4 identically distributed geometric random variables each with mean 4. This
gives E[N4 ] = 4E[N1 ] = 16.
233
We can solve this problem using Theorem 6.2 which says that
Var[W ] = Var[X] + Var[Y ] + 2 Cov [X, Y ] (1)
The first two moments of X are

' 1 ' 1−x ' 1
E [X] = 2x dy dx = 2x(1 − x) dx = 1/3 (2)
0 0 0
' 1 ' 1−x ' 1
2
E X = 2
2x dy dx = 2x2 (1 − x) dx = 1/6 (3)
0 0 0
(4)
Thus the variance of X is Var[X] = E[X 2 ] − (E[X])2 = 1/18. By symmetry, it should be apparent
that E[Y ] = E[X] = 1/3 and Var[Y ] = Var[X] = 1/18. To find the covariance, we first find the
correlation
' 1 ' 1−x ' 1
E [XY ] = 2xy dy dx = x(1 − x)2 dx = 1/12 (5)
0 0 0
The covariance is
Cov [X, Y ] = E [XY ] − E [X] E [Y ] = 1/12 − (1/3)2 = −1/36 (6)
Finally, the variance of the sum W = X + Y is
Var[W ] = Var[X] + Var[Y ] − 2 Cov [X, Y ] = 2/18 − 2/36 = 1/18 (7)
For this specific problem, it’s arguable whether it would easier to find Var[W ] by first deriving the
CDF and PDF of W . In particular, for 0 ≤ w ≤ 1,
' w ' w−x ' w
FW (w) = P [X + Y ≤ w] = 2 dy dx = 2(w − x) dx = w2 (8)
0 0 0
Hence, by taking the derivative of the CDF, the PDF of W is

2w 0 ≤ w ≤ 1
fW (w) = (9)
0 otherwise
From the PDF, the first and second moments of W are

' 1 '
1
E [W ] = 2w2 dw = 2/3 E W2 = 2w3 dw = 1/2 (10)
0 0
The variance of W is Var[W ] = E[W 2 ] − (E[W ])2 = 1/18. Not surprisingly, we get the same answer
both ways.

This problem should be in either Chapter 10 or Chapter 11.
Since each Xi has zero mean, the mean of Yn is
E [Yn ] = E [Xn + Xn−1 + Xn−2 ] /3 = 0 (1)
234
Since Yn has zero mean, the variance of Yn is

Var[Yn ] = E Yn2 (2)
1
= E (Xn + Xn−1 + Xn−2 )2 (3)
9
1
= E Xn2 + Xn−12 2
+ Xn−2 + 2Xn Xn−1 + 2Xn Xn−2 + 2Xn−1 Xn−2 (4)
9
1 4
= (1 + 1 + 1 + 2/4 + 0 + 2/4) = (5)
9 9


2 0≤x≤y≤1
fX,Y (x, y) = (1)
0 otherwise
We wish to find the PDF of W where W = X + Y . First we find the CDF of W , FW (w), but we
must realize that the CDF will require different integrations for different values of w.
Y
Y=X
w For values of 0 ≤ w ≤ 1 we look to integrate the shaded area in the figure
to the right.
Area of
Integration
' w ' w−x
2 w2
FW (w) = 2 dy dx = (2)
0 x 2
X+Y=w
X
w
For values of w in the region 1 ≤ w ≤ 2 we look to integrate over the
Y shaded region in the graph to the right. From the graph we see that
w we can integrate with respect to x first, ranging y from 0 to w/2,
Y=X thereby covering the lower right triangle of the shaded region and
leaving the upper trapezoid, which is accounted for in the second
Area of term of the following expression:
Integration
' w ' y ' 1 ' w−y
2
X+Y=w
FW (w) = 2 dx dy + 2 dx dy (3)
w
0 0 2
0
X
w w2
= 2w − 1 − (4)
2
Putting all the parts together gives the CDF FW (w) and (by taking the derivative) the PDF
fW (w).
⎧
⎪ 0 w<0 ⎧
⎪
⎨ w2 ⎨ w 0≤w≤1
0≤w≤1
FW (w) = 2
w2 fW (w) = 2−w 1≤w ≤2 (5)
⎪ ⎩
⎩ 2w − 1 − 2 1 ≤ w ≤ 2
⎪ 0 otherwise
1 w>2


1 0 ≤ x, y ≤ 1
fX,Y (x, y) = (1)
0 otherwise
235
Proceeding as in Problem 6.2.1, we must first find FW (w) by integrating over the square defined
by 0 ≤ x, y ≤ 1. Again we are forced to find FW (w) in parts as we did in Problem 6.2.1 resulting
in the following integrals for their appropriate regions. For 0 ≤ w ≤ 1,
' w ' w−x
FW (w) = dx dy = w2 /2 (2)
0 0
For 1 ≤ w ≤ 2,
' w−1 ' 1 ' 1 ' w−y
FW (w) = dx dy + dx dy = 2w − 1 − w2 /2 (3)
0 0 w−1 0
The complete CDF FW (w) is shown below along with the corresponding PDF fW (w) = dFW (w)/dw.
⎧
⎪ 0 w<0 ⎧
⎪
⎨ 2 ⎨ w 0≤w≤1
w /2 0≤w≤1
FW (w) = fW (w) = 2−w 1≤w ≤2 (4)
⎪
⎪ 2w − 1 − w2 /2 1 ≤ w ≤ 2 ⎩
⎩ 0 otherwise
1 otherwise

By using Theorem 6.5, we can find the PDF of W = X + Y by convolving the two exponential
distributions. For µ = λ,
' ∞
fW (w) = fX (x) fY (w − x) dx (1)
−∞
' w
= λe−λx µe−µ(w−x) dx (2)
0
' w
−µw
= λµe e−(λ−µ)x dx (3)
0
λµ −µw
λ−µ e − e−λw w ≥ 0
= (4)
0 otherwise
When µ = λ, the previous derivation is invalid because of the denominator term λ − µ. For µ = λ,
we have
' ∞
fW (w) = fX (x) fY (w − x) dx (5)
−∞
' w
= λe−λx λe−λ(w−x) dx (6)
0
' w
= λ2 e−λw dx (7)
2 −λw 0
λ we w≥0
= (8)
0 otherwise
Note that when µ = λ, W is the sum of two iid exponential random variables and has a second
order Erlang PDF.
236
In this problem, X and Y have joint PDF

8xy 0 ≤ y ≤ x ≤ 1
fX,Y (x, y) = (1)
0 otherwise
(∞
We can find the PDF of W using Theorem 6.4: fW (w) = −∞ fX,Y (x, w − x) dx. The only tricky
part remaining is to determine the limits of the integration. First, for w < 0, fW (w) = 0. The
two remaining cases are shown in the accompanying figure. The shaded area shows where the joint
PDF fX,Y (x, y) is nonzero. The diagonal lines depict y = w − x as a function of x. The intersection
of the diagonal line and the shaded area define our limits of integration.
For 0 ≤ w ≤ 1,
' w
y fW (w) = 8x(w − x) dx (2)
w/2
2 $w
= 4wx2 − 8x3 /3$w/2 = 2w3 /3 (3)
w
1<w<2
1 For 1 ≤ w ≤ 2,
w
0<w<1
' 1
fW (w) = 8x(w − x) dx (4)
x w/2
w 1 w 2 $1
= 4wx2 − 8x3 /3$w/2 (5)
= 4w − 8/3 − 2w3 /3 (6)
Since X + Y ≤ 2, fW (w) = 0 for w > 2. Hence the complete expression for the PDF of W is
⎧
⎨ 2w3 /3 0≤w≤1
fW (w) = 4w − 8/3 − 2w3 /3 1 ≤ w ≤ 2 (7)
⎩
0 otherwise

We first find the CDF of W following the same procedure as in the proof of Theorem 6.4.
' ∞ ' y+w
FW (w) = P [X ≤ Y + w] = fX,Y (x, y) dx dy (1)
−∞ −∞
By taking the derivative with respect to w, we obtain

' ∞ ' y+w
dFW (w) d
fW (w) = = fX,Y (x, y) dx dy (2)
dw −∞ dw −∞
' ∞
= fX,Y (w + y, y) dy (3)
−∞
With the variable substitution y = x − w, we have dy = dx and

' ∞
fW (w) = fX,Y (x, x − w) dx (4)
−∞
237
The random variables K and J have PMFs
j −α
α e
j = 0, 1, 2, . . . β k e−β
j! k = 0, 1, 2, . . .
PJ (j) = PK (k) = k! (1)
For n ≥ 0, we can find the PMF of N = J + K via

∞

P [N = n] = P [J = n − k, K = k] (2)
k=−∞
Since J and K are independent, non-negative random variables,

n
P [N = n] = PJ (n − k) PK (k) (3)
k=0
n
αn−k e−α β k e−β
= (4)
(n − k)! k!
k=0
n−k k
(α + β)n e−(α+β)
n
n! α β
= (5)
n! k!(n − k)! α + β α+β
k=0
1
The marked sum above equals 1 because it is the sum of a binomial PMF over all possible values.
The PMF of N is the Poisson PMF

(α+β)n e−(α+β)
PN (n) = n! n = 0, 1, 2, . . . (6)
0 otherwise

For a constant a > 0, a zero mean Laplace random variable X has PDF
a −a|x|
fX (x) = e −∞<x<∞ (1)
2
The moment generating function of X is
' ' ∞
a 0
a
φX (s) = E esX = esx eax dx + esx e−ax dx (2)
2 −∞ 2 0
$0 $∞
a e(s+a)x $$ a e(s−a)x $$
= $ + $ (3)
2 s+a $ 2 s−a $
−∞
0
a 1 1
= − (4)
2 s+a s−a
a2
= 2 (5)
a − s2
238
(a) By summing across the rows of the table, we see that J has PMF

0.6 j = −2
PJ (j) = (1)
0.4 j = −1
The MGF of J is φJ (s) = E[esJ ] = 0.6e−2s + 0.4e−s .
(b) Summing down the columns of the table, we see that K has PMF
⎧
⎨ 0.7 k = −1
PK (k) = 0.2 k = 0 (2)
⎩
0.1 k = 1
The MGF of K is φK (s) = 0.7e−s + 0.2 + 0.1es .
(c) To find the PMF of M = J + K, it is easist to annotate each entry in the table with the
coresponding value of M :
PJ,K (j, k) k = −1 k=0 k=1

j = −2 0.42(M = −3) 0.12(M = −2) 0.06(M = −1) (3)
j = −1 0.28(M = −2) 0.08(M = −1) 0.04(M = 0)
We obtain PM (m) by summing over all j, k such that j + k = m, yielding

⎧
⎪
⎪ 0.42 m = −3
⎨
0.40 m = −2
PM (m) = (4)
⎪
⎪ 0.14 m = −1
⎩
0.04 m = 0
(d) One way to solve this problem, is to find the MGF φM (s) and then take four derivatives.
Sometimes its better to just work with definition of E[M 4 ]:

E M4 = PM (m) m4 (5)
m
= 0.42(−3)4 + 0.40(−2)4 + 0.14(−1)4 + 0.04(0)4 = 40.434 (6)
As best I can tell, the prupose of this problem is to check that you know when not to use the
methods in this chapter.

We find the MGF by calculating E[esX ] from the PDF fX (x).
' b
1 ebs − eas
φX (s) = E esX = esX dx = (1)
a b−a s(b − a)
Now to find the first moment, we evaluate the derivative of φX (s) at s = 0.
$ $
dφX (s) $$ s bebs − aeas − ebs − eas $$
E [X] = = $ (2)
ds $s=0 (b − a)s2 $
s=0
239
Direct evaluation of the above expression at s = 0 yields 0/0 so we must apply l’Hôpital’s rule and
differentiate the numerator and denominator.

bebs − aeas + s b2 ebs − a2 eas − bebs − aeas
E [X] = lim (3)
s→0 2(b − a)s
b2 ebs − a2 eas b+a
= lim = (4)
s→0 2(b − a) 2
To find the second moment of X, we first find that the second derivative of φX (s) is

d2 φX (s) s2 b2 ebs − a2 eas − 2s bebs − aeas + 2 bebs − aeas
= (5)
ds2 (b − a)s3
Substituting s = 0 will yield 0/0 so once again we apply l’Hôpital’s rule and differentiate the
numerator and denominator.

2 d2 φX (s) s2 b3 ebs − a3 eas
E X = lim = lim (6)
s→0 ds2 s→0 3(b − a)s2
b3 − a3
= = (b2 + ab + a2 )/3 (7)
3(b − a)
In this case, it is probably simpler to find these moments without using the MGF.

2 2
Using the moment generating function of X, φX (s) = eσ s /2 . We can find the nth moment of X,
E[X n ] by taking the nth derivative of φX (s) and setting s = 0.
$
2 2 $
E [X] = σ 2 seσ s /2 $ =0 (1)
2
s=0
$
2 2 2 2 $
E X = σ 2 eσ s /2 + σ 4 s2 eσ s /2 $ = σ2. (2)
s=0
Continuing in this manner we find that

2 2 $$
E X 3 = 3σ 4 s + σ 6 s3 eσ s /2 $ =0 (3)
2 2 $$
s=0

E X 4 = 3σ 4 + 6σ 6 s2 + σ 8 s4 eσ s /2 $ = 3σ 4 . (4)
s=0
To calculate the moments of Y , we define Y = X + µ so that Y is Gaussian (µ, σ). In this case
the second moment of Y is

E Y 2 = E (X + µ)2 = E X 2 + 2µX + µ2 = σ 2 + µ2 . (5)
Similarly, the third moment of Y is

E Y 3 = E (X + µ)3 (6)

= E X 3 + 3µX 2 + 3µ2 X + µ3 = 3µσ 2 + µ3 . (7)
Finally, the fourth moment of Y is

E Y 4 = E (X + µ)4 (8)

= E X 4 + 4µX 3 + 6µ2 X 2 + 4µ3 X + µ4 (9)
= 3σ 4 + 6µ2 σ 2 + µ4 . (10)
240
The PMF of K is
1/n k = 1, 2, . . . , n
PK (k) = (1)
0 otherwise
The corresponding MGF of K is
1 s
φK (s) = E esK = e + e2 s + · · · + ens (2)
n
es
= 1 + es + e2s + · · · + e(n−1)s (3)
n
es (ens − 1)
= (4)
n(es − 1)
We can evaluate the moments of K by taking derivatives of the MGF. Some algebra will show that
dφK (s) ne(n+2)s − (n + 1)e(n+1)s + es

= (5)
ds n(es − 1)2
Evaluating dφK (s)/ds at s = 0 yields 0/0. Hence, we apply l’Hôpital’s rule twice (by twice
differentiating the numerator and twice differentiating the denominator) when we write
$
dφK (s) $$ n(n + 2)e(n+2)s − (n + 1)2 e(n+1)s + es
= lim (6)
ds $s=0 s→0 2n(es − 1)
n(n + 2)2 e(n+2)s − (n + 1)3 e(n+1)s + es
= lim = (n + 1)/2 (7)
s→0 2nes
A significant amount of algebra will show that the second derivative of the MGF is
d2 φK (s) n2 e(n+3)s − (2n2 + 2n − 1)e(n+2)s + (n + 1)2 e(n+1)s − e2s − es

= (8)
ds2 n(es − 1)3
Evaluating d2 φK (s)/ds2 at s = 0 yields 0/0. Because (es − 1)3 appears in the denominator, we
need to use l’Hôpital’s rule three times to obtain our answer.
$
d2 φK (s) $$ n2 (n + 3)3 e(n+3)s − (2n2 + 2n − 1)(n + 2)3 e(n+2)s + (n + 1)5 − 8e2s − es
$ = lim (9)
ds2 s=0 s→0 6nes
n2 (n + 3)3 − (2n2 + 2n − 1)(n + 2)3 + (n + 1)5 − 9
= (10)
6n
= (2n + 1)(n + 1)/6 (11)
We can use these results to derive two well known results. We observe that we can directly use the
PMF PK (k) to calculate the moments
1 1
n n
E [K] = k E K2 = k2 (12)
n n
k=1 k=1
Using the answers we found for E[K] and E[K 2 ], we have the formulas

n
n(n + 1)
n
n(n + 1)(2n + 1)
k= k2 = (13)
2 6
k=1 k=1
241
N is a binomial (n = 100, p = 0.4) random variable. M is a binomial (n = 50, p = 0.4) random
variable. Thus N is the sum of 100 independent Bernoulli (p = 0.4) and M is the sum of 50
independent Bernoulli (p = 0.4) random variables. Since M and N are independent, L = M + N
is the sum of 150 independent Bernoulli (p = 0.4) random variables. Hence L is a binomial
n = 150, p = 0.4) and has PMF

150
PL (l) = (0.4)l (0.6)150−l . (1)
l

Random variable Y has the moment generating function φY (s) = 1/(1 − s). Random variable V
has the moment generating function φV (s) = 1/(1 − s)4 . Y and V are independent. W = Y + V .
(a) From Table 6.1, Y is an exponential (λ = 1) random variable. For an exponential (λ) random
variable, Example 6.5 derives the moments of the exponential random variable. For λ = 1,
the moments of Y are

E [Y ] = 1, E Y 2 = 2, E Y 3 = 3! = 6. (1)
(b) Since Y and V are independent, W = Y + V has MGF

4 5
1 1 1
φW (s) = φY (s)φV (s) = = . (2)
1−s 1−s 1−s
W is the sum of five independent exponential (λ = 1) random variables X1 , . . . , X5 . (That

is, W is an Erlang (n = 5, λ = 1) random variable.) Each Xi has expected value E[X] = 1
and variance Var[X] = 1. From Theorem 6.1 and Theorem 6.3,
E [W ] = 5E [X] = 5, Var[W ] = 5 Var[X] = 5. (3)
It follows that
E W 2 = Var[W ] + (E [W ])2 = 5 + 25 = 30. (4)

In the iid random sequence K1 , K2 , . . ., each Ki has PMF
⎧
⎨ 1−p k = 0,
PK (k) = p k = 1, (1)
⎩
0 otherwise.
(a) The MGF of K is φK (s) = E[esK ] = 1 − p + pes .
(b) By Theorem 6.8, M = K1 + K2 + . . . + Kn has MGF
φM (s) = [φK (s)]n = [1 − p + pes ]n (2)
242
(c) Although we could just use the fact that the expectation of the sum equals the sum of the
expectations, the problem asks us to find the moments using φM (s). In this case,
$
dφM (s) $$ $
E [M ] = $ = n(1 − p + pes )n−1 pes $s=0 = np (3)
ds s=0
The second moment of M can be found via

$
2 dφM (s) $$
E M = (4)
ds $s=0
$
= np (n − 1)(1 − p + pes )pe2s + (1 − p + pes )n−1 es $s=0 (5)
= np[(n − 1)p + 1] (6)
The variance of M is

Var[M ] = E M 2 − (E [M ])2 = np(1 − p) = n Var[K] (7)

Based on the problem statement, the number of points Xi that you earn for game i has PMF

1/3 x = 0, 1, 2
PXi (x) = (1)
0 otherwise
(a) The MGF of Xi is

φXi (s) = E esXi = 1/3 + es /3 + e2s /3 (2)
Since Y = X1 + · · · + Xn , Theorem 6.8 implies
φY (s) = [φXi (s)]n = [1 + es + e2s ]n /3n (3)
(b) First we observe that first and second moments of Xi are

E [Xi ] = xPXi (x) = 1/3 + 2/3 = 1 (4)
x
2
E Xi2 = x PXi (x) = 12 /3 + 22 /3 = 5/3 (5)
x
Hence,
Var[Xi ] = E Xi2 − (E [Xi ])2 = 2/3. (6)
By Theorems 6.1 and 6.3, the mean and variance of Y are
E [Y ] = nE [X] = n (7)
Var[Y ] = n Var[X] = 2n/3 (8)
Another more complicated way to find the mean and variance is to evaluate derivatives of
φY (s) as s = 0.
243

2k e−2 /k! k = 0, 1, 2, . . .
PKi (k) = (1)
0 otherwise
And let Ri = K1 + K2 + . . . + Ki
(a) From Table 6.1, we find that the Poisson (α = 2) random variable K has MGF φK (s) =
s
e2(e −1) .
(b) The MGF of Ri is the product of the MGFs of the Ki ’s.

i
s −1)
φRi (s) = φK (s) = e2i(e (2)
n=1
(c) Since the MGF of Ri is of the same form as that of the Poisson with parameter, α = 2i.
Therefore we can conclude that Ri is in fact a Poisson random variable with parameter
α = 2i. That is,
(2i)r e−2i /r! r = 0, 1, 2, . . .
PRi (r) = (3)
0 otherwise
(d) Because Ri is a Poisson random variable with parameter α = 2i, the mean and variance of
Ri are then both 2i.

The total energy stored over the 31 days is
Y = X1 + X2 + · · · + X31 (1)
The random variables X1 , . . . , X31 are Gaussian and independent but not identically distributed.
However, since the sum of independent Gaussian random variables is Gaussian, we know that Y is
Gaussian. Hence, all we need to do is find the mean and variance of Y in order to specify the PDF
of Y . The mean of Y is

31
31
31(32)
E [Y ] = E [Xi ] = (32 − i/4) = 32(31) − = 868 kW-hr (2)
8
i=1 i=1
Since each Xi has variance of 100(kW-hr)2 , the variance of Y is
Var[Y ] = Var[X1 ] + · · · + Var[X31 ] = 31 Var[Xi ] = 3100 (3)
Since E[Y ] = 868 and Var[Y ] = 3100, the Gaussian PDF of Y is

1 2
fY (y) = √ e−(y−868) /6200 (4)
6200π
244
By Theorem 6.8, we know that φM (s) = [φK (s)]n .
(a) The first derivative of φM (s) is
dφM (s) dφK (s)

= n [φK (s)]n−1 (1)
ds ds
We can evaluate dφM (s)/ds at s = 0 to find E[M ].
$ $
dφM (s) $$ $
n−1 dφK (s) $
E [M ] = = n [φ (s)] = nE [K] (2)
ds $s=0 ds $s=0
K
(b) The second derivative of φM (s) is

2
d2 φM (s) dφK (s) d2 φK (s)
= n(n − 1) [φK (s)]n−2 + n [φK (s)]n−1 (3)
ds2 ds ds2
Evaluating the second derivative at s = 0 yields

$
2 d2 φM (s) $$ 2
E M = = n(n − 1) (E [K])2
+ nE K (4)
ds2 $s=0
(a) From Table 6.1, we see that the exponential random variable X has MGF
λ
φX (s) = (1)
λ−s
(b) Note that K is a geometric random variable identical to the geometric random variable X in
Table 6.1 with parameter p = 1 − q. From Table 6.1, we know that random variable K has
MGF
(1 − q)es
φK (s) = (2)
1 − qes
Since K is independent of each Xi , V = X1 + · · · + XK is a random sum of random variables.
From Theorem 6.12,
(1 − q) λ−s
λ
(1 − q)λ
φV (s) = φK (ln φX (s)) = = (3)
1− λ
q λ−s (1 − q)λ − s
We see that the MGF of V is that of an exponential random variable with parameter (1− q)λ.
The PDF of V is
(1 − q)λe−(1−q)λv v ≥ 0
fV (v) = (4)
0 otherwise
245
The number N of passes thrown has the Poisson PMF and MGF

(30)n e−30 /n! n = 0, 1, . . . s
PN (n) = φN (s) = e30(e −1) (1)
0 otherwise
Let Xi = 1 if pass i is thrown and completed and otherwise Xi = 0. The PMF and MGF of each
Xi is ⎧
⎨ 1/3 x = 0
PXi (x) = 2/3 x = 1 φXi (s) = 1/3 + (2/3)es (2)
⎩
0 otherwise
The number of completed passes can be written as the random sum of random variables
K = X1 + · · · + XN (3)
Since each Xi is independent of N , we can use Theorem 6.12 to write

s −1)
φK (s) = φN (ln φX (s)) = e30(φX (s)−1) = e30(2/3)(e (4)
We see that K has the MGF of a Poisson random variable with mean E[K] = 30(2/3) = 20,
variance Var[K] = 20, and PMF

(20)k e−20 /k! k = 0, 1, . . .
PK (k) = (5)
0 otherwise

In this problem, Y = X1 + · · · + XN is not a straightforward random sum of random variables
because N and the Xi ’s are dependent. In particular, given N = n, then we know that there were
exactly 100 heads in N flips. Hence, given N , X1 + · · · + XN = 100 no matter what is the actual
value of N . Hence Y = 100 every time and the PMF of Y is

1 y = 100
PY (y) = (1)
0 otherwise

Donovan McNabb’s passing yardage is the random sum of random variables
V + Y1 + · · · + YK (1)
where Yi has the exponential PDF

1 −y/15
15 e y≥0
fYi (y) = (2)
0 otherwise
From Table 6.1, the MGFs of Y and K are

1/15 1 s −1)
φY (s) = = φK (s) = e20(e (3)
1/15 − s 1 − 15s
From Theorem 6.12, V has MGF
φV (s) = φK (ln φY (s)) = e20(φY (s)−s) = e300s/(1−15s) (4)
246
The PDF of V cannot be found in a simple form. However, we can use the MGF to calculate the
mean and variance. In particular,
$ $
dφV (s) $$ 300s/(1−15s) 300 $
$
E [V ] = $ =e = 300 (5)
ds (1 − 15s) $s=0
2
$
s=0
2 d2 φV (s) $$
E V = (6)
ds2 $s=0
2 $
$
300 9000 $
= e300s/(1−15s) + e300s/(1−15s)
$ = 99, 000 (7)
(1 − 15s) 2 (1 − 15s) $
3
s=0
Thus, V has variance Var[V ] = E[V 2 ] − (E[V ])2 = 9, 000 and standard deviation σV ≈ 94.9.
A second way to calculate the mean and variance of V is to use Theorem 6.13 which says
E [V ] = E [K] E [Y ] = 20(15) = 200 (8)

2 2 2
Var[V ] = E [K] Var[Y ] + Var[K](E [Y ]) = (20)15 + (20)15 = 9000 (9)

Since each ticket is equally likely to have one of 466 combinations, the probability a ticket is a
winner is
1
q = 46 (1)
6
Let Xi = 1 if the ith ticket sold is a winner; otherwise Xi = 0. Since the number K of tickets sold
has a Poisson PMF with E[K] = r, the number of winning tickets is the random sum
V = X1 + · · · + XK (2)
From Appendix A,
s −1]
φX (s) = (1 − q) + qes φK (s) = er[e (3)
By Theorem 6.12,
s −1)
φV (s) = φK (ln φX (s)) = er[φX (s)−1] = erq(e (4)
Hence, we see that V has the MGF of a Poisson random variable with mean E[V ] = rq. The PMF
of V is
(rq)v e−rq /v! v = 0, 1, 2, . . .
PV (v) = (5)
0 otherwise
(a) We can view K as a shifted geometric random variable. To find the MGF, we start from first
principles with Definition 6.1:
∞
∞
p
φK (s) = esk p(1 − p)k = p [(1 − p)es ]k = (1)
1 − (1 − p)es
k=0 n=0
247
2 /2
(b) First, we need to recall that each Xi has MGF φX (s) = es+s . From Theorem 6.12, the
MGF of R is
p
φR (s) = φK (ln φX (s)) = φK (s + s2 /2) = (2)
1 − (1 − p)es+s2 /2
(c) To use Theorem 6.13, we first need to calculate the mean and variance of K:
$ $
dφK (s) $$ p(1 − p)es 2 $$ 1−p
E [K] = $ = $ = (3)
ds s=0 1 − (1 − p)e s $ p
$ s=0
$
2 d2 φK (s) $$ [1 − (1 − p)es ]es + 2(1 − p)e2s $$
E K = = p(1 − p) (4)
ds2 $s=0 [1 − (1 − p)es ]3 $
s=0
(1 − p)(2 − p)
= (5)
p2
Hence, Var[K] = E[K 2 ] − (E[K])2 = (1 − p)/p2 . Finally. we can use Theorem 6.13 to write
1−p 1−p 1 − p2
Var[R] = E [K] Var[X] + (E [X])2 Var[K] = + 2 = (6)
p p p2

The way to solve for the mean and variance of U is to use conditional expectations. Given K = k,
U = X1 + · · · + Xk and
E [U |K = k] = E [X1 + · · · + Xk |X1 + · · · + Xn = k] (1)

k
= E [Xi |X1 + · · · + Xn = k] (2)
i=1
Since Xi is a Bernoulli random variable,

⎡ ⎤

n
E [Xi |X1 + · · · + Xn = k] = P ⎣Xi = 1| Xj = k ⎦ (3)
j=1

P Xi = 1, j
=i Xj = k − 1
=
(4)
n
P j=1 Xj = k

Note that nj=1 Xj is just a binomial random variable

for n trials while j
=i Xj is a binomial
random variable for n − 1 trials. In addition, Xi and j
=i Xj are independent random variables.
This implies

P [Xi = 1] P j
=i Xj = k − 1
E [Xi |X1 + · · · + Xn = k] =
(5)
n
P X
j=1 j = k
n−1 k−1
p p (1 − p)n−1−(k−1) k
= k−1 n k = (6)
k p (1 − p)
n−k n
248
A second way is to argue that symmetry implies E[Xi |X1 + · · · + Xn = k] = γ, the same for each
i. In this case,

n
nγ = E [Xi |X1 + · · · + Xn = k] = E [X1 + · · · + Xn |X1 + · · · + Xn = k] = k (7)
i=1
Thus γ = k/n. At any rate, the conditional mean of U is

k
k
k k2
E [U |K = k] = E [Xi |X1 + · · · + Xn = k] = = (8)
n n
i=1 i=1
This says that the random variable E[U |K] = K 2 /n. Using iterated expectations, we have

E [U ] = E [E [U |K]] = E K 2 /n (9)
Since K is a binomial random variable, we know that E[K] = np and Var[K] = np(1 − p). Thus,
1 2 1
E [U ] = E K = Var[K] + (E [K])2 = p(1 − p) + np2 (10)
n n
On the other hand, V is just and ordinary random sum of independent random variables and the
mean of E[V ] = E[X]E[M ] = np2 .

Using N to denote the number of games played, we can write the total number of points earned as
the random sum
Y = X1 + X2 + · · · + XN (1)
(a) It is tempting to use Theorem 6.12 to find φY (s); however, this would be wrong since each Xi
is not independent of N . In this problem, we must start from first principles using iterated
expectations.

∞
φY (s) = E E es(X1 +···+XN ) |N = PN (n) E es(X1 +···+Xn ) |N = n (2)
n=1
Given N = n, X1 , . . . , Xn are independent so that

E es(X1 +···+Xn ) |N = n = E esX1 |N = n E esX2 |N = n · · · E esXn |N = n (3)
Given N = n, we know that games 1 through n − 1 were either wins or ties and that game n
was a loss. That is, given N = n, Xn = 0 and for i < n, Xi = 0. Moreover, for i < n, Xi has
the conditional PMF

1/2 x = 1, 2
PXi |N =n (x) = PXi |Xi
=0 (x) = (4)
0 otherwise
These facts imply

E esXn |N = n = e0 = 1 (5)
and that for i < n,

E esXi |N = n = (1/2)es + (1/2)e2s = es /2 + e2s /2 (6)
249
Now we can find the MGF of Y .
∞

φY (s) = PN (n) E esX1 |N = n E esX2 |N = n · · · E esXn |N = n (7)
n=1
∞ ∞
n−1 1 n
= PN (n) es /2 + e2s /2 = s 2s
PN (n) es /2 + e2s /2 (8)
e /2 + e /2
n=1 n=1
It follows that
∞
1 n ln[(es +e2s )/2] φN (ln[es /2 + e2s /2])
φY (s) = PN (n) e = (9)
es /2 + e2s /2 es /2 + e2s /2
n=1
The tournament ends as soon as you lose a game. Since each game is a loss with probability
1/3 independent of any previous game, the number of games played has the geometric PMF
and corresponding MGF

(2/3)n−1 (1/3) n = 1, 2, . . . (1/3)es
PN (n) = φN (s) = (10)
0 otherwise 1 − (2/3)es
Thus, the MGF of Y is

1/3
φY (s) = (11)
1− (es+ e2s )/3
(b) To find the moments of Y , we evaluate the derivatives of the MGF φY (s). Since
dφY (s) es + 2e2s

= (12)
ds 9 [1 − es /3 − e2s /3]2
we see that $
dφY (s) $$ 3
E [Y ] = = =3 (13)
ds $s=0 9(1/3)2
If you’re curious, you may notice that E[Y ] = 3 precisely equals E[N ]E[Xi ], the answer you
would get if you mistakenly assumed that N and each Xi were independent. Although this
may seem like a coincidence, its actually the result of theorem known as Wald’s equality.
The second derivative of the MGF is
d2 φY (s) (1 − es /3 − e2s /3)(es + 4e2s ) + 2(es + 2e2s )2 /3
= (14)
ds2 9(1 − es /3 − e2s /3)3

$
2
d2 φY (s) $$ 5/3 + 6
E Y = 2 $ = = 23 (15)
ds s=0 1/3
The variance of Y is Var[Y ] = E[Y 2 ] − (E[Y ])2 = 23 − 9 = 14.
250
We know that the waiting time, W is uniformly distributed on [0,10] and therefore has the following
PDF.
1/10 0 ≤ w ≤ 10
fW (w) = (1)
0 otherwise
We also know that the total time is 3 milliseconds plus the waiting time, that is X = W + 3.
(a) The expected value of X is E[X] = E[W + 3] = E[W ] + 3 = 5 + 3 = 8.
(b) The variance of X is Var[X] = Var[W + 3] = Var[W ] = 25/3.
(c) The expected value of A is E[A] = 12E[X] = 96.

(d) The standard deviation of A is σA = Var[A] = 12(25/3) = 10.
(e) P [A > 116] = 1 − Φ( 116−96

10 ) = 1 − Φ(2) = 0.02275.
10 ) = Φ(−1) = 1 − Φ(1) = 0.1587

(f) P [A < 86] = Φ( 86−96

Knowing that the probability that voice call occurs is 0.8 and the probability that a data call occurs
is 0.2 we can define the random variable Di as the number of data calls in a single telephone call.
It is obvious that for any i there are only two possible values for Di , namely 0 and 1. Furthermore
for all i the Di ’s are independent and identically distributed withe the following PMF.
⎧
⎨ 0.8 d = 0
PD (d) = 0.2 d = 1 (1)
⎩
0 otherwise
From the above we can determine that
E [D] = 0.2 Var [D] = 0.2 − 0.04 = 0.16 (2)
With these facts, we can answer the questions posed by the problem.
(a) E[K100 ] = 100E[D] = 20

√
(b) Var[K100 ] = 100 Var[D] = 16 = 4

(c) P [K100 ≥ 18] = 1 − Φ 18−20
4 = 1 − Φ(−1/2) = Φ(1/2) = 0.6915
(d) P [16 ≤ K100 ≤ 24] = Φ( 24−20

4 ) − Φ( 4 ) = Φ(1) − Φ(−1) = 2Φ(1) − 1 = 0.6826
16−20
(a) Let X1 , . . . , X120 denote the set of call durations (measured in minutes) during the month.
From the problem statement, each X − I is an exponential (λ) random variable with E[Xi ] =
1/λ = 2.5 min and Var[Xi ] = 1/λ2 = 6.25 min2 . The total number of minutes used during
the month is Y = X1 + · · · + X120 . By Theorem 6.1 and Theorem 6.3,
E [Y ] = 120E [Xi ] = 300 Var[Y ] = 120 Var[Xi ] = 750. (1)
251
The subscriber’s bill is 30 + 0.4(y − 300)+ where x+ = x if x ≥ 0 or x+ = 0 if x < 0. the
subscribers bill is exactly $36 if Y = 315. The probability the subscribers bill exceeds $36
equals

Y − 300 315 − 300 15
P [Y > 315] = P > =Q √ = 0.2919. (2)
σY σY 750
(b) If the actual call duration is Xi , the subscriber is billed for Mi = Xi minutes. Because
each Xi is an exponential (λ) random variable, Theorem 3.9 says that Mi is a geometric (p)
random variable with p = 1 − e−λ = 0.3297. Since Mi is geometric,
1 1−p
E [Mi ] = = 3.033, Var[Mi ] = = 6.167. (3)
p p2
The number of billed minutes in the month is B = M1 + · · · + M120 . Since M1 , . . . , M120 are
iid random variables,
E [B] = 120E [Mi ] = 364.0, Var[B] = 120 Var[Mi ] = 740.08. (4)
Similar to part (a), the subscriber is billed $36 if B = 315 minutes. The probability the
subscriber is billed more than $36 is

B − 364 315 − 365
P [B > 315] = P √ > √ = Q(−1.8) = Φ(1.8) = 0.964. (5)
740.08 740.08

In Problem 6.2.6, we learned that a sum of iid Poisson random variables is a Poisson random
variable. Hence Wn is a Poisson random variable with mean E[Wn ] = nE[K] = n. Thus Wn has
variance Var[Wn ] = n and PMF
w −n
n e /w! w = 0, 1, 2, . . .
PWn (w) = (1)
0 otherwise
All of this implies that we can exactly calculate
P [Wn = n] = PWn (n) = nn e−n /n! (2)
Since we can perform the exact calculation, using a central limit theorem may seem silly; however
for large n, calculating nn or n! is difficult for large n. Moreover, it’s interesting to see how good
the approximation is. In this case, the approximation is

n + 0.5 − n n − 0.5 − n 1
P [Wn = n] = P [n ≤ Wn ≤ n] ≈ Φ √ −Φ √ = 2Φ √ − 1 (3)
n n 2 n
The comparison of the exact calculation and the approximation are given in the following table.
P [Wn = n] n = 1 n = 4 n = 16 n = 64
exact 0.3679 0.1954 0.0992 0.0498 (4)
approximate 0.3829 0.1974 0.0995 0.0498
252
(a) Since the number of requests N has expected value E[N ] = 300 and variance Var[N ] = 300,
we need C to satisfy

N − 300 C − 300
P [N > C] = P √ > √ (1)
300 300

C − 300
=1−Φ √ = 0.05. (2)
300
From Table 3.1, we note that Φ(1.65) = 0.9505. Thus,
√
C = 300 + 1.65 300 = 328.6. (3)
(b) For C = 328.6, the exact probability of overload is
P [N > C] = 1 − P [N ≤ 328] = 1 − poissoncdf(300,328) = 0.0516, (4)
which shows the central limit theorem approximation is reasonable.
(c) This part of the problem could be stated more carefully. Re-examining Definition 2.10 for
the Poisson random variable and the accompanying discussion in Chapter 2, we observe that
the webserver has an arrival rate of λ = 300 hits/min, or equivalently λ = 5 hits/sec. Thus
in a one second interval, the number of requests N is a Poisson (α = 5) random variable.
However, since the server “capacity” in a one second interval is not precisely defined, we will
make the somewhat arbitrary definition that the server capacity is C = 328.6/60 = 5.477
packets/sec. With this somewhat arbitrary definition, the probability of overload in a one
second interval is

P N > C = 1 − P N ≤ 5.477 = 1 − P N ≤ 5 . (5)
Because the number of arrivals in the interval is small, it would be a mistake to use the
Central Limit Theorem to estimate this overload probability. However, the direct calculation
of the overload probability is not hard. For E[N ] = α = 5,

5
5
αn
1 − P N ≤ 5 = 1 − PN (n) = 1 − e−α = 0.3840. (6)
n!
n=0 n=0
(d) Here we find the smallest C such that P [N ≤ C] ≥ 0.95. From the previous step, we know
that C > 5. Since N is a Poisson (α = 5) random variable, we need to find the smallest C
such that
C
P [N ≤ C] = αn e−α /n! ≥ 0.95. (7)
n=0
Some experiments with poissoncdf(alpha,c) will show that P [N ≤ 8] = 0.9319 while

P [N ≤ 9] = 0.9682. Hence C = 9.
253
(e) If we use the Central Limit theorem to estimate the overload probability in a one second
interval, we would use the facts that E[N ] = 5 and Var[N ] = 5 to estimate the the overload
probability as
5−5
1−P N ≤5 =1−Φ √ = 0.5 (8)
5
which overestimates the overload probability by roughly 30 percent. We recall from Chapter 2
that a Poisson random is the limiting case of the (n, p) binomial random variable when n is
large and np = α.In general, for fixed p, the Poisson and binomial PMFs become closer as n
increases. Since large n is also the case for which the central limit theorem applies, it is not
surprising that the the CLT approximation for the Poisson (α) CDF is better when α = np
is large.
Comment: Perhaps a more interesting question is why the overload probability in a one-second
interval is so much higher than that in a one-minute interval? To answer this, consider a T -second
interval in which the number of requests NT is a Poisson (λT ) random variable while the server
capacity is cT hits. In the earlier problem parts, c = 5.477 hits/sec. We make the assumption that
the server system is reasonably well-engineered in that c > λ. (We will learn in Chapter 12 that to
assume otherwise means that the backlog of requests will grow without bound.) Further, assuming
T is fairly large, we use the CLT to estimate the probability of overload in a T -second interval as
√
NT − λT cT − λT
P [NT ≥ cT ] = P √ ≥ √ =Q k T , (9)
λT λT
√
where k = (c − λ)/ λ. As long as c > λ, the overload probability decreases with increasing T .
In fact, the overload probability goes rapidly to zero as T becomes large. The reason is that the
gap cT − λT between server capacity cT and the expected number of requests λT √ grows linearly
in T while the standard deviation of the number of requests grows proportional to T . However,
one should add that the definition of a T -second overload is somewhat arbitrary. In fact, one can
argue that as T becomes large, the requirement for no overloads simply becomes less stringent. In
Chapter 12, we will learn techniques to analyze a system such as this webserver in terms of the
average backlog of requests and the average delay in serving in serving a request. These statistics
won’t depend on a particular time period T and perhaps better describe the system performance.
(a) The number of tests L needed to identify 500 acceptable circuits is a Pascal (k = 500, p = 0.8)
random variable, which has expected value E[L] = k/p = 625 tests.
(b) Let K denote the number of acceptable circuits in n = 600 tests. Since K is binomial
(n = 600, p = 0.8), E[K] = np = 480 and Var[K] = np(1 − p) = 96. Using the CLT, we
estimate the probability of finding at least 500 acceptable circuits as

K − 480 20 20
P [K ≥ 500] = P √ ≥√ ≈Q √ = 0.0206. (1)
96 96 96
(c) Using Matlab, we observe that
1.0-binomialcdf(600,0.8,499)
ans =
0.0215
254
(d) We need to find the smallest value of n such that the binomial (n, p) random variable K
satisfies P [K ≥ 500] ≥ 0.9. Since E[K] = np and Var[K] = np(1−p), the CLT approximation
yields 0 1
K − np 500 − np
P [K ≥ 500] = P ≥ ≈ 1 − Φ(z) = 0.90. (2)
np(1 − p) np(1 − p)

where z = (500 − np)/ np(1 − p). It follows that 1 − Φ(z) = Φ(−z) ≥ 0.9, implying
z = −1.29. Since p = 0.8, we have that

np − 500 = 1.29 np(1 − p). (3)
Equivalently, for p = 0.8, solving the quadratic equation

2
500 1−p
n− = (1.29)2 n (4)
p p
we obtain n = 641.3. Thus we should test n = 642 circuits.

2 /2
The N [0, 1] random variable Z has MGF φZ (s) = es . Hence the Chernoff bound for Z is
2 /2 2 /2−sc
P [Z ≥ c] ≤ min e−sc es = min es (1)
s≥0 s≥0
2 /2−sc
We can minimize es by minimizing the exponent s2 /2 − sc. By setting
d 2
s /2 − sc = 2s − c = 0 (2)
ds
2
we obtain s = c. At s = c, the upper bound is P [Z ≥ c] ≤ e−c /2 . The table below compares this
upper bound to the true probability. Note that for c = 1, 2 we use Table 3.1 and the fact that
Q(c) = 1 − Φ(c).
c=1 c=2 c=3 c=4 c=5

Chernoff bound 0.606 0.135 0.011 3.35 × 10−4 3.73 × 10−6 (3)
Q(c) 0.1587 0.0228 0.0013 3.17 × 10−5 2.87 × 10−7
We see that in this case, the Chernoff bound typically overestimates the true probability by roughly
a factor of 10.

For an N [µ, σ 2 ] random variable X, we can write
P [X ≥ c] = P [(X − µ)/σ ≥ (c − µ)/σ] = P [Z ≥ (c − µ)/σ] (1)
Since Z is N [0, 1], we can apply the result of Problem 6.8.1 with c replaced by (c − µ)/σ. This
yields
2 2
P [X ≥ c] = P [Z ≥ (c − µ)/σ] ≤ e−(c−µ) /2σ (2)
255
From Appendix A, we know that the MGF of K is
s −1)
φK (s) = eα(e (1)
The Chernoff bound becomes

s −1) s −1)−sc
P [K ≥ c] ≤ min e−sc eα(e = min eα(e (2)
s≥0 s≥0
Since ey is an increasing function, it is sufficient to choose s to minimize h(s) = α(es − 1) − sc.

Setting dh(s)/ds = αes − c = 0 yields es = c/α or s = ln(c/α). Note that for c < α, the minimizing
s is negative. In this case, we choose s = 0 and the Chernoff bound is P [K ≥ c] ≤ 1. For c ≥ α,
applying s = ln(c/α) yields P [K ≥ c] ≤ e−α (αe/c)c . A complete expression for the Chernoff bound
is
1 c<α
P [K ≥ c] ≤ (3)
αc ec e−α /cc c ≥ α

This problem is solved completely in the solution to Quiz 6.8! We repeat that solution here. Since
W = X1 + X2 + X3 is an Erlang (n = 3, λ = 1/2) random variable, Theorem 3.11 says that for any
w > 0, the CDF of W satisfies

2
(λw)k e−λw
FW (w) = 1 − (1)
k!
k=0
Equivalently, for λ = 1/2 and w = 20,
P [W > 20] = 1 − FW (20) (2)

−10 10 102
=e 1+ + = 61e−10 = 0.0028 (3)
1! 2!

Let Wn = X1 + · · · + Xn . Since Mn (X) = Wn /n, we can write
P [Mn (X) ≥ c] = P [Wn ≥ nc] (1)
Since φWn (s) = (φX (s))n , applying the Chernoff bound to Wn yields
n
P [Wn ≥ nc] ≤ min e−snc φWn (s) = min e−sc φX (s) (2)
s≥0 s≥0
For y ≥ 0, y n is a nondecreasing function of y. This implies that the value of s that minimizes
e−sc φX (s) also minimizes (e−sc φX (s))n . Hence
n
P [Mn (X) ≥ c] = P [Wn ≥ nc] ≤ min e−sc φX (s) (3)
s≥0
256
Note that Wn is a binomial (10n , 0.5) random variable. We need to calculate
P [Bn ] = P [0.499 × 10n ≤ Wn ≤ 0.501 × 10n ] (1)

= P [Wn ≤ 0.501 × 10 ] − P [Wn < 0.499 × 10 ] .
n n
(2)
A complication is that the event Wn < w is not the same as Wn ≤ w when w is an integer. In this
case, we observe that
P [Wn < w] = P [Wn ≤ w − 1] = FWn (w − 1) (3)
Thus
9 :
P [Bn ] = FWn (0.501 × 10n ) − FWn 0.499 × 109 − 1 (4)
For n = 1, . . . , N , we can calculate P [Bn ] in this Matlab program:
function pb=binomialcdftest(N);
pb=zeros(1,N);
for n=1:N,
w=[0.499 0.501]*10^n;
w(1)=ceil(w(1))-1;
pb(n)=diff(binomialcdf(10^n,0.5,w));
end
Unfortunately, on this user’s machine (a Windows XP laptop), the program fails for N = 4. The
problem, as noted earlier is that binomialcdf.m uses binomialpmf.m, which fails for a binomial
(10000, p) random variable. Of course, your mileage may vary. A slightly better solution is to use
the bignomialcdf.m function, which is identical to binomialcdf.m except it calls bignomialpmf.m
rather than binomialpmf.m. This enables calculations for larger values of n, although at some cost
in numerical accuracy. Here is the code:
function pb=bignomialcdftest(N);
pb=zeros(1,N);
for n=1:N,
w=[0.499 0.501]*10^n;
w(1)=ceil(w(1))-1;
pb(n)=diff(bignomialcdf(10^n,0.5,w));
end
For comparison, here are the outputs of the two programs:

>> binomialcdftest(4)
ans =
0.2461 0.0796 0.0756 NaN
>> bignomialcdftest(6)
ans =
0.2461 0.0796 0.0756 0.1663 0.4750 0.9546
The result 0.9546 for n = 6 corresponds to the exact probability in Example 6.15 which used the
CLT to estimate the probability as 0.9544. Unfortunately for this user, for n = 7, bignomialcdftest(7)
failed.
257
The Erlang (n, λ = 1) random variable X has expected value E[X] = n/λ = n and variance
Var[X] = n/λ2 = n. The PDF of X as well as the PDF of a Gaussian random variable Y with the
same expected value and variance are
⎧ n−1 −x
⎨ x e
x≥0 1 2
fX (x) = (n − 1)! fY (x) = √ e−x /2n (1)
⎩ 2πn
0 otherwise
function df=erlangclt(n); From the forms of the functions, it not likely to be apparent that
r=3*sqrt(n); fX (x) and fY (x) are similar. The following program plots fX (x)
x=(n-r):(2*r)/100:n+r; and fY (x) for values of x within three standard deviations of the
fx=erlangpdf(n,1,x); expected value n. Below are sample outputs of erlangclt(n) for
fy=gausspdf(n,sqrt(n),x);
n = 4, 20, 100.
plot(x,fx,x,fy);
df=fx-fy;
In the graphs we will see that as n increases, the Erlang PDF becomes increasingly similar to
the Gaussian PDF of the same expected value and variance. This is not surprising since the Erlang
(n, λ) random variable is the sum of n of exponential random variables and the CLT says that the
Erlang CDF should converge to a Gaussian CDF as n gets large.
0.1
0.2 0.04
fX(x) fY(x)
fX(x) fY(x)
fX(x) fY(x)
0.15
0.05
0.1 0.02
0.05
0 0 0
0 5 10 10 20 30 80 100 120
x x x
erlangclt(4) erlangclt(20) erlangclt(100)
On the other hand, the convergence should be viewed with some caution. For example, the
mode (the peak value) of the Erlang PDF occurs at x = n − 1 while the mode of the Gaussian PDF
is at x = n. This difference only appears to go away for n = 100 because the graph x-axis range
is expanding. More important, the two PDFs are quite different far away from the center of the
distribution. The Erlang PDF is always zero for x < 0 while the Gaussian PDF is always positive.
For large postive x, the two distributions do not have the same exponential decay. Thus it’s not a
good idea to use the CLT to estimate probabilities of rare events such as {X > x} for extremely
large values of x.

In this problem, we re-create the plots of Figure 6.3 except we use the binomial PMF and corre-
sponding Gaussian PDF. Here is a Matlab program that compares the binomial (n, p) PMF and
the Gaussian PDF with the same expected value and variance.
258
function y=binomcltpmf(n,p)
x=-1:17;
xx=-1:0.05:17;
y=binomialpmf(n,p,x);
std=sqrt(n*p*(1-p));
clt=gausspdf(n*p,std,xx);
hold off;
pmfplot(x,y,’\it x’,’\it p_X(x) f_X(x)’);
hold on; plot(xx,clt); hold off;
Here are the output plots for p = 1/2 and n = 2, 4, 8, 16.
1 0.4
fX(x)
fX(x)
0.5 0.2
pX(x)
pX(x)
0 0
−5 0 5 10 15 20 −5 0 5 10 15 20
x x
binomcltpmf(2,0.5) binomcltpmf(4,0.5)
0.4 0.2
fX(x)
fX(x)
0.2 0.1
pX(x)
pX(x)
0 0
−5 0 5 10 15 20 −5 0 5 10 15 20
x x
binomcltpmf(8,0.5) binomcltpmf(16,0.5)
To see why the values of the PDF and PMF are roughly the same, consider the Gaussian random
variable Y . For small ∆,
FY (x + ∆/2) − FY (x − ∆/2)
fY (x) ∆ ≈ . (1)
∆
For ∆ = 1, we obtain
fY (x) ≈ FY (x + 1/2) − FY (x − 1/2) . (2)
Since the Gaussian CDF is approximately the same as the CDF of the binomial (n, p) random
variable X, we observe for an integer x that
fY (x) ≈ FX (x + 1/2) − FX (x − 1/2) = PX (x) . (3)
Although the equivalence in heights of the PMF and PDF is only an approximation, it can be
useful for checking the correctness of a result.

Since the conv function is for convolving signals in time, we treat PX1 (x) and PX2 (x2 )x, or as
though they were signals in time starting at time x = 0. That is,

px1 = PX1 (0) PX1 (1) · · · PX1 (25) (1)

px2 = PX2 (0) PX2 (1) · · · PX2 (100) (2)
259
%convx1x2.m In particular, between its minimum and maximum values,
sw=(0:125); the vector px2 must enumerate all integer values, including
px1=[0,0.04*ones(1,25)]; those which have zero probability. In addition, we write down
px2=zeros(1,101); sw=0:125 directly based on knowledge that the range enu-
px2(10*(1:10))=10*(1:10)/550;
merated by px1 and px2 corresponds to X1 + X2 having a
pw=conv(px1,px2);
h=pmfplot(sw,pw,... minimum value of 0 and a maximum value of 125.
’\itw’,’\itP_W(w)’);
set(h,’LineWidth’,0.25);
The resulting plot will be essentially identical to Figure 6.4. One final note, the command
set(h,’LineWidth’,0.25) is used to make the bars of the PMF thin enough to be resolved indi-
vidually.
sx1=(1:10);px1=0.1*ones(1,10); Since the mdgrid function extends naturally to higher

sx2=(1:20);px2=0.05*ones(1,20); dimensions, this solution follows the logic of sumx1x2
sx3=(1:30);px3=ones(1,30)/30; in Example 6.19.
[SX1,SX2,SX3]=ndgrid(sx1,sx2,sx3);
[PX1,PX2,PX3]=ndgrid(px1,px2,px3);
SW=SX1+SX2+SX3;
PW=PX1.*PX2.*PX3;
sw=unique(SW);
pw=finitepmf(SW,PW,sw);
h=pmfplot(sw,pw,’\itw’,’\itP_W(w)’);
The output of sumx1x2x3 is the plot of the PMF of W shown below. We use the command
set(h,’LineWidth’,0.25) to ensure that the bars of the PMF are thin enough to be resolved
individually.
0.04
0.03
PW(w)
0.02
0.01
0
0 10 20 30 40 50 60
w
function [pw,sw]=sumfinitepmf(px,sx,py,sy); sumfinitepmf generalizes the method of Ex-

[SX,SY]=ndgrid(sx,sy); ample 6.19. The only difference is that the
[PX,PY]=ndgrid(px,py); PMFs px and py and ranges sx and sy are
SW=SX+SY;PW=PX.*PY; not hard coded, but instead are function in-
sw=unique(SW);
puts.
pw=finitepmf(SW,PW,sw);
260
As an example, suppose X is a discrete uniform (0, 20) random variable and Y is an independent
discrete uniform (0, 80) random variable. The following program sum2unif will generate and plot
the PMF of W = X + Y .
%sum2unif.m
sx=0:20;px=ones(1,21)/21;
sy=0:80;py=ones(1,81)/81;
[pw,sw]=sumfinitepmf(px,sx,py,sy);
h=pmfplot(sw,pw,’\it w’,’\it P_W(w)’);
Here is the graph generated by sum2unif.
0.015
0.01
PW(w)
0.005
0
0 10 20 30 40 50 60 70 80 90 100
w
261

Recall that X1 , X2 . . . Xn are independent exponential random variables with mean value µX = 5
so that for x ≥ 0, FX (x) = 1 − e−x/5 .
2
(a) Using Theorem 7.1, σM 2 /n. Realizing that σ 2 = 25, we obtain
= σX
n (x) X
2
σX 25
Var[M9 (X)] = = (1)
9 9
(b)
P [X1 ≥ 7] = 1 − P [X1 ≤ 7] (2)

−7/5 −7/5
= 1 − FX (7) = 1 − (1 − e )=e ≈ 0.247 (3)
(c) First we express P [M9 (X) > 7] in terms of X1 , . . . , X9 .
P [M9 (X) > 7] = 1 − P [M9 (X) ≤ 7] = 1 − P [(X1 + . . . + X9 ) ≤ 63] (4)
Now the probability that M9 (X) > 7 can be approximated using the Central Limit Theorem
(CLT).
P [M9 (X) > 7] = 1 − P [(X1 + . . . + X9 ) ≤ 63] (5)

63 − 9µX
≈1−Φ √ = 1 − Φ(6/5) (6)
9σX
Consulting with Table 3.1 yields P [M9 (X) > 7] ≈ 0.1151.

2 =3
X1 , X2 . . . Xn are independent uniform random variables with mean value µX = 7 and σX
(a) Since X1 is a uniform random variable, it must have a uniform PDF over an interval [a, b].
From Appendix A, we can look up that µX = (a+b)/2 and that Var[X] = (b−a)2 /12. Hence,
given the mean and variance, we obtain the following equations for a and b.
(b − a)2 /12 = 3 (a + b)/2 = 7 (1)
Solving these equations yields a = 4 and b = 10 from which we can state the distribution of
X.
1/6 4 ≤ x ≤ 10
fX (x) = (2)
0 otherwise
(b) From Theorem 7.1, we know that
Var[X] 3
Var[M16 (X)] = = (3)
16 16
262
(c)
' ∞ ' 10
P [X1 ≥ 9] = fX1 (x) dx = (1/6) dx = 1/6 (4)
9 9
(d) The variance of M16 (X) is much less than Var[X1 ]. Hence, the PDF of M16 (X) should
be much more concentrated about E[X] than the PDF of X1 . Thus we should expect
P [M16 (X) > 9] to be much less than P [X1 > 9].
P [M16 (X) > 9] = 1 − P [M16 (X) ≤ 9] = 1 − P [(X1 + · · · + X16 ) ≤ 144] (5)
By a Central Limit Theorem approximation,

144 − 16µX
P [M16 (X) > 9] ≈ 1 − Φ √ = 1 − Φ(2.66) = 0.0039 (6)
16σX
As we predicted, P [M16 (X) > 9] P [X1 > 9].

This problem is in the wrong section since the standard error isn’t defined until Section 7.3. However
is we peek ahead to this section, the problem isn’t very hard. Given thesample mean estimate
Mn (X), the standard error is defined as the standard deviation en = Var[Mn (X)]. In our
2
problem, we use samples Xi to generate Yi = Xi . For the sample mean Mn (Y ), we need to find
the standard error )
Var[Y ]
en = Var[Mn (Y )] = . (1)
n
Since X is a uniform (0, 1) random variable,
' 1
2
E [Y ] = E X = x2 dx = 1/3, (2)
0
' 1

E Y 2 = E X4 = x4 dx = 1/5. (3)
0
Thus Var[Y ] = 1/5 − (1/3)2 = 4/45 and the sample mean Mn (Y ) has standard error
)
4
en = . (4)
45n
(a) Since Yn = X2n−1 + (−X2n ), Theorem 6.1 says that the expected value of the difference is
E [Y ] = E [X2n−1 ] + E [−X2n ] = E [X] − E [X] = 0 (1)
(b) By Theorem 6.2, the variance of the difference between X2n−1 and X2n is
Var[Yn ] = Var[X2n−1 ] + Var[−X2n ] = 2 Var[X] (2)
263
(c) Each Yn is the difference of two samples of X that are independent of the samples used by
any other Ym . Thus Y1 , Y2 , . . . is an iid random sequence. By Theorem 7.1, the mean and
variance of Mn (Y ) are
E [Mn (Y )] = E [Yn ] = 0 (3)

Var[Yn ] 2 Var[X]
Var[Mn (Y )] = = (4)
n n

If the average weight of a Maine black bear is 500 pounds with standard deviation equal to 100
pounds, we can use the Chebyshev inequality to upper bound the probability that a randomly
chosen bear will be more then 200 pounds away from the average.
Var[W ] 1002
P [|W − E [W ] | ≥ 200] ≤ ≤ = 0.25 (1)
2002 2002

We know from the Chebyshev inequality that
2
σX
P [|X − E [X] | ≥ c] ≤ (1)
c2
Choosing c = kσX , we obtain
1
P [|X − E [X] | ≥ kσ] ≤ (2)
k2
The actual probability the Gaussian random variable Y is more than k standard deviations from
its expected value is
P [|Y − E [Y ]| ≥ kσY ] = P [Y − E [Y ] ≤ −kσY ] + P [Y − E [Y ] ≥ kσY ] (3)

Y − E [Y ]
= 2P ≥k (4)
σY
= 2Q(k) (5)
The following table compares the upper bound and the true probability:
k=1 k=2 k=3 k=4 k=5

Chebyshev bound 1 0.250 0.111 0.0625 0.040 (6)
2Q(k) 0.317 0.046 0.0027 6.33 × 10−5 5.73 × 10−7
The Chebyshev bound gets increasingly weak as k goes up. As an example, for k = 4, the bound
exceeds the true probability by a factor of 1,000 while for k = 5 the bound exceeds the actual
probability by a factor of nearly 100,000.

The hard part of this problem is to derive the PDF of the sum W = X1 + X2 + X3 of iid uniform
(0, 30) random variables. In this case, we need to use the techniques of Chapter 6 to convolve the
three PDFs. To simplify our calculations, we will instead find the PDF of V = Y1 + Y2 + Y3 where
264
the Yi are iid uniform (0, 1) random variables. By Theorem 3.20 to conclude that W = 30V is the
sum of three iid uniform (0, 30) random variables.
To start, let V2 = Y1 + Y2 . Since each Y1 has a PDF shaped like a unit area pulse, the PDF of
V2 is the triangular function
1 ⎧
⎨ v 0≤v≤1
fV (v)
2
0.5
fV2 (v) = 2−v 1<v ≤2 (1)
⎩
0 0 otherwise
0 1 2
v
The PDF of V is the convolution integral
' ∞
fV (v) = fV2 (y) fY3 (v − y) dy (2)
−∞
' 1 ' 2
= yfY3 (v − y) dy + (2 − y)fY3 (v − y) dy. (3)
0 1
Evaluation of these integrals depends on v through the function

1 v−1<v <1
fY3 (v − y) = (4)
0 otherwise
To compute the convolution, it is helpful to depict the three distinct cases. In each case, the square
“pulse” is fY3 (v − y) and the triangular pulse is fV2 (y).
1 1 1
0.5 0.5 0.5
0 0 0
−1 0 1 2 3 −1 0 1 2 3 −1 0 1 2 3
0≤v<1 1≤v<2 2≤v<3
From the graphs, we can compute the convolution for each case:
' v
1
0≤v<1: fV3 (v) = y dy = v 2 (5)
0 2
' 1 ' v
3
1≤v<2: fV3 (v) = y dy + (2 − y) dy = −v 2 + 3v − (6)
v−1 1 2
' 2
(3 − v)2
2≤v<3: fV3 (v) = (2 − y) dy = (7)
v−1 2
To complete the problem, we use Theorem 3.20 to observe that W = 30V3 is the sum of three iid
uniform (0, 30) random variables. From Theorem 3.19,
⎧
⎪ 2
⎪ (w/30) /60 0 ≤ w < 30,
⎨
1 [−(w/30) + 3(w/30) − 3/2]/30 30 ≤ w < 60,
2
fW (w) = fV3 (v3 ) v/30 = (8)
30 ⎪
⎪ [3 − (w/30)]2 /60 60 ≤ w < 90,
⎩
0 otherwise.
265
Finally, we can compute the exact probability
' $90
1 9
(3 − w/30)3 $$ 1
P [W ≥ 75] = 0[3 − (w/30)] dw = −
2
$ = 48 (9)
60 75 6 75
For comparison, the Markov inequality indicated that P [W < 75] ≤ 3/5 and the Chebyshev in-
equality showed that P [W < 75] ≤ 1/4. As we see, both inequalities are quite weak in this case.

On each roll of the dice, a success, namely snake eyes, occurs with probability p = 1/36. The
number of trials, R, needed for three successes is a Pascal (k = 3, p) random variable with
E [R] = 3/p = 108, Var[R] = 3(1 − p)/p2 = 3780. (1)
(a) By the Markov inequality,
E [R] 54
P [R ≥ 250] ≤ = = 0.432. (2)
250 125
(b) By the Chebyshev inequality,
P [R ≥ 250] = P [R − 108 ≥ 142] = P [|R − 108| ≥ 142] (3)

Var[R]
≤ = 0.1875. (4)
(142)2

(c) The exact value is P [R ≥ 250] = 1 − 249

r=3 PR (r). Since there is no way around summing the
Pascal PMF to find the CDF, this is what pascalcdf does.
>> 1-pascalcdf(3,1/36,249)
ans =
0.0299
Thus the Markov and Chebyshev inequalities are valid bounds but not good estimates of
P [R ≥ 250].

For an an arbitrary Gaussian (µ, σ) random variable Y ,
P [µ − σ ≤ Y ≤ µ + σ] = P [−σ ≤ Y − µ ≤ σ] (1)

Y −µ
= P −1 ≤ ≤1 (2)
σ
= Φ(1) − Φ(−1) = 2Φ(1) − 1 = 0.6827. (3)
Note that Y can be any Gaussian random variable, including, for example, Mn (X) when X is
Gaussian. When X is not Gaussian, the same claim holds to the extent that the central limit
theorem promises that Mn (X) is nearly Gaussian for large n.
266
It should seem obvious that the result is true since Var[R̂n ] going to zero implies the probablity
that R̂n differs from E[R̂n ] is going to zero. Similarly, the difference between E[R̂n ] and r is also
going to zero deterministically. Hence it ought to follow that R̂n is converging to r in probability.
Here are the details:
We must show that limN →∞ P [|R̂n − r| ≥ ] = 0. First we note that R̂n being asymptotically
unbiased implies that limn→∞ E[R̂n ] = r. Equivalently, given > 0, there exists n0 such that
|E[R̂n ] − r| ≤ 2 /2 for all n ≥ n0 .
Second, we observe that
$ $2 $ $2 $ $2 $ $2
$ $ $ $ $ $ $ $
$R̂n − r$ = $(R̂n − E R̂n ) + (E R̂n − r)$ ≤ $R̂n − E R̂n $ + $E R̂n − r$ . (1)
Thus for all n ≥ n0 , $ $2 $ $2

$ $ $ $
$R̂n − r$ ≤≤ $R̂n − E R̂n $ + 2 /2. (2)
It follows for n ≥ n0 that

$ $2 $ $2 $ $2
$ $ $ $ $ $
P $R̂n − r$ ≥ ≤ P
2
$R̂n − E R̂n $ + /2 ≥ = P
2 2
$R̂n − E R̂n $ ≥ /2
2
(3)
By the Chebyshev inequality, we have that

$ $2
$ $ Var[R̂n ]
P $R̂n − E R̂n $ ≥ /2 ≤
2
√ (4)
(/ 2)2
Combining these facts, we see for n ≥ n0 that

$ $2
$ $ Var[R̂n ]
P $R̂n − r$ ≥ ≤
2
√ . (5)
(/ 2)2
It follows that $ $2
$ $ Var[R̂n ]
lim P $R̂n − r$ ≥ ≤ lim
2
√ = 0. (6)
n→∞ n→∞ (/ 2)2
This proves that R̂n is a consistent estimator.

This problem is really very simple. If we let Y = X1 X2 and for the ith trial, let Yi = X1 (i)X2 (i),
then R̂n = Mn (Y ), the sample mean of random variable Y . By Theorem 7.5, Mn (Y ) is unbiased.
Since Var[Y ] = Var[X1 X2 ] < ∞, Theorem 7.7 tells us that Mn (Y ) is a consistent sequence.
(a) Since the expectation of a sum equals the sum of the expectations also holds for vectors,
1 1
n n
E [M(n)] = E [X(i)] = µX = µX . (1)
n n
i=1 i=1
267

(b) The jth component of M(n) is Mj (n) = n1 ni=1 Xj (i), which is just the sample mean of Xj .
Defining Aj = {|Mj (n) − µj | ≥ c}, we observe that

P max |Mj (n) − µj | ≥ c = P [A1 ∪ A2 ∪ · · · ∪ Ak ] . (2)
j=1,...,k
Applying the Chebyshev inequality to Mj (n), we find that
Var[Mj (n)] σj2

P [Aj ] ≤ = . (3)
c2 nc2
By the union bound,

1 2
k k
P max |Mj (n) − µj | ≥ c ≤ P [Aj ] ≤ 2 σj (4)
j=1,...,k nc
j=1 j=1

k
Since 2
j=1 σj < ∞, limn→∞ P [maxj=1,...,k |Mj (n) − µj | ≥ c] = 0.

Note that we can write Yk as

X2k−1 − X2k 2 X2k − X2k−1 2 (X2k − X2k−1 )2
Yk = + = (1)
2 2 2
Hence,
1 2
E [Yk ] = E X2k − 2X2k X2k−1 + X2k−1
2
= E X 2 − (E [X])2 = Var[X] (2)
2
Next we observe that Y1 , Y2 , . . . is an iid random sequence. If this independence is not obvious,
consider that Y1 is a function of X1 and X2 , Y2 is a function of X3 and X4 , and so on. Since
X1 , X2 , . . . is an idd sequence, Y1 , Y2 , . . . is an iid sequence. Hence, E[Mn (Y )] = E[Y ] = Var[X],
implying Mn (Y ) is an unbiased estimator of Var[X]. We can use Theorem 7.5 to prove that Mn (Y )
is consistent if we show that Var[Y ] is finite. Since Var[Y ] ≤ E[Y 2 ], it is sufficient to prove that
E[Y 2 ] < ∞. Note that
4 − 4X 3 X
2k 2k−1 + 6X2k X2k−1 − 4X2k X2k−1 + X2k−1
X2k 2 2 3 4
Yk2 = (3)
4
Taking expectations yields
1 3 2 2
E Yk2 = E X 4 − 2E X 3 E [X] + E X (4)
2 2
Hence, if the first four moments of X are finite, then Var[Y ] ≤ E[Y 2 ] < ∞. By Theorem 7.5, the
sequence Mn (Y ) is consistent.
268
(a) From Theorem 6.2, we have

n
n−1
n
Var[X1 + · · · + Xn ] = Var[Xi ] + 2 Cov [Xi , Xj ] (1)
i=1 i=1 j=i+1
Note that Var[Xi ] = σ 2 and for j > i, Cov[Xi , Xj ] = σ 2 aj−i . This implies

n−1
n
Var[X1 + · · · + Xn ] = nσ 2 + 2σ 2 aj−i (2)
i=1 j=i+1

n−1

= nσ 2 + 2σ 2 a + a2 + · · · + an−i (3)
i=1
2aσ
2 n−1
= nσ 2 + (1 − an−i ) (4)
1−a
i=1
With some more algebra, we obtain
2aσ 2 2aσ 2
Var[X1 + · · · + Xn ] = nσ 2 + (n − 1) − a + a2 + · · · + an−1 (5)
1−a 1−a
2
2
2
n(1 + a)σ 2aσ a
= − − 2σ 2
(1 − an−1 ) (6)
1−a 1−a 1−a
Since a/(1 − a) and 1 − an−1 are both nonnegative,

1+a
Var[X1 + · · · + Xn ] ≤ nσ 2
(7)
1−a
(b) Since the expected value of a sum equals the sum of the expected values,
E [X1 ] + · · · + E [Xn ]
E [M (X1 , . . . , Xn )] = =µ (8)
n
The variance of M (X1 , . . . , Xn ) is
Var[X1 + · · · + Xn ] σ 2 (1 + a)
Var[M (X1 , . . . , Xn )] = ≤ (9)
n2 n(1 − a)
Applying the Chebyshev inequality to M (X1 , . . . , Xn ) yields
Var[M (X1 , . . . , Xn )] σ 2 (1 + a)
P [|M (X1 , . . . , Xn ) − µ| ≥ c] ≤ ≤ (10)
c2 n(1 − a)c2
(c) Taking the limit as n approaches infinity of the bound derived in part (b) yields
σ 2 (1 + a)
lim P [|M (X1 , . . . , Xn ) − µ| ≥ c] ≤ lim =0 (11)
n→∞ n→∞ n(1 − a)c2
Thus
lim P [|M (X1 , . . . , Xn ) − µ| ≥ c] = 0 (12)
n→∞
269
(a) Since the expectation of the sum equals the sum of the expectations,
1 n
1 n
E R̂(n) = E X(m)X (m) = R = R. (1)
n n
m=1 m=1
(b) This proof follows the method used to solve Problem 7.3.4. The i, jth element of R̂(n) is
1
n
R̂i,j (n) = n m=1 Xi (m)Xj (m), which is just the sample mean of Xi Xj . Defining the event
3$ $ 4
$ $
Ai,j = $R̂i,j (n) − E [Xi Xj ]$ ≥ c , (2)
we observe that
$ $
$ $
P max $R̂i,j (n) − E [Xi Xj ]$ ≥ c = P [∪i,j Ai,j ] . (3)
i,j
Applying the Chebyshev inequality to R̂i,j (n), we find that
Var[R̂i,j (n)] Var[Xi Xj ]

P [Ai,j ] ≤ = . (4)
c2 nc2
By the union bound,
$ $
$ $ 1
P max $R̂i,j (n) − E [Xi Xj ]$ ≥ c ≤ P [Ai,j ] ≤ 2 Var[Xi Xj ] (5)
i,j nc
i,j i,j
By the result of Problem 4.11.8, Xi Xj , the product of jointly Gaussian random variables, has
finite variance. Thus

k
k
Var[Xi Xj ] = Var[Xi Xj ] ≤ k 2 max Var[Xi Xj ] < ∞. (6)
i,j
i,j i=1 j=1
It follows that
$ $
$ $ k 2 maxi,j Var[Xi Xj ]
lim P max $R̂i,j (n) − E [Xi X − j]$ ≥ c ≤ lim =0 (7)
n→∞ i,j n→∞ nc2

⎧
⎨ 0.1 x = 0
PX (x) = 0.9 x = 1 (1)
⎩
0 otherwise
(a) E[X] is in fact the same as PX (1) because X is a Bernoulli random variable.
(b) We can use the Chebyshev inequality to find
P [|M90 (X) − PX (1)| ≥ .05] = P [|M90 (X) − E [X] | ≥ .05] ≤ α (2)
In particular, the Chebyshev inequality states that
σX2
.09
α= 2
= = 0.4 (3)
90(.05) 90(.05)2
270
(c) Now we wish to find the value of n such that P [|Mn (X) − PX (1)| ≥ .03] ≤ .01. From the
Chebyshev inequality, we write
2
σX
0.1 = . (4)
n(.03)2
2 = 0.09, solving for n yields n = 100.
Since σX

X1 , X2 , . . . are iid random variables each with mean 75 and standard deviation 15.
(a) We would like to find the value of n such that
P [74 ≤ Mn (X) ≤ 76] = 0.99 (1)
When we know only the mean and variance of Xi , our only real tool is the Chebyshev
inequality which says that
P [74 ≤ Mn (X) ≤ 76] = 1 − P [|Mn (X) − E [X]| ≥ 1] (2)

Var [X] 225
≥1− =1− ≥ 0.99 (3)
n n
This yields n ≥ 22,500.
(b) If each Xi is a Gaussian, the sample mean, Mn (X) will also be Gaussian with mean and
variance
E [Mn (X)] = E [X] = 75 (4)

Var [Mn (X)] = Var [X] /n = 225/n (5)
In this case,

76 − µ 74 − µ
P [74 ≤ Mn (X) ≤ 76] = Φ −Φ (6)
σ σ
√ √
= Φ( n /15) − Φ(− n /15) (7)
√
= 2Φ( n /15) − 1 = 0.99 (8)
Thus, n = 1,521.
Since even under the Gaussian assumption, the number of samples n is so large that even if the
Xi are not Gaussian, the sample mean may be approximated by a Gaussian. Hence, about 1500
samples probably is about right. However, in the absence of any information about the PDF of Xi
beyond the mean and variance, we cannot make any guarantees stronger than that given by the
Chebyshev inequality.
(a) Since XA is a Bernoulli (p = P [A]) random variable,
E [XA ] = P [A] = 0.8, Var[XA ] = P [A] (1 − P [A]) = 0.16. (1)
271
1
n
(b) Let XA,i to denote XA on the ith trial. Since P̂n (A) = Mn (XA ) = n i=1 XA,i ,
1
n
P [A] (1 − P [A])
Var[P̂n (A)] = 2
Var[XA,i ] = . (2)
n n
i=1
(c) Since P̂100 (A) = M100 (XA ), we can use Theorem 7.12(b) to write
$ $
$ $ Var[XA ] 0.16
P $P̂100 (A) − P [A]$ < c ≥ 1 − 2
=1− = 1 − α. (3)
100c 100c2
For c = 0.1, α = 0.16/[100(0.1)2 ] = 0.16. Thus, with 100 samples, our confidence coefficient
is 1 − α = 0.84.
(d) In this case, the number of samples n is unknown. Once again, we use Theorem 7.12(b) to
write $ $
$ $ Var[XA ] 0.16
P $P̂n (A) − P [A]$ < c ≥ 1 − =1− = 1 − α. (4)
nc2 nc2
For c = 0.1, we have confidence coefficient 1 − α = 0.95 if α = 0.16/[n(0.1)2 ] = 0.05, or
n = 320.

Since E[X] = µX = p and Var[X] = p(1 − p), we use Theorem 7.12(b) to write
p(1 − p)
P [|M100 (X) − p| < c] ≥ 1 − = 1 − α. (1)
100c2
For confidence coefficient 0.99, we require
p(1 − p)
≤ 0.01 or c≥ p(1 − p). (2)
100c2
Since p is unknown, we must ensure that the constraint is met
for every value of p. The worst case
occurs at p = 1/2 which maximizes p(1 − p). In this case, c = 1/4 = 1/2 is the smallest value of
c for which we have confidence coefficient of at least 0.99.
If M100 (X) = 0.06, our interval estimate for p is
M100 (X) − c < p < M100 (X) + c. (3)
Since p ≥ 0, M100 (X) = 0.06 and c = 0.5 imply that our interval estimate is
0 ≤ p < 0.56. (4)
Our interval estimate is not very tight because because 100 samples is not very large for a confidence
coefficient of 0.99.

First we observe that the interval estimate can be expressed as
$ $
$ $
$P̂n (A) − P [A]$ < 0.05. (1)
272
Since P̂n (A) = Mn (XA ) and E[Mn (XA )] = P [A], we can use Theorem 7.12(b) to write
$ $
$ $ Var[XA ]
P $P̂n (A) − P [A]$ < 0.05 ≥ 1 − . (2)
n(0.05)2
Note that Var[XA ] = P [A](1 − P [A]) ≤ 0.25. Thus for confidence coefficient 0.9, we require that
Var[XA ] 0.25
1− 2
≥1− ≥ 0.9. (3)
n(0.05) n(0.05)2
This implies n ≥ 1,000 samples are needed.

Both questions can be answered using the following equation from Example 7.6:
$ $ P [A] (1 − P [A])
$ $
P $P̂n (A) − P [A]$ ≥ c ≤ (1)
nc2
The unusual part of this problem is that we are given the true value of P [A]. Since P [A] = 0.01,
we can write $ $ 0.0099
$ $
P $P̂n (A) − P [A]$ ≥ c ≤ (2)
nc2
(a) In this part, we meet the requirement by choosing c = 0.001 yielding
$ $ 9900
$ $
P $P̂n (A) − P [A]$ ≥ 0.001 ≤ (3)
n
Thus to have confidence level 0.01, we require that 9900/n ≤ 0.01. This requires n ≥ 990,000.
(b) In this case, we meet the requirement by choosing c = 10−3 P [A] = 10−5 . This implies
$ $ P [A] (1 − P [A]) 9.9 × 107
$ $ 0.0099
P $P̂n (A) − P [A]$ ≥ c ≤ = = (4)
nc2 n10−10 n
The confidence level 0.01 is met if 9.9 × 107 /n = 0.01 or n = 9.9 × 109 .

Since the relative frequency of the error event E is P̂n (E) = Mn (XE ) and E[Mn (XE )] = P [E], we
can use Theorem 7.12(a) to write
$ $ Var[X ]
$ $ E
P $P̂n (A) − P [E]$ ≥ c ≤ . (1)
nc2
Note that Var[XE ] = P [E](1 − P [E]) since XE is a Bernoulli (p = P [E]) random variable. Using
the additional fact that P [E] ≤ and the fairly trivial fact that 1 − P [E] ≤ 1, we can conclude that
Var[XE ] = P [E] (1 − P [E]) ≤ P [E] ≤ . (2)
Thus $ $ Var[X ]
$ $ E
P $P̂n (A) − P [E]$ ≥ c ≤ 2
≤ 2. (3)
nc nc
273
In this problem, we have to keep straight that the Poisson expected value α = 1 is a different
α than the confidence coefficient 1 − α. That said, we will try avoid using α for the confidence
coefficient. Using X to denote the Poisson (α = 1) random variable, the trace of the sample mean
is the sequence M1 (X), M2 (X), . . . The confidence interval estimate of α has the form
Mn (X) − c ≤ α ≤ Mn (X) + c. (1)
The confidence coefficient of the estimate based on n samples is
P [Mn (X) − c ≤ α ≤ Mn (X) + c] = P [α − c ≤ Mn (X) ≤ α + c] (2)

= P [−c ≤ Mn (X) − α ≤ c] . (3)
Since Var[Mn (X)] = Var[X]/n = 1/n, the 0.9 confidence interval shrinks with increasing n. In
particular, c = cn will be a decreasing sequence. Using a Central Limit Theorem approximation, a
0.9 confidence implies
0 1
−cn Mn (X) − α cn
0.9 = P ≤ ≤ (4)
1/n 1/n 1/n
√ √ √
= Φ(cn n) − Φ(−cn n) = 2Φ(cn n) − 1. (5)
√ √
Equivalently, Φ(cn n) = 0.95 or cn = 1.65/ n.
Thus, as a function of the number of samples n, we plot three functions: the sample mean
√ √
Mn (X), and the upper limit Mn (X) + 1.65/ n and lower limit Mn (X) − 1.65/ n of the 0.9
confidence interval. We use the Matlab function poissonmeanseq(n) to generate these sequences
for n sample values.
function M=poissonmeanseq(n);
x=poissonrv(1,n);
nn=(1:n)’;
M=cumsum(x)./nn;
r=(1.65)./sqrt(nn);
plot(nn,M,nn,M+r,nn,M-r);
Here are two output graphs:
2 3
2
1
Mn(X)
Mn(X)
1
0
0
−1
−1
−2 −2
0 20 40 60 0 200 400 600
n n
poissonmeanseq(60) poissonmeanseq(600)
274
For a Bernoulli (p = 1/2) random variable X, the sample mean Mn (X) is the fraction of successes
in n Bernoulli trials. That is, Mn (X) = Kn /n where Kn is a binomial (n, p = 1/2) random variable.
Thus the probability the sample mean is within one standard error of (p = 1/2) is
√ √
n n n n
pn = P − ≤ Kn ≤ + (1)
2 2 2 2
√ √
n n n n
= P Kn ≤ + − P Kn < − (2)
2 2 2 2
√ % √ &
n n n n
= FKn + − FKn − −1 (3)
2 2 2 2
Here is a Matlab function that graphs pn as a function of n for N steps alongside the output
graph for bernoullistderr(50).
function p=bernoullistderr(N); 1
p=zeros(1,N); 0.9
for n=1:N,
r=[ceil((n-sqrt(n))/2)-1; ... 0.8
pn
(n+sqrt(n))/2];
0.7
p(n)=diff(binomialcdf(n,0.5,r));
end 0.6
plot(1:N,p);
ylabel(’\it p_n’); 0.5
0 10 20 30 40 50
xlabel(’\it n’); n
The alternating up-down sawtooth pattern is a consequence of the CDF of Kn jumping up at

√ √
integer values. In particular, depending on the value of n/2, every other value of (n ± n)/2
exceeds the next integer and increases pn . The less frequent but larger jumps in pn occur when
√
n is an integer. For very large n, the central lmit theorem takes over since the CDF of Mn (X)
converges to a Gaussian CDF. In this case, the sawtooth pattern dies out and pn will converge
to 0.68, the probability that a Gaussian random variable is within one standard deviation of its
expected value.
Finally, we note that the sawtooth pattern is essentially the same as the sawtooth pattern
observed in Quiz 7.5 where we graphed the fraction of 1,000 Bernoulli sample mean traces that
were within one standard error of the true expected value. For 1,000 traces in Quiz 7.5 was so
large, the fraction of the number of traces within one standard error was always very close to the
actual probability pn .

First, we need to determine whether the relative performance of the two estimators depends on the
actual value of λ. To address this, we observe that if Y is an exponential (1) random variable, then
Theorem 3.20 tells us that X = Y /λ is an exponential (λ) random variable. Thus if Y1 , Y2 , . . . are
iid samples of Y , then Y1 /λ, Y2 /λ, . . . are iid samples of X. Moreover, the sample mean of X is
1
n
1
Mn (X) = Yi = Mn (Y ) (1)
nλ λ
i=1
Similarly, the sample variance of X satisfies

2
1 1 V (Y )
n n
Yi 1
Vn (X) = (Xi − Mn (X))2 = − Mn (Y ) = n 2 (2)
n−1 n−1 λ λ λ
i=1 i=1
275
We can conclude that
λ λ
λ̂ = λ̃ = (3)
Mn (Y ) Vn (Y )
For λ = 1, the estimators λ̂ and λ̃ are just scaled versions of the estimators for the case λ = 1.
Hence it is sufficient to consider only the λ = 1 case. The function z=lamest(n,m) returns the
estimation errors for m trials of each estimator where each trial uses n iid exponential (1) samples.
function z=lamest(n,m); In lamest.m, each column of matrix x represents one trial.

x=exponentialrv(1,n*m); Note that mx is a row vector such that mx(i) is the sample
x=reshape(x,n,m); mean for trial i. The matrix MX has mx(i) for every element
mx=sum(x)/n; in column i. Thus vx is a row vector such that vx(i) is the
MX=ones(n,1)*mx;
sample variance for trial i.
vx=sum((x-MX).^2)/(n-1);
z=[(1./mx); (1./sqrt(vx))]-1;
Finally, z is a 2 × m matrix such that column i of z records the estimation errors for trial i. If
λ̂i and λ̃i are the estimates for for trial i, then z(1,i) is the error Ẑi = λ̂i − 1 while z(2,i) is the
error Z̃i = λ̃i − 1.
Now that we can simulate the errors generated by each estimator, we need to determine which
estimator is better. We start by using the commands
z=lamest(1000,1000);
plot(z(1,:),z(2,:),’bd’)
to perform 1,000 trials, each using 1,000 samples. The plot command generates a scatter plot of
the error pairs (Ẑi , Z̃i ) for each trial. Here is an example of the resulting scatter plot:
0.2
0.1
z(2,i)
−0.1
−0.2
−0.1 −0.05 0 0.05 0.1 0.15
z(1,i)
In the scatter plot, each diamond marks an independent pair (Ẑ, Z̃) where Ẑ is plotted on the x-axis
and Z̃ is plotted on the y-axis. (Although it is outside the scope of this solution, it is interesting
to note that the errors Ẑ and Z̃ appear to be positively correlated.) From the plot, it may not be
obvious that one estimator is better than the other. However, by reading the axis ticks carefully,
one can observe that it appears that typical values for Ẑ are in the interval (−0.05, 0.05) while
typical values for Z̃ are in the interval (−0.1, 0.1). This suggests that Ẑ may be superior. To verify
this observation, we calculate the sample mean for each squared errors
1 2 1 2
m m
Mm (Ẑ 2 ) = Ẑi Mm (Z̃ 2 ) = Z̃i (4)
m m
i=1 i=1
From our Matlab experiment with m = 1,000 trials, we calculate
276
>> sum(z.^2,2)/1000
ans =
0.0010
0.0021
That is, M1,000 (Ẑ 2 ) = 0.0010 and M1,000 (Z̃ 2 ) = 0.0021. In fact, one can show (with a lot of work)
for large m that
Mm (Ẑ 2 ) ≈ 1/m Mm (Z̃ 2 ) = 2/m (5)
and that
Mm (Z̃ 2 )
lim = 2. (6)
m→∞ Mm (Ẑ 2 )
In short, the mean squared error of the λ̃ estimator is twice that of the λ̂ estimator.

For the sample covariance matrix
1
n
R̂(n) = X(m)X (m), (1)
n
m=1
we wish to use a Matlab simulation to estimate the probability

$ $
$ $
p(n) = P max $R̂ij − Iij $ ≥ 0.05 . (2)
i,j
(In the original printing, 0.05 was 0.01 but that requirement demanded that n be so large that
most installations of Matlab would grind to a halt on the calculations.)
The Matlab program uses a matrix algebra identity that may (or may not) be familiar. For a
matrix
X = x(1) x(2) · · · x(n) , (3)
with columns x(i), we can write

n
XX = x(i)x (i). (4)
i=1
function p=diagtest(n,t); The Matlab function diagtest(n,t) per-

y=zeros(t,1); p=[ ]; forms the calculation by generating t sam-
for ntest=n, ple covariance matrices, each using n vectors
disp(ntest) x(i) in each sample covariance matrix. These
for i=1:t,
n vectors are generated by the gaussvector
X=gaussvector(0,eye(10),ntest);
R=X*(X’)/ntest; function. The vector x(i) is the ith column
y(i)=(max(max(abs(R-eye(10))))>= 0.05); of the matrix X.
end
p=[p, sum(y)/t];
end
semilogy(n,p);
xlabel(’\it n’); ylabel(’\it p(n)’);
The function diagtest estimates p(n) as the fraction of t trials for which the threshold 0.05 is
exceeded. In addition, if the input n to diagtest is a vector, then the experiment is repeated for
each value in the vector n.
The following commands, estimate p(n) for n ∈ {10, 100, 1000, 10000} using t = 2000 trials:
277
n=[10 100 1000 10000];
p=diagtest(n,2000);
The output is
p=
1.0000 1.0000 1.0000 0.0035
We see that p(n) goes from roughly 1 to almost 0 in going from n = 1,000 to n = 10,000. To
investigate this transition more carefully, we execute the commands
nn=1000:500:10000;
p=diagtest(nn,2000);
The output is shown in the following graph. We use a semilog plot to emphasize differences when
p(n) is close to zero.
0
10
−1
10
p(n)
−2
10
−3
10
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
n
Beyond n = 1,000, the probability p(n) declines rapidly. The “bumpiness” of the graph for large n
occurs because the probability p(n) is small enough that out of 2,000 trials, the 0.05 threshold is
exceeded only a few times.
Note that if x has dimension greater than 10, then the value of n needed to ensure that p(n) is
small would increase.

The difficulty in this problem is that although E[X] exists, EX 2 and higher order moments are
infinite. Thus Var[X] is also infinite. It also follows for any finite n that the sample mean Mn (X)
has infinite variance. In this case, we cannot apply the Chebyshev inequality to the sample mean
to show the convergence in probability of Mn (X) to E[X].
If limn→∞ P [|Mn (X) − E[X]| ≥ ] = p, then there are two distinct possibilities:
• p > 0, or
• p = 0 but the Chebyshev inequality isn’t a sufficient powerful technique to verify this fact.
To resolve whether p = 0 (and the sample mean converges to the expected value) one can spend
time trying to prove either p = 0 or p > 0. At this point, we try some simulation experiments to
see if the experimental evidence points one way or the other.
As requested by the problem, we implement a Matlab function samplemeantest(n,a) to
simulate one hundred traces of the sample mean when E[X] = a. Each trace is a length n sequence
M1 (X), M2 (X), . . . , Mn (X).
278
function mx=samplemeantest(n,a); The n × 100 matrix x consists of iid samples of
u=rand(n,100); X. Taking cumulative sums along each column of
x=a-2+(1./sqrt(1-u)); x, and dividng row i by i, each column of mx is
d=((1:n)’)*ones(1,100); a length n sample mean trace. we then plot the
mx=cumsum(x)./d;
traces.
plot(mx);
xlabel(’\it n’); ylabel(’\it M_n(X)’);
axis([0 n a-1 a+1]);
The following graph was generated by samplemeantest(1000,5):
5.5
Mn(X)
4.5
4
0 100 200 300 400 500 600 700 800 900 1000
n
Frankly, it is difficult to draw strong conclusions from the graph. If the sample sequences Mn (X)
are converging to E[X], the convergence is fairly slow. Even after averaging 1,000 samples, typical
values for the sample mean appear to range from a − 0.5 to a + 0.5. There may also be outlier
sequences which are still off the charts since we truncated the y-axis range. On the other hand, the
sample mean sequences do not appear to be diverging (which is also possible since Var[X] = ∞.)
Note the above graph was generated using 105 sample values. Repeating the experiment with more
samples, say samplemeantest(10000,5), will yield a similarly inconclusive result. Even if your
version of Matlab can support the generation of 100 times as many samples, you won’t know for
sure whether the sample mean sequence always converges. On the other hand, the experiment is
probably enough that if you pursue the analysis, you should start by trying to prove that p = 0.
(This will make a a fun problem for the third edition!)
279

Assuming the coin is fair, we must choose a rejection region R such that α = P [R] = 0.05. We can
choose a rejection region R = {L > r}. What remains is to choose r so that P [R] = 0.05. Note
that L > l if we first observe l tails in a row. Under the hypothesis that the coin is fair, l tails in a
row occurs with probability
P [L > l] = (1/2)l (1)
Thus, we need
P [R] = P [L > r] = 2−r = 0.05 (2)
Thus, r = − log2 (0.05) = log2 (20) = 4.32. In this case, we reject the hypothesis that the coin is
fair if L ≥ 5. The significance level of the test is α = P [L > 4] = 2−4 = 0.0625 which close to but
not exactly 0.05.
The shortcoming of this test is that we always accept the hypothesis that the coin is fair
whenever heads occurs on the first, second, third or fourth flip. If the coin was biased such that
the probability of heads was much higher than 1/2, say 0.8 or 0.9, we would hardly ever reject the
hypothesis that the coin is fair. In that sense, our test cannot identify that kind of biased coin.
(a) We wish to develop a hypothesis test of the form

P [|K − E [K] | > c] = 0.05 (1)
to determine if the coin we’ve been flipping is indeed a fair one. We would like to find the
value of c, which will determine the upper and lower limits on how many heads we can get
away from the expected number out of 100 flips and still accept our hypothesis. Under our
fair coin hypothesis, the expected number of heads, and the standard deviation of the process
are

E [K] = 50, σK = 100 · 1/2 · 1/2 = 5. (2)
Now in order to find c we make use of the central limit theorem and divide the above inequality
through by σK to arrive at

|K − E [K] | c
P > = 0.05 (3)
σK σK
Taking the complement, we get

c K − E [K] c
P − ≤ ≤ = 0.95 (4)
σK σK σK
Using the Central Limit Theorem we can write

c −c c
Φ −Φ = 2Φ − 1 = 0.95 (5)
σK σK σK
This implies Φ(c/σK ) = 0.975 or c/5 = 1.96. That is, c = 9.8 flips. So we see that if we
observe more then 50 + 10 = 60 or less then 50 − 10 = 40 heads, then with significance level
α ≈ 0.05 we should reject the hypothesis that the coin is fair.
280
(b) Now we wish to develop a test of the form
P [K > c] = 0.01 (6)
Thus we need to find the value of c that makes the above probability true. This value will
tell us that if we observe more than c heads, then with significance level α = 0.01, we should
reject the hypothesis that the coin is fair. To find this value of c we look to evaluate the CDF
k

100
FK (k) = (1/2)100 . (7)
i
i=0
Computation reveals that c ≈ 62 flips. So if we observe 62 or greater heads, then with a

significance level of 0.01 we should reject the fair coin hypothesis. Another way to obtain
this result is to use a Central Limit Theorem approximation. First, we express our rejection
region in terms of a zero mean, unit variance random variable.

K − E [K] c − E [K]
P [K > c] = 1 − P [K ≤ c] = 1 − P ≤ = 0.01 (8)
σK σK
Since E[K] = 50 and σK = 5, the CLT approximation is

c − 50
P [K > c] ≈ 1 − Φ = 0.01 (9)
5
From Table 3.1, we have (c − 50)/5 = 2.35 or c = 61.75. Once again, we see that we reject
the hypothesis if we observe 62 or more heads.

A reasonable test would reject the null hypothesis that the plant is operating normally if one or
more of the chips fail the one-day test. Exactly how many shold be tested and how many failures
N are needed to reject the null hypothesis would depend on the significance level of the test.
(a) The lifetime of a chip is X, an exponential (λ) random variable with λ = (T /200)2 . The
probability p that a chip passes the one-day test is
p = P [X ≥ 1/365] = e−λ/365 . (1)
For an m chip test, the significance level of the test is
α = P [N > 0] = 1 − P [N = 0] = 1 − pm = 1 − e−mλ/365 . (2)
(b) At T = 100◦ , λ = 1/4 and we obtain a significance level of α = 0.01 if
365 ln(0.99) 3.67

m=− = = 14.74. (3)
λ λ
In fact, at m = 15 chips, the significance level is α = 0.0102.
(c) Raising T raises the failure rate λ = (T /200)2 and thus lowers m = 3.67/λ. In essence, raising
the temperature makes a “tougher” test and thus requires fewer chips to be tested for the
same significance level.
281
(a) The rejection region is R = {T > t0 }. The duration of a voice call has exponential PDF

(1/3)e−t/3 t ≥ 0,
fT (t) = (1)
0 otherwise.
The significance level of the test is
' ∞
α = P [T > t0 ] = fT (t) dt = e−t0 /3 . (2)
t0
(b) The significance level is α = 0.05 if t0 = −3 ln α = 8.99 minutes.

In order to test just a small number of pacemakers, we test n pacemakers and we reject the null
hypothesis if any pacemaker fails the test. Moreover, we choose the smallest n such that we meet
the required significance level of the test.
The number of pacemakers that fail the test is X, a binomial (n, q0 = 10−4 ) random variable.
The significance level of the test is
α = P [X > 0] = 1 − P [X = 0] = 1 − (1 − q0 )n . (1)
For a significance level α = 0.01, we have that

ln(1 − α)
n= = 100.5. (2)
ln(1 − q0 )
Comment: For α = 0.01, keep in mind that there is a one percent probability that a normal
factory will fail the test. That is, a test failure is quite unlikely if the factory is operating normally.

Since the null hypothesis H0 asserts that the two exams have the same mean and variance, we
reject H0 if the difference in sample means is large. That is, R = {|D| ≥ d0 }.
Under H0 , the two sample means satisfy
σ2 100
E [MA ] = E [MB ] = µ, Var[MA ] = Var[MB ] = = (1)
n n
Since n is large, it is reasonable to use the Central Limit Theorem to approximate MA and MB as
Gaussian random variables. Since MA and MB are independent, D is also Gaussian with
200
E [D] = E [MA ] − E [MB ] = 0 Var[D] = Var[MA ] + Var[MB ] = . (2)
n
Under the Gaussian assumption, we can calculate the significance level of the test as
α = P [|D| ≥ d0 ] = 2 (1 − Φ(d0 /σD )) . (3)

For α = 0.05, Φ(d0 /σD ) = 0.975, or d0 = 1.96σD = 1.96 200/n. If n = 100 students take each
exam, then d0 = 2.77 and we reject the null hypothesis that the exams are the same if the sample
means differ by more than 2.77 points.
282
For the MAP test, we must choose acceptance regions A0 and A1 for the two hypotheses H0 and
H1 . From Theorem 8.2, the MAP rule is
PN |H0 (n) P [H1 ]
n ∈ A0 if ≥ ; n ∈ A1 otherwise. (1)
PN |H1 (n) P [H0 ]
Since PN |Hi (n) = λni e−λi /n!, the MAP rule becomes
n
λ0 P [H1 ]
n ∈ A0 if e−(λ0 −λ1 ) ≥ ; n ∈ A1 otherwise. (2)
λ1 P [H0 ]
By taking logarithms and assuming λ1 > λ0 yields the final form of the MAP rule
λ1 − λ0 + ln(P [H0 ] /P [H1 ])
n ∈ A0 if n ≤ n∗ = ; n ∈ A1 otherwise. (3)
ln(λ1 /λ0 )
From the MAP rule, we can get the ML rule by setting the a priori probabilities to be equal. This
yields the ML rule
λ1 − λ0
n ∈ A0 if n ≤ n∗ = ; n ∈ A1 otherwise. (4)
ln(λ1 /λ0 )

Hypotheses H0 and H1 have a priori probabilities P [H0 ] = 0.8 and P [H1 ] = 0.2 and likelihood
functions

(1/3)e−t/3 t ≥ 0, (1/µD )e−t/µD t ≥ 0,
fT |H0 (t) = fT |H1 (t) = (1)
otherwise, otherwise,
The acceptance regions are A0 = {t|T ≤ t0 } and A1 = {t|t > t0 }.

(a) The false alarm probability is
' ∞
PFA = P [A1 |H0 ] = fT |H0 (t) dt = e−t0 /3 . (2)
t0
(b) The miss probability is

' t0
PMISS = P [A0 |H1 ] = fT |H1 (t) dt = 1 − e−t0 /µD . (3)
0
(c) From Theorem 8.6, the maximum likelihood decision rule is

fT |H0 (t)
t ∈ A0 if ≥ 1; t ∈ A1 otherwise. (4)
fT |H1 (t)
After some algebra, this rule simplifies to

ln(µD /3)
t ∈ A0 if t ≤ tM L = ; t ∈ A1 otherwise. (5)
1/3 − 1/µD
When µD = 6 minutes, tM L = 6 ln 2 = 4.16 minutes. When µD = 10 minutes, tM L =
(30/7) ln(10/3) = 5.16 minutes.
283
(d) The ML rule is the same as the MAP rule when P [H0 ] = P [H1 ]. When P [H0 ] > P [H1 ], the
MAP rule (which minimizes the probability of an error) should enlarge the A0 acceptance
region. Thus we would expect tMAP > tM L .
(e) From Theorem 8.2, the MAP rule is
fT |H0 (t) P [H1 ] 1

t ∈ A0 if ≥ = ; t ∈ A1 otherwise. (6)
fT |H1 (t) P [H0 ] 4
This rule simplifies to

ln(4µD /3)
t ∈ A0 if t ≤ tMAP = ; t ∈ A1 otherwise. (7)
1/3 − 1/µD
When µD = 6 minutes, tMAP = 6 ln 8 = 12.48 minutes. When µD = 10 minutes, tM L =
(30/7) ln(40/3) = 11.1 minutes.
(f) For a given threshold t0 , we learned in parts (a) and (b) that
PFA = e−t0 /3 , ¶MISS = 1 − e−t0 /µD . (8)
The Matlab program rocvoicedataout graphs both receiver operating curves. The program
and the resulting ROC are shown here.
t=0:0.05:30; 1
PFA= exp(-t/3); µD=6
0.8 µD=10
PMISS6= 1-exp(-t/6);
PMISS10=1-exp(-t/10); 0.6
PMISS
plot(PFA,PMISS6,PFA,PMISS10);
0.4
legend(’\mu_D=6’,’\mu_D=10’);
xlabel(’\itP_{\rmFA}’); 0.2
ylabel(’\itP_{\rmMISS}’);
0
0 0.5 1
PFA
As one might expect, larger µD resulted in reduced PMISS for the same PFA .

By Theorem 8.5, the decision rule is
PN |H0 (n)
n ∈ A0 if L(n) = ≥ γ; n ∈ A1 otherwise, (1)
PN |H1 (n)

where where γ is the largest possible value such that L(n)<γ PN |H0 (n) ≤ α.
Given H0 , N is Poisson (a0 = 1,000) while given H1 , N is Poisson (a1 = 1,300). We can solve
for the acceptance set A0 by observing that n ∈ A0 if
PN |H0 (n) an e−a0 /n!
= 0n −a1 ≥ γ. (2)
PN |H1 (n) a1 e /n!
Cancelling common factors and taking the logarithm, we find that n ∈ A0 if

a0
n ln ≥ (a0 − a1 ) + ln γ. (3)
a1
284
Since ln(a0 /a1 ) < 0, divind through reverses the inequality and shows that
(a0 − a1 ) + ln γ (a1 − a0 ) − ln γ
n ∈ A0 if n ≤ n∗ = = ; n ∈ A1 otherwise (4)
ln(a0 /a1 ) ln(a1 /a0 )
However, we still need to determine the constant γ. In fact, it is easier to work with the threshold
n∗ directly. Note that L(n) < γ if and only if n > n∗ . Thus we choose the smallest n∗ such that

P [N > n∗ |H0 ] = PN |H0 (n) α ≤ 10−6 . (5)
n>n∗
To find n∗ a reasonable approach would be to use Central Limit Theorem approximation since
given H0 , N is a Poisson (1,000) random variable, which has the same PDF as the sum on 1,000
independent Poisson (1) random variables. Given H0 , N has expected value a0 and variance a0 .
From the CLT,
∗
∗ N − a0 n∗ − a0 n − a0
P [N > n |H0 ] = P √ > √ |H0 ≈ Q √ ≤ 10−6 . (6)
a0 a0 a0
From Table 3.2, Q(4.75) = 1.02 × 10−6 and Q(4.76) < 10−6 , implying
√
n∗ = a0 + 4.76 a0 = 1150.5. (7)
On the other hand, perhaps the CLT should be used with some caution since α = 10−6 implies we
are using the CLT approximation far from the center of the distribution. In fact, we can check out
answer using the poissoncdf functions:
>> nstar=[1150 1151 1152 1153 1154 1155];
>> (1.0-poissoncdf(1000,nstar))’
ans =
1.0e-005 *
0.1644 0.1420 0.1225 0.1056 0.0910 0.0783
>>
Thus we see that n∗ 1154. Using this threshold, the miss probability is
P [N ≤ n∗ |H1 ] = P [N ≤ 1154|H1 ] = poissoncdf(1300,1154) = 1.98 × 10−5 . (8)
Keep in mind that this is the smallest possible PMISS subject to the constraint that PFA ≤ 10−6 .
(a) Given H0 , X is Gaussian (0, 1). Given H1 , X is Gaussian (4, 1). From Theorem 8.2, the
MAP hypothesis test is
2 /2
fX|H0 (x) e−x
x ∈ A0 if = ; x ∈ A1 otherwise. (1)
fX|H1 (x) P [H1 ]
e−(x−4)2 /2 ≥
P [H0 ]
Since a target is present with probability P [H1 ] = 0.01, the MAP rule simplifies to

1 P [H1 ]
x ∈ A0 if x ≤ xMAP = 2 − ln = 3.15; x ∈ A1 otherwise. (2)
4 P [H0 ]
285
The false alarm and miss probabilities are
PFA = P [X ≥ xMAP |H0 ] = Q(xMAP ) = 8.16 × 10−4 (3)

PMISS = P [X < xMAP |H1 ] = Φ(xMAP − 4) = 1 − Φ(0.85) = 0.1977. (4)
The average cost of the MAP policy is
E [CMAP ] = C10 PFA P [H0 ] + C01 PMISS P [H1 ] (5)

−4
= (1)(8.16 × 10 4
)(0.99) + (10 )(0.1977)(0.01) = 19.77. (6)
(b) The cost of a false alarm is C10 = 1 unit while the cost of a miss is C01 = 104 units. From
Theorem 8.3, we see that the Minimum Cost test is the same as the MAP test except the
P [H0 ] is replaced by C10 P [H0 ] and P [H1 ] is replaced by C01 P [H1 ]. Thus, we see from thr
MAP test that the minimum cost test is

1 C01 P [H1 ]
x ∈ A0 if x ≤ xMC = 2 − ln = 0.846; x ∈ A1 otherwise. (7)
4 C10 P [H0 ]
The false alarm and miss probabilities are
PFA = P [X ≥ xMC |H0 ] = Q(xMC ) = 0.1987 (8)

−4
PMISS = P [X < xMC |H1 ] = Φ(xMC − 4) = 1 − Φ(3.154) = 8.06 × 10 . (9)
The average cost of the minimum cost policy is
E [CMC ] = C10 PFA P [H0 ] + C01 PMISS P [H1 ] (10)

= (1)(0.1987)(0.99) + (104 )(8.06 × 10−4 )(0.01) = 0.2773. (11)
Because the cost of a miss is so high, the minimum cost test greatly reduces the miss proba-
bility, resulting in a much lower average cost than the MAP test.

Given H0 , X is Gaussian (0, 1). Given H1 , X is Gaussian (v, 1). By Theorem 8.4, the Neyman-
Pearson test is
2
fX|H0 (x) e−x /2
x ∈ A0 if L(x) = = −(x−v)2 /2 ≥ γ; x ∈ A1 otherwise. (1)
fX|H1 (x) e
2 −(x−v)2 ]/2 2 /2
x ∈ A0 if L(x) = e−[x = e−vx+v ≥ γ; x ∈ A1 otherwise. (2)
Taking logarithms, the Neyman-Pearson rule becomes

v 1
x ∈ A0 if x ≤ x0 = − ln γ; x ∈ A1 otherwise. (3)
2 v
The choice of γ has a one-to-one correspondence with the choice of the threshold x0 . Moreoever
L(x) ≥ γ if and only if x ≤ x0 . In terms of x0 , the false alarm probability is
PFA = P [L(X) < γ|H0 ] = P [X ≥ x0 |H0 ] = Q(x0 ). (4)
Thus we choose x0 such that Q(x0 ) = α.
286
Given H0 , Mn (T ) has expected value E[V ]/n = 3/n and variance Var[V ]/n = 9/n. Given H1 ,
Mn (T ) has expected value E[D]/n = 6/n and variance Var[D]/n = 36/n.
(a) Using a Central Limit Theorem approximation, the false alarm probability is
0 1
Mn (T ) − 3 t0 − 3 √
PFA = P [Mn (T ) > t0 |H0 ] = P > = Q( n[t0 /3 − 1]). (1)
9/n 9/n
(b) Again, using a CLT Approximation, the miss probability is

0 1
Mn (T ) − 6 t0 − 6 √
PMISS = P [Mn (T ) ≤ t0 |H1 ] = P ≤ = Φ( n[t0 /6 − 1]). (2)
36/n 36/n
(c) From Theorem 8.6, the maximum likelihood decision rule is
fMn (T )|H0 (t)

t ∈ A0 if ≥ 1; t ∈ A1 otherwise. (3)
fMn (T )|H1 (t)
We will see shortly that using a CLT approximation for the likelihood functions is something
of a detour. Nevertheless, with a CLT approximation, the likelihood functions are
) )
n −n(t−3)2 /18 n −n(t−6)2 /72
fMn (T )|H0 (t) = e fMn (T )|H1 (t) = e (4)
18π 72π
From the CLT approximation, the ML decision rule is

) 2
72 e−n(t−3) /18 −n[4(t−3)2 −(t−6)2 ]/72
t ∈ A0 if −n(t−6)2 /72 = 2e ≥ 1; t ∈ A1 otherwise. (5)
18 e
After some algebra, this rule simplifies to
24 ln 2
t ∈ A0 if t2 − 4t − ≤ 0; t ∈ A1 otherwise. (6)
n
Since the quadratic t2 −4t−24 ln(2)/n has two zeros, we use the quadratic formula to find the
roots. One root corresponds to a negative value of t and can be discarded since Mn (T ) ≥ 0.
Thus the ML rule (for n = 9) becomes

t ∈ A0 if t ≤ tM L = 2 + 2 1 + 6 ln(2)/n = 4.42; t ∈ A1 otherwise. (7)
The negative root of the quadratic is the result of the Gaussian assumption which allows for
a nonzero probability that Mn (T ) will be negative. In this case, hypothesis H1 which has
higher variance becomes more likely. However, since Mn (T ) ≥ 0, we can ignore this root since
it is just an artifact of the CLT approximation.
In fact, the CLT approximation gives an incorrect answer. Note that Mn (T ) = Yn /n where
Yn is a sum of iid exponential random variables. Under hypothesis H0 , Yn is an Erlang
(n, λ0 = 1/3) random variable. Under hypothesis H1 , Yn is an Erlang (n, λ1 = 1/6) random
variable. Since Mn (T ) = Yn /n is a scaled version of Yn , Theorem 3.20 tells us that given
287
hypothesis Hi , Mn (T ) is an Erlang (n, nλi ) random variable. Thus Mn (T ) has likelihood
functions
(nλi )n tn−1 e−nλi t
(n−1)! t≥0
fMn (T )|Hi (t) = (8)
0 otherwise
Using the Erlang likelihood functions, the ML rule becomes
n
fMn (T )|H0 (t) λ0
t ∈ A0 if = e−n(λ0 −λ1 )t ≥ 1; t ∈ A1 otherwise. (9)
fMn (T )|H1 (t) λ1
ln(λ0 /λ1 )
t ∈ A0 if t ≤ tML = = 6 ln 2 = 4.159; t ∈ A1 otherwise. (10)
λ0 − λ1
Since 6 ln 2 = 4.159, this rule is not the same as the rule derived using a CLT approximation.
Using the exact Erlang PDF, the ML rule does not depend on n. Moreoever, even if n → ∞,
the exact Erlang-derived rule and the CLT approximation rule remain different. In fact, the
CLT-based rule is simply an approximation to the correct rule. This highlights that we should
first check whether a CLT approximation is necessary before we use it.
(d) In this part, we will use the exact Erlang PDFs to find the MAP decision rule. From 8.2, the
MAP rule is
n
fMn (T )|H0 (t) λ0 P [H1 ]
t ∈ A0 if = e−n(λ0 −λ1 )t ≥ ; t ∈ A1 otherwise. (11)
fMn (T )|H1 (t) λ1 P [H0 ]
Since P [H0 ] = 0.8 and P [H1 ] = 0.2, the MAP rule simplifies to
ln λλ01 − 1
ln PP [H 1]
n [H0 ] ln 4
t ∈ A0 if t ≤ tMAP = = 6 ln 2 + ; t ∈ A1 otherwise. (12)
λ0 − λ1 n
For n = 9, tMAP = 5.083.
(e) Although we have seen it is incorrect to use a CLT approximation to derive the decision
rule, the CLT approximation used in parts (a) and (b) remains a good way to estimate the
false alarm and miss probabilities. However, given Hi , Mn (T ) is an Erlang (n, nλi ) random
variable. In particular, given H0 , Mn (T ) is an Erlang (n, n/3) random variable while given
H1 , Mn (T ) is an Erlang (n, n/6). Thus we can also use erlangcdf for an exact calculation
of the false alarm and miss probabilities. To summarize the results of parts (a) and (b), a
threshold t0 implies that
√
PFA = P [Mn (T ) > t0 |H0 ] = 1-erlangcdf(n,n/3,t0) ≈ Q( n[t0 /3 − 1]), (13)
√
PMISS = P [Mn (T ) ≤ t0 |H1 ] = erlangcdf(n,n/6,t0) ≈ Φ( n[t0 /6 − 1]). (14)
(15)
Here is a program that generates the receiver operating curve.
288
%voicedatroc.m
t0=1:0.1:8’;
n=9;
PFA9=1.0-erlangcdf(n,n/3,t0);
PFA9clt=1-phi(sqrt(n)*((t0/3)-1));
PM9=erlangcdf(n,n/6,t0);
PM9clt=phi(sqrt(n)*((t0/6)-1));
n=16;
PFA16=1.0-erlangcdf(n,n/3,t0);
PFA16clt=1.0-phi(sqrt(n)*((t0/3)-1));
PM16=erlangcdf(n,n/6,t0);
PM16clt=phi(sqrt(n)*((t0/6)-1));
plot(PFA9,PM9,PFA9clt,PM9clt,PFA16,PM16,PFA16clt,PM16clt);
axis([0 0.8 0 0.8]);
legend(’Erlang n=9’,’CLT n=9’,’Erlang n=16’,’CLT n=16’);
Here are the resulting ROCs.
0.8
Erlang n=9
0.6 CLT n=9
Erlang n=16
0.4 CLT n=16
0.2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Both the true curve and CLT-based approximations are shown. The graph makes it clear
that the CLT approximations are somewhat innaccurate. It is also apparent that the ROC
for n = 16 is clearly better than for n = 9.

This problem is a continuation of Problem 8.2.6. In this case, we say a call is a “success” if T > t0 .
The success probability depends on which hypothesis is true. In particular,
p0 = P [T > t0 |H0 ] = e−t0 /3 , p1 = P [T > t0 |H1 ] = e−t0 /6 . (1)
Under hypothesis Hi , K has the binomial (n, pi ) PMF

n k
PK|Hi (k) = p (1 − pi )n−k . (2)
k i
(a) A false alarm occurs if K > k0 under hypothesis H0 . The probability of this event is
n
n k
PFA = P [K > k0 |H0 ] = p (1 − p0 )n−k , (3)
k 0
k=k0 +1
(b) From Theorem 8.6, the maximum likelihood decision rule is

PK|H0 (k) pk (1 − p0 )n−k
k ∈ A0 if = 0k ≥ 1; k ∈ A1 otherwise. (4)
PK|H1 (k) p1 (1 − p1 )n−k
289

p0 /(1 − p0 ) 1 − p1
k ∈ A0 if k ln ≥ ln n; k ∈ A1 otherwise. (5)
p1 /(1 − p1 ) 1 − p0
To proceed further, we need to know if p0 < p1 or if p0 ≥ p1 . For t0 = 4.5,
p0 = e−1.5 = 0.2231 < e−0.75 = 0.4724 = p1 . (6)
In this case, the ML rule becomes

1−p0
ln 1−p1
k ∈ A0 if k ≤ kM L = n = (0.340)n; k ∈ A1 otherwise. (7)
p1 /(1−p1 )
ln p0 /(1−p0 )
For n = 16, kM L = 5.44.

(c) From Theorem 8.2, the MAP test is
PK|H0 (k) pk (1 − p0 )n−k P [H1 ]
k ∈ A0 if = 0k ≥ ; k ∈ A1 otherwise. (8)
PK|H1 (k) p1 (1 − p1 )n−k P [H0 ]
with P [H0 ] = 0.8 and P [H1 ] = 0.2, this rule simplifies to

p0 /(1 − p0 ) 1 − p1
k ∈ A0 if k ln ≥ ln n; k ∈ A1 otherwise. (9)
p1 /(1 − p1 ) 1 − p0
For t0 = 4.5, p0 = 0.2231 < p1 = 0.4724, the MAP rule becomes

ln 1−p 0
1−p1 + ln 4
k ∈ A0 if k ≤ kMAP = n = (0.340)n + 1.22; k ∈ A1 otherwise. (10)
ln pp10 /(1−p1)
/(1−p0 )
For n = 16, kMAP = 6.66.

(d) For threshold k0 , the false alarm and miss probabilities are
PFA = P [K > k0 |H0 ] = 1-binomialcdf(n,p0,k0) (11)
PMISS = P [K ≤ k0 |H1 ] = binomialcdf(n,p1,k0) (12)
The ROC is generated by evaluating PFA and PMISS for each value of k0 . Here is a Matlab
program that does this task and plots the ROC.
function [PFA,PMISS]=binvoicedataroc(n);
t0=[3; 4.5];
p0=exp(-t0/3); p1=exp(-t0/6);
k0=(0:n)’;
PFA=zeros(n+1,2);
for j=1:2,
PFA(:,j) = 1.0-binomialcdf(n,p0(j),k0);
PM(:,j)=binomialcdf(n,p1(j),k0);
end
plot(PFA(:,1),PM(:,1),’-o’,PFA(:,2),PM(:,2),’-x’);
legend(’t_0=3’,’t_0=4.5’);
axis([0 0.8 0 0.8]);
xlabel(’\itP_{\rmFA}’);
ylabel(’\itP_{\rmMISS}’);
290
and here is the resulting ROC:
0.8
t0=3
0.6 t0=4.5
PMISS
0.4
0.2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
PFA
As we see, the test works better with threshold t0 = 4.5 than with t0 = 3.

Given hypothesis H0 that X = 0, Y = W is an exponential (λ = 1) random variable. Given
hypothesis H1 that X = 1, Y = V + W is an Erlang (n = 2, λ = 1) random variable. That is,
−y −y
e y ≥ 0, ye y ≥ 0,
fY |H0 (y) = fY |H1 (y) = (1)
The probability of a decoding error is minimized by the MAP rule. Since P [H0 ] = P [H1 ] = 1/2,
the MAP rule is
fY |H0 (y) e−y P [H1 ]
y ∈ A0 if = −y ≥ = 1; y ∈ A1 otherwise. (2)
fY |H1 (y) ye P [H0 ]
Thus the MAP rule simplifies to
y ∈ A0 if y ≤ 1; y ∈ A1 otherwise. (3)
The probability of error is
PERR = P [Y > 1|H0 ] P [H0 ] + P [Y ≤ 1|H1 ] P [H1 ] (4)

' '
1 ∞ −y 1 1 −y
= e dy + ye dy (5)
2 1 2 0
e−1 1 − 2e−1 1 − e−1
= + = . (6)
2 2 2

Given hypothesis Hi , K has the binomial PMF

n k
PK|Hi (k) = q (1 − qi )n−k . (1)
k i
(a) The ML rule is
k ∈ A0 if PK|H0 (k) > PK|H1 (k) ; k ∈ A1 otherwise. (2)
291
When we observe K = k ∈ {0, 1, . . . , n}, plugging in the conditional PMF’s yields the rule

n k n k
k ∈ A0 if q0 (1 − q0 )n−k
> q (1 − q1 )n−k ; k ∈ A1 otherwise. (3)
k k 1
Cancelling common factors, taking the logarithm of both sides, and rearranging yields
k ∈ A0 if k ln q0 + (n − k) ln(1 − q0 ) > k ln q1 + (n − k) ln(1 − q1 ); k ∈ A1 otherwise. (4)
By combining all terms with k, the rule can be simplified to

q1 /(1 − q1 ) 1 − q0
k ∈ A0 if k ln < n ln ; k ∈ A1 otherwise. (5)
q0 /(1 − q0 ) 1 − q1
Note that q1 > q0 implies q1 /(1 − q1 ) > q0 /(1 − q0 ). Thus, we can rewrite our ML rule as
ln[(1 − q0 )/(1 − q1 )]
k ∈ A0 if k < k ∗ = n ; k ∈ A1 otherwise. (6)
ln[q1 /q0 ] + ln[(1 − q0 )/(1 − q1 )]
(b) Let k ∗ denote the threshold given in part (a). Using n = 500, q0 = 10−4 , and q1 = 10−2 , we
have
ln[(1 − 10−4 )/(1 − 10−2 )]
k ∗ = 500 ≈ 1.078 (7)
ln[10−2 /10−4 ] + ln[(1 − 10−4 )/(1 − 10−2 )]
Thus the ML rule is that if we observe K ≤ 1, then we choose hypothesis H0 ; otherwise, we
choose H1 . The false alarm probability is
PFA = P [A1 |H0 ] = P [K > 1|H0 ] (8)

= 1 − PK|H0 (0) − PK|H1 (1) (9)
= 1 − (1 − q0 ) 500
− 500q0 (1 − q0 ) 499
= 0.0012 (10)
and the miss probability is
PMISS = P [A0 |H1 ] = P [K ≤ 1|H1 ] (11)

= PK|H1 (0) + PK|H1 (1) (12)
= (1 − q1 ) 500
+ 500q1 (1 − q1 ) 499
= 0.0398. (13)
(c) In the test of Example 8.8, the geometric random variable N , the number of tests needed
to find the first failure, was used. In this problem, the binomial random variable K, the
number of failures in 500 tests, was used. We will call these two procedures the geometric
(N ) (N )
and the binomial tests. Also, we will use PFA and PMISS to denote the false alarm and miss
(K) (K)
probabilities using the geometric test. We also use PFA and PMISS for the error probabilities
of the binomial test. From Example 8.8, we have the following comparison:
geometric test binomial test (14)

(N ) (K)
PFA = 0.0045, PFA = 0.0012, (15)
(N ) (K)
PMISS = 0.0087, PMISS = 0.0398 (16)
292
When making comparisons between tests, we want to judge both the reliability of the test
as well as the cost of the testing procedure. With respect to the reliability, we see that the
conditional error probabilities appear to be comparable in that
(N ) (K)
PFA PMISS
(K)
= 3.75 but (N )
= 4.57. (17)
PFA PMISS
Roughly, the false alarm probability of the geometric test is about four times higher than
that of the binomial test. However, the miss probability of the binomial test is about four
times that of the geometric test. As for the cost of the test, it is reasonable to assume the
cost is proportional to the number of disk drives that are tested. For the geometric test of
Example 8.8, we test either until the first failure or until 46 drives pass the test. For the
binomial test, we test until either 2 drives fail or until 500 drives the pass the test! You can,
if you wish, calculate the expected number of drives tested under each test method for each
hypothesis. However, it isn’t necessary in order to see that a lot more drives will be tested
using the binomial test. If we knew the a priori probabilities P [Hi ] and also the relative costs
of the two type of errors, then we could determine which test procedure was better. However,
without that information, it would not be unreasonable to conclude that the geometric test
offers performance roughly comparable to that of the binomial test but with a significant
reduction in the expected number of drives tested.

The key to this problem is to observe that
P [A0 |H0 ] = 1 − P [A1 |H0 ] , P [A1 |H1 ] = 1 − P [A0 |H1 ] . (1)
The total expected cost can be written as

E C = P [A1 |H0 ] P [H0 ] C10

+ (1 − P [A1 |H0 ])P [H0 ] C00 (2)

+ P [A0 |H1 ] P [H1 ] C01 + (1 − P [A0 |H1 ])P [H1 ] C11 . (3)
Rearranging terms, we have

E C = P [A1 |H0 ] P [H0 ] (C10

− C00
) + P [A0 |H1 ] P [H1 ] (C01
− C11 )

+ P [H0 ] C00 + P [H1 ] C11 . (4)
+ P [H ]C does not depend on the acceptance sets A and A , the decision rule
Since P [H0 ]C00 1 11 0 1
that minimizes E[C ] is the same decision rule that minimizes

E C = P [A1 |H0 ] P [H0 ] (C10

− C00
) + P [A0 |H1 ] P [H1 ] (C01
− C11 ). (5)
The decision rule that minimizes E[C ] is the same as the minimum cost test in Theorem 8.3 with
the costs C01 and C10 replaced by the differential costs C01 − C and C − C .
11 10 00

Since the three hypotheses H0 , H1 , and H2 are equally likely, the MAP and ML hypothesis tests
are the same. From Theorem 8.8, the MAP rule is
x ∈ Am if fX|Hm (x) ≥ fX|Hj (x) for all j. (1)
293
2 , the conditional PDF of X given H is
Since N is Gaussian with zero mean and variance σN i
1 2 /2σ 2
fX|Hi (x) = e−(x−a(i−1)) N . (2)
2
2πσN
Thus, the MAP rule is
x ∈ Am if (x − a(m − 1))2 ≤ (x − a(j − 1))2 for all j. (3)
This implies that the rule for membership in A0 is
x ∈ A0 if (x + a)2 ≤ x2 and (x + a)2 ≤ (x − a)2 . (4)
x ∈ A0 if x ≤ −a/2. (5)
Similar rules can be developed for A1 and A2 . These are:
x ∈ A1 if −a/2 ≤ x ≤ a/2 (6)

x ∈ A2 if x ≥ a/2 (7)
To summarize, the three acceptance regions are
A0 = {x|x ≤ −a/2} A1 = {x| − a/2 < x ≤ a/2} A2 = {x|x > a/2} (8)
Graphically, the signal space is one dimensional and the acceptance regions are
A0 A1 A2
s0 s1 s2 X
-a 0 a
Just as in the QPSK system of Example 8.13, the additive Gaussian noise dictates that the ac-
ceptance region Ai is the set of observations x that are closer to si = (i − 1)a than any other
sj .

(1) (2)
Let the components of sijk be denoted by sijk and sijk so that given hypothesis Hijk ,
0 (1) 1
X1 s N1
= ijk
(2) + (1)
X2 s N2
ijk
As in Example 8.13, we will assume N1 and N2 are iid zero mean Gaussian random variables with
variance σ 2 . Thus, given hypothesis Hijk , X1 and X2 are independent and the conditional joint
PDF of X1 and X2 is
fX1 ,X2 |Hijk (x1 , x2 ) = fX1 |Hijk (x1 ) fX2 |Hijk (x2 ) (2)
1 −(x1 −s(1) 2 2 (2) 2
ijk ) /2σ e−(x2 −sijk ) /2σ
2
= 2
e (3)
2πσ
1 −[(x1 −s(1) 2 (2) 2
ijk ) +(x2 −sijk ) ]/2σ
2
= 2
e (4)
2πσ
294
In terms of the distance x − sijk between vectors
0 (1)
1
x sijk
x= 1 sijk = (2) (5)
x2 sijk
we can write
1 −x−sijk 2 /2σ2
fX1 ,X2 |Hi (x1 , x2 ) = e (6)
2πσ 2
Since all eight symbols s000 , . . . , s111 are equally likely, the MAP and ML rules are
x ∈ Aijk if fX1 ,X2 |Hijk (x1 , x2 ) ≥ fX1 ,X2 |Hi j k (x1 , x2 ) for all other Hi j k . (7)
x ∈ Aijk if x − sijk ≤ x − sijk for all other i j k . (8)
This means that Aijk is the set of all vectors x that are closer to sijk than any other signal.
Graphically, to find the boundary between points closer to sijk than si j k , we draw the line seg-
ment connecting sijk and si j k . The boundary is then the perpendicular bisector. The resulting
boundaries are shown in this figure:
X2
A110 A100
s110 s100
A010 A000
s010 s000
X1
A011 A001
s011 s001
A111 A101
s111 s101

In Problem 8.3.1, we found the MAP acceptance regions were
A0 = {x|x ≤ −a/2} A1 = {x| − a/2 < x ≤ a/2} A2 = {x|x > a/2} (1)
To calculate the probability of decoding error, we first calculate the conditional error probabilities
P [DE |Hi ] = P [X ∈ Ai |Hi ] (2)
Given Hi , recall that X = a(i − 1) + N . This implies

a
P [X ∈ A0 |H0 ] = P [−a + N > −a/2] = P [N > a/2] = Q (3)
2σN

a
P [X ∈ A1 |H1 ] = P [N < −a/2] + P [N > a/2] = 2Q (4)
2σN

a
P [X ∈ A2 |H2 ] = P [a + N < a/2] = P [N < −a/2] = Q (5)
2σN
295
Since the three hypotheses H0 , H1 , and H2 each have probability 1/3, the probability of error is

2
4 a
P [DE ] = P [X ∈ Ai |Hi ] P [Hi ] = Q (6)
3 2σN
i=0

Let Hi denote the hypothesis that symbol ai was transmitted. Since the four hypotheses are equally
likely, the ML tests will maximize the probability of a correct decision. Given Hi , N1 and N2 are
independent and thus X1 and X2 are independent. This implies
fX1 ,X2 |Hi (x1 , x2 ) = fX1 |Hi (x1 ) fX2 |Hi (x2 ) (1)
1 −(x1 −si1 )2 /2σ2 −(x2 −si2 )2 /2σ2
= e e (2)
2πσ 2
1 −[(x1 −si1 )2 +(x2 −si2 )2 ]/2σ2
= e (3)
2πσ 2
From Definition 8.2 the acceptance regions Ai for the ML multiple hypothesis test must satisfy
(x1 , x2 ) ∈ Ai if fX1 ,X2 |Hi (x1 , x2 ) ≥ fX1 ,X2 |Hj (x1 , x2 ) for all j. (4)
Equivalently, the ML acceptance regions are
(x1 , x2 ) ∈ Ai if (x1 − si1 )2 + (x2 − si2 )2 ≤ (x1 − sj1 )2 + (x2 − sj2 )2 for all j (5)
In terms of the vectors x and si , the acceptance regions are defined by the rule
x ∈ Ai if x − si 2 ≤ x − sj 2 (6)
Just as in the case of QPSK, the acceptance region Ai is the set of vectors x that are closest to si .

From the signal constellation depicted in Problem 8.3.5, each signal sij1 is below the x-axis while
each signal sij0 is above the x-axis. The event B3 of an error in the third bit occurs if we transmit
a signal sij1 but the receiver output x is above the x-axis or if we transmit a signal sij0 and the
receiver output is below the x-axis. By symmetry, we need only consider the case when we transmit
one of the four signals sij1 . In particular,
• Given H011 or H001 , X2 = −1 + N2
• Given H101 or H111 , X2 = −2 + N2
This implies
P [B3 |H011 ] = P [B3 |H001 ] = P [−1 + N2 > 0] = Q(1/σN ) (1)

P [B3 |H101 ] = P [B3 |H111 ] = P [−2 + N2 > 0] = Q(2/σN ) (2)
Assuming all four hypotheses are equally likely, the probability of an error decoding the third bit
is
P [B3 |H011 ] + P [B3 |H001 ] + P [B3 |H101 ] + P [B3 |H111 ]
P [B3 ] = (3)
4
Q(1/σN ) + Q(2/σN )
= (4)
2
296
(a) Hypothesis Hi is that X = si +N, where N is a Gaussian random vector independent of which
signal was transmitted. Thus, given Hi , X is a Gaussian (si , σ 2 I) random vector. Since X is
two-dimensional,
1 − 1 (x−si ) σ2 I−1 (x−si ) 1 − 12 x−si 2
fX|Hi (x) = 2
e 2 = e 2σ . (1)
2πσ 2πσ 2
Since the hypotheses Hi are equally likely, the MAP and ML rules are the same and achieve
the minimum probability of error. In this case, from the vector version of Theorem 8.8, the
MAP rule is
x ∈ Am if fX|Hm (x) ≥ fX|Hj (x) for all j. (2)
X2 Using the conditional PDFs fX|Hi (x), the MAP rule becomes
A2
s2
s1 A1 x ∈ Am if x − sm 2 ≤ x − sj 2 for all j. (3)
s0 A0
X1 In terms of geometry, the interpretation is that all vectors x
sM-1 closer to sm than to any other signal sj are assigned to Am . In
AM-1 this problem, the signal constellation (i.e., the set of vectors si )
sM-2
AM-2 is the set of vectors on the circle of radius E. The acceptance
regions are the “pie slices” around each signal vector.
(b) Consider the following sketch to determine d.

X2
Geometrically, the largest d such that x − si ≤ d defines the largest

d circle around si that can be inscribed into the pie slice Ai . By symme-
/2 X1 try, this is the same for every Ai , hence we examine A0 . Each pie √ slice
E1/2 s0
has angle θ = 2π/M . Since the length √ of each signal
√ vector is E,
the sketch shows that sin(θ/2) = d/ E. Thus d = E sin(π/M ).
(c) By symmetry, PERR is the same as the conditional probability of error 1−P [Ai |Hi ], no matter
which si is transmitted. Let B denote a circle of radius d at the origin and let Bi denote the
circle of radius d around si . Since B0 ⊂ A0 ,
P [A0 |H0 ] = P [X ∈ A0 |H0 ] ≥ P [X ∈ B0 |H0 ] = P [N ∈ B] . (4)
Since the components of N are iid Gaussian (0, σ 2 ) random variables,

'' ''
1 2 2 2
P [N ∈ B] = fN1 ,N2 (n1 , n2 ) dn1 dn2 = 2
e−(n1 +n2 )/2σ dn1 dn2 . (5)
B 2πσ B
297
By changing to polar coordinates,
' d ' 2π
1 2 2
P [N ∈ B] = 2
e−r /2σ r dθ dr (6)
2πσ 0 0
' d
1 2 2
= 2 re−r /2σ r dr (7)
σ 0
2
$
2 $d 2 2 2 2
= −e−r /2σ $ = 1 − e−d /2σ = 1 − e−E sin (π/M )/2σ (8)
0
Thus
2 (π/M )/2σ 2
PERR = 1 − P [A0 |H0 ] ≤ 1 − P [N ∈ B] = e−E sin . (9)
(a) In Problem 8.3.4, we found that in terms of the vectors x and si , the acceptance regions are
defined by the rule
x ∈ Ai if x − si 2 ≤ x − sj 2 for all j. (1)
Just as in the case of QPSK, the acceptance region Ai is the set of vectors x that are closest
to si . Graphically, these regions are easily found from the sketch of the signal constellation:
X2
s7 s3
s5 s2
s6 s4 s1 s0
X1
s9 s8 s1 2 s1 4
s1 0 s1 3
s11 s1 5
(b) For hypothesis A1 , we see that the acceptance region is
A1 = {(X1 , X2 )|0 < X1 ≤ 2, 0 < X2 ≤ 2} (2)
298
Given H1 , a correct decision is made if (X1 , X2 ) ∈ A1 . Given H1 , X1 = 1 + N1 and X2 =
1 + N2 . Thus,
P [C|H1 ] = P [(X1 , X2 ) ∈ A1 |H1 ] (3)

= P [0 < 1 + N1 ≤ 2, 0 < 1 + N2 ≤ 2] (4)
= (P [−1 < N1 ≤ 1])2 (5)
= (Φ(1/σN ) − Φ(−1/σN )) 2
(6)
= (2Φ(1/σN ) − 1) 2
(7)
(c) Surrounding each signal si is an acceptance region Ai that is no smaller than the acceptance
region A1 . That is,
P [C|Hi ] = P [(X1 , X2 ) ∈ Ai |H1 ] (8)

≥ P [−1 < N1 ≤ 1, −1 < N2 ≤ 1] (9)
= (P [−1 < N1 ≤ 1]) = P [C|H1 ] .
2
(10)
This implies

15
P [C] = P [C|Hi ] P [H1 ] (11)
i=0
15
15
≥ P [C|H1 ] P [Hi ] = P [C|H1 ] P [Hi ] = P [C|H1 ] (12)
i=0 i=0

Let pi = P [Hi ]. From Theorem 8.8, the MAP multiple hypothesis test is
(x1 , x2 ) ∈ Ai if pi fX1 ,X2 |Hi (x1 , x2 ) ≥ pj fX1 ,X2 |Hj (x1 , x2 ) for all j (1)
From Example 8.13, the conditional PDF of X1 , X2 given Hi is

1 −[(x1 −√E cos θi )2 +(x2 −√E sin θi )2 ]/2σ2
fX1 ,X2 |Hi (x1 , x2 ) = e (2)
2πσ 2
Using this conditional joint PDF, the MAP rule becomes
• (x1 , x2 ) ∈ Ai if for all j,
√ √
(x1 − E cos θi )2 + (x2 − E sin θi )2
−
√ 2σ 2 √
(x1 − E cos θj )2 + (x2 − E sin θj )2 pj
+ 2
≥ ln . (3)
2σ pi
Expanding the squares and using the identity cos2 θ + sin2 θ = 1 yields the simplified rule
• (x1 , x2 ) ∈ Ai if for all j,
σ2 pj
x1 [cos θi − cos θj ] + x2 [sin θi − sin θj ] ≥ √ ln (4)
E pi
299
Note that the MAP rules define linear constraints in x1 and x2 . Since θi = π/4 + iπ/2, we use the
following table to enumerate the constraints:
cos θi sin θi
√ √
i=0 1/ √2 1/√2
i=1 −1/√2 1/ √2 (5)
i=2 −1/√ 2 −1/√2
i=3 1/ 2 −1/ 2
To be explicit, to determine whether (x1 , x2 ) ∈ Ai , we need to check the MAP rule for each j = i.
Thus, each Ai is defined by three constraints. Using the above table, the acceptance regions are
• (x1 , x2 ) ∈ A0 if
σ2 p1 σ2 p3 σ2 p2
x1 ≥ √ ln x2 ≥ √ ln x1 + x2 ≥ √ ln (6)
2E p0 2E p0 2E p0
• (x1 , x2 ) ∈ A1 if
σ2 p1 σ2 p2 σ2 p3
x1 ≤ √ ln x2 ≥ √ ln − x1 + x2 ≥ √ ln (7)
2E p0 2E p1 2E p1
• (x1 , x2 ) ∈ A2 if
σ2 p2 σ2 p2 σ2 p2
x1 ≤ √ ln x2 ≤ √ ln x1 + x2 ≥ √ ln (8)
2E p3 2E p1 2E p0
• (x1 , x2 ) ∈ A3 if
σ2 p2 σ2 p3 σ2 p2
x1 ≥ √ ln x2 ≤ √ ln − x1 + x2 ≥ √ ln (9)
2E p3 2E p0 2E p3
Using the parameters
σ = 0.8 E=1 p0 = 1/2 p1 = p2 = p3 = 1/6 (10)
the acceptance regions for the MAP rule are
A0 = {(x1 , x2 )|x1 ≥ −0.497, x2 ≥ −0.497, x1 + x2 ≥ −0.497} (11)

A1 = {(x1 , x2 )|x1 ≤ −0.497, x2 ≥ 0, −x1 + x2 ≥ 0} (12)
A2 = {(x1 , x2 )|x1 ≤ 0, x2 ≤ 0, x1 + x2 ≥ −0.497} (13)
A3 = {(x1 , x2 )|x1 ≥ 0, x2 ≤ −0.497, −x1 + x2 ≥ 0} (14)
Here is a sketch of these acceptance regions:
300
X2
A1 s1 s0 A0
X1
A2 s2 s3 A3
Note that the boundary between A1 and A3 defined by −x1 + x2 ≥ 0 plays no role because of the
high value of p0 .
(a) First we note that

⎡√ ⎤⎡ ⎤ ⎡√ ⎤
p1 X1 p1 X 1
⎢ .. ⎥ ⎢ .. ⎥ ⎢ .. ⎥
P1/2 X = ⎣ . ⎦⎣ . ⎦ = ⎣ . ⎦. (1)
√ √
pk Xk pk X k
Since each Si is a column vector,

⎡√ ⎤
p1 X 1
⎢ ⎥ √ √
SP1/2 X = S1 · · · Sk ⎣ ... ⎦ = p1 X1 S1 + · · · + pk Xk Sk . (2)
√
pk X k

k √
Thus Y = SP1/2 X + N = i=1 pi Xi Si + N.

(b) Given the observation Y = y, a detector must decide which vector X = X1 · · · Xk
was (collectively) sent by the k transmitters. A hypothesis Hj must specify whether Xi = 1
or Xi = −1 for each i. That is, a hypothesis Hj corresponds to a vector xj ∈ Bk which
has ±1 components. Since there are 2k such vectors, there are 2k hypotheses which we can
enumerate as H1 , . . . , H2k . Since each Xi is independently and equally likely to be ±1, each
hypothesis has probability 2−k . In this case, the MAP and and ML rules are the same and
achieve minimum probability of error. The MAP/ML rule is
y ∈ Am if fY|Hm (y) ≥ fY|Hj (y) for all j. (3)
Under hypothesis Hj , Y = SP1/2 xj + N is a Gaussian (SP1/2 xj , σ 2 I) random vector. The

conditional PDF of Y is
1 1 2
− 21 (y−SP1/2 xj ) (σ 2 I)−1 (y−SP1/2 xj ) −y−SP1/2 xj /2σ 2
fY|Hj (y) = e = e . (4)
(2πσ 2 )n/2 (2πσ 2 )n/2
301
The MAP rule is
2 2
y ∈ Am if e−y−SP ≥ e−y−SP
1/2 x /2σ 2 1/2 x /2σ 2
m j
for all j, (5)
or equivalently,
; ; ; ;
; ; ; ;
y ∈ Am if ;y − SP1/2 xm ; ≤ ;y − SP1/2 xj ; for all j. (6)
; ;
That is, we choose the vector x∗ = xm that minimizes the distance ;y − SP1/2 xj ; among all
vectors xj ∈ Bk . Since this vector x∗ is a function of the observation y, this is described by
the math notation ; ;
; ;
x∗ (y) = arg min ;y − SP1/2 x; , (7)
x∈Bk
where arg minx g(x) returns the argument x that minimizes g(x).
; ;
(c) To implement this detector, we must evaluate ;y − SP1/2 x; for each x ∈ Bk . Since there 2k
vectors in Bk , we have to evaluate 2k hypotheses. Because the number of hypotheses grows
exponentially with the number of users k, the maximum likelihood detector is considered to
be computationally intractable for a large number of users k.

A short answer is that the decorrelator cannot be the same as the optimal maximum likelihood
(ML) detector. If they were the same, that means we have reduced the 2k comparisons of the
optimal detector to a linear transformation followed by k single bit comparisons.
However, as this is not a satisfactory answer, we will build a simple example with k = 2 users
and precessing gain n = 2 to show the difference between the ML detector
and the decorrelator. In
particular, suppose user 1 transmits with code vector S1 = 1 0 and user transmits with code

vector S2 = cos θ sin θ In addition, we assume that the users powers are p1 = p2 = 1. In this
case, P = I and
1 cos θ
S= . (1)
0 sin θ
For the ML detector, there are four hypotheses corresponding to each possible transmitted bit of
each user. Using Hi to denote the hypothesis that X = xi , we have

1 −1
X = x1 = X = x3 = (2)
1 1

1 −1
X = x2 = X = x4 = (3)
−1 −1
When X = xi , Y = yi + N where yi = Sxi . Thus under hypothesis Hi , Y = yi + N is a Gaussian

(yi , σ 2 I) random vector with PDF
1 −(y−yi ) (σ2 I)−1 (y−yi )/2 1 − y−yi 2 /2σ2
fY|Hi (y) = e = e . (4)
2πσ 2 2πσ 2
With the four hypotheses equally likely, the MAP and ML detectors are the same and minimize
the probability of error. From Theorem 8.8, this decision rule is
y ∈ Am if fY|Hm (y) ≥ fY|Hj (y) for all j. (5)
302
y ∈ Am if y − ym ≤ y − yj for all j. (6)
It is useful to show these acceptance sets graphically. In this plot, the area around yi is the
acceptance set Ai and the dashed lines are the boundaries between the acceptance sets.
Y2
A3 A1
y3 y1
1 + cos θ −1 + cos θ
y1 = y3 = (7)
Y1 sin θ sin θ

1 − cos θ −1 − cos θ
y2 = y4 = (8)
y4 y2 − sin θ − sin θ
A4 A2
The probability of a correct decision is

'
1
4
P [C] = fY|Hi (y) dy. (9)
4 Ai
i=1
(Even though the components of Y are conditionally independent given Hi , the four integrals
Ai fY|Hi (y) dy cannot be represented in a simple form. Moreoever, they cannot even be rep-
resented by the Φ(·) function. Note that the probability of a correct decision is the probability that
the bits X1 and X2 transmitted by both users are detected correctly.
The probability of a bit error is still somewhat more complex. For example if X1 = 1, then
hypotheses H1 and H3 are equally likely. The detector guesses X̂1 = 1 if Y ∈ A1 ∪ A3 . Given
X1 = 1, the conditional probability of a correct decision on this bit is
1 1
P X̂1 = 1|X1 = 1 = P [Y ∈ A1 ∪ A3 |H1 ] + P [Y ∈ A1 ∪ A3 |H3 ] (10)
2' 2 '
1 1
= fY|H1 (y) dy + f (y) dy (11)
2 A1 ∪A3 2 A1 ∪A3 Y|H3
By comparison, the decorrelator does something simpler. Since S is a square invertible matrix,

−1 −1 −1 −1 1 1 − cos θ
(S S) S = S (S ) S = S = (12)
sin θ 0 1
We see that the components of Ỹ = S−1 Y are

cos θ Y2
Ỹ1 = Y1 − Y2 , Ỹ2 = . (13)
sin θ sin θ
Assuming (as in earlier sketch) that 0 < θ < π/2, the decorrelator bit decisions are
Y2
A3 y3 y1 A1
cos θ
X̂1 = sgn (Ỹ1 ) = sgn Y1 − Y2 (14)
Y1 sin θ

Y2
X̂2 = sgn (Ỹ2 ) = sgn = sgn (Y2 ). (15)
A4 y4 y2 A2 sin θ
303
Because we chose a coordinate system such that S1 lies along the x-axis, the effect of the decorrela-
tor on the rule for bit X2 is particularly easy to understand. For bit X2 , we just check whether the
vector Y is in the upper half plane. Generally, the boundaries of the decorrelator decision regions
are determined by straight lines, they are easy to implement and probability of error is easy to
calculate. However, these regions are suboptimal in terms of probability of error.

Under hypothesis Hi , the conditional PMF of X is

(1 − pi )px−1
i /(1 − p20
i ) x = 1, 2, . . . , 20,
PX|Hi (x) = (1)
0 otherwise,
where p0 = 0.99 and p1 = 0.9. It follows that for x0 = 0, 1, . . . , 19 that
1 − pi
20
1 − pi x0
P [X > x0 |Hi ] = px−1 = pi + · · · + p19 (2)
1 − p20
i x=x0 +1
i
1 − pi20 i
px0 (1 − pi )
= i 1 + p i + · · · + p 19−x0
(3)
1 − p20
i
i
pxi 0 (1 − p20−x
i
0
) pxi 0 − p20i
= = (4)
1 − p20
i 1 − p 20
i
We note that the above formula is also correct for x0 = 20. Using this formula, the false alarm and
miss probabilities are
px0 0 − p20
PFA = P [X > x0 |H0 ] = 0
, (5)
1 − p20
0
1 − px1 0
PMISS = 1 − P [X > x0 |H1 ] = (6)
1 − p20
1
The Matlab program rocdisc(p0,p1) returns the false alarm and miss probabilities and also
plots the ROC. Here is the program and the output for rocdisc(0.9,0.99):
function [PFA,PMISS]=rocdisc(p0,p1); 1
x=0:20;
0.8
PFA= (p0.^x-p0^(20))/(1-p0^(20));
P MISS
PMISS= (1.0-(p1.^x))/(1-p1^(20)); 0.6

plot(PFA,PMISS,’k.’);
0.4
xlabel(’\itP_{\rm FA}’);
ylabel(’\itP_{\rm MISS}’); 0.2
0
0 0.5 1
P FA
From the receiver operating curve, we learn that we have a fairly lousy sensor. No matter how
we set the threshold x0 , either the false alarm probability or the miss probability (or both!) exceed
0.5.

From Example 8.7, the probability of error is

σ p v σ p v
PERR = pQ ln + + (1 − p)Φ ln − . (1)
2v 1 − p σ 2v 1 − p σ
304
It is straightforward to use Matlab to plot PERR as a function of p. The function bperr calculates
PERR for a vector p and a scalar signal to noise ratio snr corresponding to v/σ. A second program
bperrplot(snr) plots PERR as a function of p. Here are the programs
function perr=bperr(p,snr); function pe=bperrplot(snr);
%Problem 8.4.2 Solution p=0.02:0.02:0.98;
r=log(p./(1-p))/(2*snr); pe=bperr(p,snr);
perr=(p.*(qfunction(r+snr))) ... plot(p,pe);
+((1-p).*phi(r-snr)); xlabel(’\it p’);
ylabel(’\it P_{ERR}’);
Here are three outputs of bperrplot for the requested SNR values.
−24
x 10
0.8 0.2 8
0.6 0.15
6
PERR
PERR
PERR
0.4 0.1
4
0.2 0.05
0 0 2
0 0.5 1 0 0.5 1 0 0.5 1
p p p
bperrplot(0.1) bperrplot(0.1) bperrplot(0.1)
In all three cases, we see that PERR is maximum at p = 1/2. When p = 1/2, the optimal (minimum
probability of error) decision rule is able to exploit the one hypothesis having higher a priori
probability than the other.
This gives the wrong impression that one should consider building a communication system
with p = 1/2. To see this, consider the most extreme case in which the error probability goes to
zero as p → 0 or p → 1. However, in these extreme cases, no information is being communicated.
When p = 0 or p = 1, the detector can simply guess the transmitted bit. In fact, there is no need
to tansmit a bit; however, it becomes impossible to transmit any information.
Finally, we note that v/σ is an SNR voltage ratio. For communication systems, it is common to
measure SNR as a power ratio. In particular, v/σ = 10 corresponds to a SNR of 10 log1 0(v 2 /σ 2 ) =
20 dB.

With v = 1.5 and d = 0.5, it appeared in Example 8.14 that T = 0.5 was best among the values
tested. However, it also seemed likely the error probability Pe would decrease for larger values of
T . To test this possibility we use sqdistor with 100,000 transmitted bits by trying the following:
>> T=[0.4:0.1:1.0];Pe=sqdistor(1.5,0.5,100000,T);
>> [Pmin,Imin]=min(Pe);T(Imin)
ans =
0.80000000000000
Thus among {0.4, 0.5, · · · , 1.0}, it appears that T = 0.8 is best. Now we test values of T in the
neighborhood of 0.8:
>> T=[0.70:0.02:0.9];Pe=sqdistor(1.5,0.5,100000,T);
>>[Pmin,Imin]=min(Pe);T(Imin)
ans =
0.78000000000000
305
This suggests that T = 0.78 is best among these values. However, inspection of the vector Pe shows
that all values are quite close. If we repeat this experiment a few times, we obtain:
>> T=[0.70:0.02:0.9];Pe=sqdistor(1.5,0.5,100000,T);
ans =
0.78000000000000
>> T=[0.70:0.02:0.9];Pe=sqdistor(1.5,0.5,100000,T);
ans =
0.80000000000000
>> T=[0.70:0.02:0.9];Pe=sqdistor(1.5,0.5,100000,T);
ans =
0.76000000000000
>> T=[0.70:0.02:0.9];Pe=sqdistor(1.5,0.5,100000,T);
ans =
0.78000000000000
This suggests that the best value of T is in the neighborhood of 0.78. If someone were paying you
to find the best T , you would probably want to do more testing. The only useful lesson here is that
when you try to optimize parameters using simulation results, you should repeat your experiments
to get a sense of the variance of your results.

Since the a priori probabilities P [H0 ] and P [H1 ] are unknown, we use a Neyamn-Pearson formula-
tion to find the ROC. For a threshold γ, the decision rule is
fX|H0 (x)
x ∈ A0 if ≥ γ; x ∈ A1 otherwise. (1)
fX|H1 (x)
Using the given conditional PDFs, we obtain
2 )/16
x ∈ A0 if e−(8x−x ≥ γx/4; x ∈ A1 otherwise. (2)
Taking logarithms yields
x ∈ A0 if x2 − 8x ≥ 16 ln(γ/4) + 16 ln x; x ∈ A1 otherwise. (3)
With some more rearranging,
x ∈ A0 if (x − 4)2 ≥ 16 ln(γ/4) + 16 +16 ln x; x ∈ A1 otherwise. (4)

γ0
When we plot the functions f (x) = (x − 4)2 and g(x) = γ0 + 16 ln x, we see that there exist x1 and
x2 such that f (x1 ) = g(x1 ) and f (x2 ) = g(x2 ). In terms of x1 and x2 ,
A0 = [0, x1 ] ∪ [x2 , ∞), A1 = (x1 , x2 ). (5)
Using a Taylor series expansion of ln x around x = x0 = 4, we can show that
g(x) = γ0 + 16 ln x ≤ h(x) = γ0 + 16(ln 4 − 1) + 4x. (6)
Since h(x)
√ is linear, we can use the quadratic formula to solve f (x) = h(x), yielding a solution
x̄2 = 6 + 4 + 16 ln 4 + γ0 . One can show that x2 ≤ x̄2 . √
In the example shown below corresponding
to γ = 1 shown here, x1 = 1.95, x2 = 9.5 and x̄2 = 6 + 20 = 10.47.
306
60
40
20
0
f(x)
−20 g(x)
h(x)
−40
0 2 4 6 8 10 12
Given x1 and x2 , the false alarm and miss probabilities are

' 2
1 −x/2
PFA = P [A1 |H0 ] = e dx = e−x1 /2 − e−x2 /2 , (7)
x1 2
' x2
x −x2 /16 2 2
PMISS = 1 − P [A1 |H1 ] = 1 − e dx = 1 − e−x1 /16 + e−x2 /16 (8)
x1 8
To calculate the ROC, we need to find x1 and x2 . Rather than find them exactly, we calculate f (x)
and g(x) for discrete steps over the interval [0, 1 + x̄2 ] and find the discrete values closest to x1 and
x2 . However, for these approximations to x1 and x2 , we calculate the exact false alarm and miss
probabilities. As a result, the optimal detector using the exact x1 and x2 cannot be worse than the
ROC that we calculate.
In terms of Matlab, we divide the work into the functions gasroc(n) which generates the
ROC by calling [x1,x2]=gasrange(gamma) to calculate x1 and x2 for a given value of γ.
function [pfa,pmiss]=gasroc(n); function [x1,x2]=gasrange(gamma);
a=(400)^(1/(n-1)); g=16+16*log(gamma/4);
k=1:n; xmax=7+ ...
g=0.05*(a.^(k-1)); sqrt(max(0,4+(16*log(4))+g));
pfa=zeros(n,1); dx=xmax/500;
pmiss=zeros(n,1); x=dx:dx:4;
for k=1:n, y=(x-4).^2-g-16*log(x);
[x1,x2]=gasrange(g(k)); [ym,i]=min(abs(y));
pmiss(k)=1-(exp(-x1^2/16)... x1=x(i);
-exp(-x2^2/16)); x=4:dx:xmax;
pfa(k)=exp(-x1/2)-exp(-x2/2); y=(x-4).^2-g-16*log(x);
end [ym,i]=min(abs(y));
plot(pfa,pmiss); x2=x(i);
ylabel(’P_{\rm MISS}’);
xlabel(’P_{\rm FA}’);
The argment n of gasroc(n) generates the ROC for n values of γ, ranging from from 1/20 to
20 in multiplicative steps. Here is the resulting ROC:
1
0.8
0.6
P MISS
0.4
0.2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
P FA
307
After all of this work, we see that the sensor is not particularly good in the the ense that no matter
how we choose the thresholds, we cannot reduce both the miss and false alarm probabilities under
30 percent.

X2
A2
s2 In the solution to Problem 8.3.6, we found that the signal constellation
s1 A1 and acceptance regions shown in the adjacent figure. We could solve
s0 A0 this problem by a general simulation of an M -PSK system. This would
X1
include a random sequence of data sysmbols, mapping symbol i to
sM-1
AM-1 vector si , adding the noise vector N to produce the receiver output
sM-2 X = si + N.
AM-2
However, we are only asked to find the probability of symbol error, but not the probability that
symbol i is decoded as symbol j at the receiver. Because of the symmetry of the signal constellation
and the acceptance regions, the probability of symbol error is the same no matter what symbol is
transmitted.
N2
Thus it is simpler to assume that s0 is transmitted every time and
check that the noise vector N is in the pie slice around s0 . In fact by
/2 N1 translating s0 to the origin, we obtain the “pie slice” geometry shown
(-E1/2, 0) (0,0)
in the figure. Because the lines marking the boundaries of the pie slice
have slopes ± tan θ/2.
The pie slice region is given by the constraints

√ √
N2 ≤ tan(θ/2) N1 + E , N2 ≥ − tan(θ/2) N1 + E . (1)
We can rearrange these inequalities to express them in vector form as

− tan θ/2 1 N1 1 √
≤ E tan θ/2. (2)
− tan θ/2 −1 N2 1
Finally, since each Ni has variance σ 2 , we define the Gaussian (0, I) random vector Z = N/σ and
write our constraints as
− tan θ/2 1 Z1 1 √
≤ γ tan θ/2, (3)
− tan θ/2 −1 Z2 1
where γ = E/σ 2 is the signal to noise ratio of the system.
The Matlab “simulation” simply generates many pairs Z1 Z2 and checks what fraction
meets these constraints. the function mpsksim(M,snr,n) simulates the M -PSK system with SNR
snr for n bit transmissions. The script mpsktest graphs the symbol error probability for M =
8, 16, 32.
308
function Pe=mpsksim(M,snr,n); %mpsktest.m;
%Problem 8.4.5 Solution: snr=10.^((0:30)/10);
%Pe=mpsksim(M,snr,n) n=500000;
%n bit M-PSK simulation Pe8=mpsksim(8,snr,n);
t=tan(pi/M); Pe16=mpsksim(16,snr,n);
A =[-t 1; -t -1]; Pe32=mpsksim(32,snr,n);
Z=randn(2,n); loglog(snr,Pe8,snr,Pe16,snr,Pe32);
PC=zeros(length(snr)); legend(’M=8’,’M=16’,’M=32’,3);
for k=1:length(snr),
B=(A*Z)<=t*sqrt(snr(k));
PC(k)=sum(min(B))/n;
end
Pe=1-PC;

In mpsksim, each column of the matrix Z corresponds to a pair of noise variables Z1 Z2 .
The code B=(A*Z)<=t*sqrt(snr(k)) checks whether each pair of noise variables is in the pie slice
region. That is, B(1,j) and B(2,j) indicate if the ith pair meets the first and second constraints.
Since min(B) operates on each column of B, min(B) is a row vector indicating which pairs of noise
variables passed the test.
Here is the output of mpsktest:
0
10
M=8
M=16
M=32
−5
10
0 1 2 3
10 10 10 10
The curves for M = 8 and M = 16 end prematurely because for high SNR, the error rate is so low
that no errors are generated in 500,000 symbols. In this case, the measured Pe is zero and since
log 0 = −∞, the loglog function simply ignores the zero values.

When the transmitted bit vector is X = x, the received signal vector Y = SP1/2 x+N is a Gaussian
(SP1/2 x, σ 2 I) random vector with conditional PDF
1 2
e−y−SP x /2σ .
1/2 2
fY|X (y|x) = 2 k/2
(1)
(2πσ )
The transmitted data vector x belongs to the set Bk of all binary ±1 vectors of length k. In
principle, we can enumerate the vectors in Bk as x0 , x1 , . . . , x2k −1 . Moreover, each possible data
vector xm represents a hypothesis. Since there are 2k possible data vectors, there are 2k acceptance
sets Am . The set Am is the set of all vectors y such that the decision rule is to guess X̂ = xm ,
Our normal procedure is to write a decision rule as “y ∈ Am if . . . ” however this problem has so
many has so many hypotheses that it is more staightforward to refer to a hypothesis X = xm by
the function x̂(y) which returns the vector xm when y ∈ Am . In short, x̂(y) is our best guess as
to which vector x was transmitted when y is received.
309
Because each hypotheses has a priori probability 2−k , the probability of error is minimized by
the maximum likelihood (ML) rule
x̂(y) = arg max fY|X (y|x) . (2)

x∈Bk
Keep in mind that arg maxx g(x) returns the argument x that maximizes g(x). In any case, the
form of fY|X (y|x) implies that the ML rule should minimize the negative exponent of fY|X (y|x).
That is, the ML rule is
; ;
; ;
x̂(y) = arg min ;y − SP1/2 x; (3)
x∈Bk
= arg min (y − SP1/2 x) (y − SP1/2 x) (4)

x∈Bk
= arg min y y − 2y SP1/2 x + x P1/2 S SP1/2 x (5)

x∈Bk
Since the term y y is the same for every x, we can define the function
h(x) = −2y SP1/2 x + x P1/2 S SP1/2 x, (6)
In this case, the ML rule can be expressed as x̂(y) = arg minx∈Bk h(x). We use Matlab to evaluate
h(x) for each x ∈ Bk . Since for k = 10, Bk has 210 = 1024 vectors, it is desirable to make the
calculation as easy as possible. To this end, we define w = SP1/2 x and and we write, with some
abuse of notation, h(·) as a function of w:
h(w) = −2y w + w w (7)
Still, given y, we need to evaluate h(w) for each vector w. In Matlab, this will be convenient
because we can form the matrices X and W with columns consisting of all possible vectors x and
w. In Matlab, it is easy to calculate w w by operating on the matrix W without looping through
all columns w.
function X=allbinaryseqs(n) In terms of Matlab, we start by defining X=allbinaryseqs(n)
%See Problem 8.4.6 which returns an n×2n matrix X, corresponding to X, such that
%X: n by 2^n matrix of all the columns of X enumerate all possible binary ±1 sequences of
%length n binary vectors length n. How allbinaryseqs works will be clear by generating
%Thanks to Jasvinder Singh
the matrices A and P and reading the help for bitget.
A=repmat([0:2^n-1],[n,1]);
P=repmat([1:n]’,[1,2^n]);
X = bitget(A,P);
X=(2*X)-1;
function S=randomsignals(n,k); Next is a short program that generates k random sig-

%S is an n by k matrix, columns are nals, each of length n. Each random signal is just a
%random unit length signal vectors binary ±1 sequence normalized to have length 1.
S=(rand(n,k)>0.5);
S=((2*S)-1.0)/sqrt(n);
Next, for a set of signal vectors (spreading sequences in CDMA parlance) given by the n × k
matrix S, err=cdmasim(S,P,m) simulates the transmission of a frame of m symbols through a k
user CDMA system with additive Gaussian noise. A “symbol,” is just a vector x corresponding to
the k transmitted bits of the k users.
310
In addition, the function Pe=rcdma(n,k,snr,s,m) runs cdmasim for the pairs of values of users
k and SNR snr. Here is the pair of functions:
function err=cdmasim(S,P,m); function Pe=rcdma(n,k,snr,s,m);
%err=cdmasim(P,S,m); %Pe=rcdma(n,k,snr,s,m);
%S= n x k matrix of signals %R-CDMA simulation:
%P= diag matrix of SNRs (power % proc gain=n, users=k
% normalized by noise variance) % rand signal set/frame
%See Problem 8.4.6 % s frames, m symbols/frame
k=size(S,2); %number of users %See Problem 8.4.6 Solution
n=size(S,1); %processing gain [K,SNR]=ndgrid(k,snr);
X=allbinaryseqs(k);%all data Pe=zeros(size(SNR));
Phalf=sqrt(P); for j=1:prod(size(SNR)),
W=S*Phalf*X; p=SNR(j);k=K(j);
WW=sum(W.*W); e=0;
err=0; for i=1:s,
for j=1:m, S=randomsignals(n,k);
s=duniformrv(1,2^k,1); e=e+cdmasim(S,p*eye(k),m);
y=S*Phalf*X(:,s)+randn(n,1); end
[hmin,imin]=min(-2*y’*W+WW); Pe(j)=e/(s*m*k);
err=err+sum(X(:,s)~=X(:,imin)); % disp([p k e Pe(j)]);
end end
In cdmasim, the kth diagonal element of P is the “power” pk of user k. Technically, we assume
that the additive Gaussian noise variable have variance 1, and thus pk is actually the signal to
noise ratio of user k. In addition, WW is a length 2k row vector, with elements w w for each possible
w. For each of the m random data symbols, represented by x (or X(:,s) in Matlab), cdmasim
calculates a received signal y (y). Finally, hmin is the minimum h(w) and imin is the index of
the column of W that minimizes h(w). Thus imin is also the index of the minimizing column of
X. Finally, cdmasim compares x̂(y) and the transmitted vector x bit by bit and counts the total
number of bit errors.
The function rcdma repeats cdmasim for s frames, with a random signal set for each frame.
Dividing the total number of bit errors over s frames by the total number of transmitted bits, we
find the bit error rate Pe . For an SNR of 4 dB and processing gain 16, the requested tests are
generated with the commands
>> n=16;
>> k=[2 4 8 16];
>> Pe=rcdma(n,k,snr,100,1000);
>>Pe
Pe =
0.0252 0.0272 0.0385 0.0788
>>
To answer part (b), the code for the matched filter (MF) detector is much simpler because
there is no need to test 2k hypotheses for every transmitted symbol. Just as for the case of the
ML detector, we define a function err=mfcdmasim(S,P,m) that simulates the MF detector for m
symbols for a given set of signal vectors S. In mfcdmasim, there is no need for looping. The mth
transmitted symbol is represented by the mth column of X and the corresponding received signal
is given by the mth column of Y. The matched filter processing can be applied to all m columns
at once. A second function Pe=mfrcdma(n,k,snr,s,m) cycles through all combinations of users k
and SNR snr and calculates the bit error rate for each pair of values. Here are the functions:
311
function err=mfcdmasim(S,P,m); function Pe=mfrcdma(n,k,snr,s,m);
%err=mfcdmasim(P,S,m); %Pe=rcdma(n,k,snr,s,m);
%S= n x k matrix of signals %R-CDMA, MF detection
%P= diag matrix of SNRs % proc gain=n, users=k
% SNR=power/var(noise) % rand signal set/frame
%See Problem 8.4.6b % s frames, m symbols/frame
k=size(S,2); %no. of users %See Problem 8.4.6 Solution
n=size(S,1); %proc. gain [K,SNR]=ndgrid(k,snr);
Phalf=sqrt(P); Pe=zeros(size(SNR));
X=randombinaryseqs(k,m); for j=1:prod(size(SNR)),
Y=S*Phalf*X+randn(n,m); p=SNR(j);kt=K(j);
XR=sign(S’*Y); e=0;
err=sum(sum(XR ~= X)); for i=1:s,
S=randomsignals(n,kt);
e=e+mfcdmasim(S,p*eye(kt),m);
end
Pe(j)=e/(s*m*kt);
disp([snr kt e]);
end
Here is a run of mfrcdma.
>> pemf=mfrcdma(16,k,4,1000,1000);
4 2 73936
4 4 264234
4 8 908558
4 16 2871356
>> pemf’
ans =
0.0370 0.0661 0.1136 0.1795
>>
The following plot compares the maximum likelihood (ML) and matched filter (MF) detectors.
0.2
ML
0.15 MF
0.1
0.05
0
0 2 4 6 8 10 12 14 16
As the ML detector offers the minimum probability of error, it should not surprising that it has a
lower bit error rate. Although the MF detector is worse, the reduction in detctor complexity makes
it attractive. In fact, in practical CDMA-based cellular phones, the processing gain ranges from
roughly 128 to 512. In such case, the complexity of the ML detector is prohibitive and thus only
matched filter detectors are used.

For the CDMA system of Problem 8.3.10, the received signal resulting from the transmissions of k
users was given by
Y = SP1/2 X + N (1)
312
√ √
where S is an n × k matrix with ith column Si and P1/2 = diag[ p1 , . . . , pk ] is a k × k diagonal
matrix of received powers, and N is a Gaussian (0, σ 2 I) Gaussian noise vector.
(a) When S has linearly independent columns, S S is invertible. In this case, the decorrelating
detector applies a transformation to Y to generate
Ỹ = (S S)−1 S Y = P1/2 X + Ñ, (2)
where Ñ = (S S)−1 S N is still a Gaussian noise vector with expected value E[Ñ] = 0.
Decorrelation separates the signals in that the ith component of Ỹ is
√
Ỹi = pi Xi + Ñi . (3)
This is the same as a single user-receiver output of the binary communication system of
Example 8.6. The single-user decision rule X̂i = sgn (Ỹi ) for the transmitted bit Xi has
probability of error
!) "
√ pi
Pe,i = P Ỹi > 0|Xi = −1 = P − pi + Ñi > 0 = Q . (4)
Var[Ñi ]
However, since Ñ = AN where A = (S S)−1 S , Theorem 5.16 tells us that Ñ has covariance
matrix CÑ = ACN A . We note that the general property that (B−1 ) = (B )−1 implies that
A = S((S S) )−1 = S(S S)−1 . These facts imply
CÑ == (S S)−1 S (σ 2 I)S(S S)−1 = σ 2 (S S)−1 . (5)
Note that S S is called the correlation matrix since its i, jth entry is Si Sj is the correlation
between the signal of user i and that of user j. Thus Var[Ñi ] = σ 2 (S S)−1
ii and the probability
of bit error for user i is for user i is
!) " !) "
pi pi
Pe,i = Q =Q . (6)
Var[Ñi ] (S S)−1
ii
To find the probability of error for a randomly chosen but, we average over the bits of all
users and find that !) "
1 1
k k
pi
Pe = Pe,i = Q . (7)
k
i=1
k (S S)−1
i=1 ii
(b) When S S is not invertible, the detector flips a coin to decide each bit. In this case, Pe,i = 1/2
and thus Pe = 1/2.
(c) When S is chosen randomly, we need to average over all possible matrices S to find the average
probability of bit error. However, there are 2kn possible matrices S and averaging over all
of them is too much work. Instead, we randomly generate m matrices S and estimate the
average Pe by averaging over these m matrices.
A function berdecorr uses this method to evaluate the decorrelator BER. The code has a
lot of lines because it evaluates the BER using m signal sets for each combination of users
k and SNRs snr. However, because the program generates signal sets and calculates the
BER asssociated with each, there is no need for the simulated transmission of bits. Thus
the program runs quickly. Since there are only 2n distinct columns for matrix S, it is quite
313
possible to generate signal sets that are not linearly independent. In this case, berdecorr
assumes the “flip a coin” rule is used. Just to see whether this rule dominates the error
probability, we also display counts of how often S is rank deficient.
Here is the (somewhat tedious) code:
function Pe=berdecorr(n,k,snr,m);
%Problem 8.4.7 Solution: R-CDMA with decorrelation
%proc gain=n, users=k, average Pe for m signal sets
count=zeros(1,length(k)); %counts rank<k signal sets
Pe=zeros(length(k),length(snr)); snr=snr(:)’;
for mm=1:m,
for i=1:length(k),
S=randomsignals(n,k(i)); R=S’*S;
if (rank(R)<k(i))
count(i)=count(i)+1;
Pe(i,:)=Pe(i,:)+0.5*ones(1,length(snr));
else
G=diag(inv(R));
Pe(i,:)=Pe(i,:)+sum(qfunction(sqrt((1./G)*snr)))/k(i);
end
end
end
disp(’Rank deficiency count:’);disp(k);disp(count);
Pe=Pe/m;
Running berdecorr with processing gains n = 16 and n = 32 yields the following output:
>> k=[1 2 4 8 16 32];

>> pe16=berdecorr(16,k,4,10000);
Rank deficiency count:
1 2 4 8 16 32
0 2 2 12 454 10000
>> pe16’
ans =
0.0228 0.0273 0.0383 0.0755 0.3515 0.5000
>> pe32=berdecorr(32,k,4,10000);
Rank deficiency count:
1 2 4 8 16 32
0 0 0 0 0 0
>> pe32’
ans =
0.0228 0.0246 0.0290 0.0400 0.0771 0.3904
>>
As you might expect, the BER increases as the number of users increases. This occurs
because the decorrelator must suppress a large set of interferers. Also, in generating 10,000
signal matrices S for each value of k we see that rank deficiency is fairly uncommon, however
it occasionally occurs for processing gain n = 16, even if k = 4 or k = 8. Finally, here is a
plot of these same BER statistics for n = 16 and k ∈ {2, 4, 8, 16}. Just for comparison, on the
same graph is the BER for the matched filter detector and the maximum likelihood detector
found in Problem 8.4.6.
314
0.4
ML
0.3 Decorr
MF
Pe
0.2
0.1
0
0 2 4 6 8 10 12 14 16
k
We see from the graph that the decorrelator is better than the matched filter for a small
number of users. However, when the number of users k is large (relative to the processing
gain n), the decorrelator suffers because it must suppress all interfering users. Finally, we
note that these conclusions are specific to this scenario when all users have equal SNR. When
some users have very high SNR, the decorrelator is good for the low-SNR user because it
zeros out the interference from the high-SNR user.

Each transmitted symbol sb1 b2 b3 corresponds to the transmission of three bits given by the vector

b = b1 b2 b3 . Note that sb1 b2 b3 is a two dimensional vector with components s(1) (2)
b1 b2 b3 sb1 b2 b3
.
The key to this problem is the mapping from bits to symbol components and then back to bits.
From the signal constellation shown with Problem 8.3.2, we make the following observations:
• sb1 b2 b3 is in the right half plane if b2 = 0; otherwise it is in the left half plane.
• sb1 b2 b3 is in the upper half plane if b3 = 0; otherwise it is in the lower half plane.
• There is an inner ring and an outer ring of signals. sb1 b2 b3 is in the inner ring if b1 = 0;
otherwise it is in the outer ring.

Given a bit vector b, we use these facts by first using b2 and b3 to map b = b1 b2 b3 to an
inner ring signal vector
3 4
s ∈ 1 1 , −1 1 , −1 −1 , 1 −1 . (1)
In the next step we scale s by (1 + b1 ). If b1 = 1, then s is stretched to the outer ring. Finally, we
add a Gaussian noise vector N to generate the received signal X = sb1 b2 b3 + N.
X2 In the solution to Problem 8.3.2, we found that the accep-
tance set for the hypothesis Hb1 b2 b3 that sb1 b2 b3 is transmitted
A110 A100 is the set of signal space points closest to sb1 b2 b3 . Graphically,
s110 s100 these acceptance sets are given in the adjacent figure. These
A010 A000 acceptance sets correspond an inverse mapping of the re-

s010 s000 ceived signal vector X to a bit vector guess b̂ = b̂1 b̂2 b̂3
X1 using the following rules:
A011 A001
s001 • b̂2 = 1 if X1 < 0; otherwise b̂2 = 0.
s011
A111 A101 • b̂3 = 1 if X2 < 0; otherwise b̂3 = 0.
s111 s101
√
• If |X1 | + |X2 | > 3 2/2, then b̂1 = 1; otherwise b̂1 = 0.
315
We implement these steps with the function [Pe,ber]=myqam(sigma,m) which simulates the
transmission of m symbols for each value of the vector sigma. Each column of B corresponds to a
bit vector b. Similarly, each column of S and X corresponds to a transmitted signal s and received
signal X. We calculate both the symbol decision errors that are made as well as the bit decision
errors. Finally, a script myqamplot.m plots the symbol error rate Pe and bit error rate ber as a
function of sigma. Here are the programs:
function [Pe,ber]=myqam(sigma,m); %myqamplot.m
Pe=zeros(size(sigma)); ber=Pe; sig=10.^(0.2*(-8:0));
B=reshape(bernoullirv(0.5,3*m),3,m); [Pe,ber]=myqam(sig,1e6);
%S(1,:)=1-2*B(2,:); loglog(sig,Pe,’-d’, ...
%S(2,:)=1-2*B(3,:); sig,ber,’-s’);
S=1-2*B([2; 3],:); legend(’SER’,’BER’,4);
S=([1;1]*(1+B(1,:))).*S;
N=randn(2,m);
for i=1:length(sigma),
X=S+sigma(i)*N;
BR=zeros(size(B));
BR([2;3],:)=(X<0);
BR(1,:)=sum(abs(X))>(3/sqrt(2));
E=(BR~=B);
Pe(i)=sum(max(E))/m;
ber(i)=sum(sum(E))/(3*m);
end
Note that we generate the bits and transmitted signals, and normalized noise only once. However
for each value of sigma, we rescale the additive noise, recalculate the received signal and receiver
bit decisions. The output of myqamplot is shown in this figure:
0
10
SER
BER
−5
10
−2 −1 0
10 10 10
Careful reading of the figure will show that the ratio of the symbol error rate to the bit error rate is
always very close to 3. This occurs because in the acceptance set for b1b2 b3 , the adjacent acceptance
sets correspond to a one bit difference. Since the usual type of symbol error occurs when the vector
X is in the adjacent set, a symbol error typically results in one bit being in error but two bits being
received correctly. Thus the bit error rate is roughly one third the symbol error rate.
(a) For the M -PSK communication system with additive Gaussian noise, Aj denoted the hypoth-
esis that signal sj was transmitted. The solution to Problem 8.3.6 derived the MAP decision
rule
316
X2
A2
s2
x ∈ Am if x − sm 2 ≤ x − sj 2 for all j. (1)
s1 A1
s0 A0 In terms of geometry, the interpretation is that all vectors x
X1
closer to sm than to any other signal sj are assigned to Am . In
sM-1
AM-1 this problem, the signal constellation (i.e., the set of vectors si )
sM-2 is the set of vectors on the circle of radius E. The acceptance
AM-2
regions are the “pie slices” around each signal vector.
We observe that
x − sj 2 = (x − sj ) (x − sj ) = x x − 2x sj + sj s . (2)
Since all the signals are on the same circle, sj sj is the same for all j. Also, x x is the same
for all j. Thus
min x − sj 2 = min −x sj = max x sj . (3)
j j j
Since x sj= x sj cos φ where φ is the angle between x and sj . Thus maximizing x sj is
equivalent to minimizing the angle between x and sj .
(b) In Problem 8.4.5, we estimated the probability of symbol error without building a complete
simulation of the M -PSK system. In this problem, we need to build a more complete simula-
tion to determine the probabilities Pij . By symmetry, it is sufficient to transmit s0 repeatedly
and count how often the receiver guesses sj . This is done by the functionp=mpskerr(M,snr,n).
function p=mpskerr(M,snr,n); Note that column i of S is the signal si−1 . The kth
%Problem 8.4.5 Solution: column of X corresponds to Xk = s0 + Nk , the re-
%Pe=mpsksim(M,snr,n) ceived signal for the kth transmission. Thus y(k)
%n bit M-PSK simulation corresponds to maxj Xk sj and e(k) reports the re-
t=(2*pi/M)*(0:(M-1));
ceiver decision for the kth transmission. The vector
S=sqrt(snr)*[cos(t);sin(t)];
X=repmat(S(:,1),1,n)+randn(2,n); p calculates the relative frequency of each receiver
[y,e]=max(S’*X); decision.
p=countequal(e-1,(0:(M-1)))’/n;

The next step is to translate the vector P00 P01 · · · P0,M −1 (corresponding to p in
Matlab) into an entire matrix P with elements Pij . The symmetry of the phase rotiation
dictates that each row of P should be a one element cyclic rotation of the previous row. More-
over, by symmetry we observe that P01 = P0,M −1 , P02 = P0,M −2 and so on. However, because
p is derived from a simulation experiment, it will exhibit this symmetry only approximately.
function P=mpskmatrix(p); Our ad hoc (and largely unjustified) solution is to take

M=length(p); the average of estimates of probabilities that symmetry
r=[0.5 zeros(1,M-2)]; says should be identical. (Why this is might be a good
A=toeplitz(r)+... thing to do would make an interesting exam problem.) In
hankel(fliplr(r));
mpskmatrix(p), the matrix A implements the averaging.
A=[zeros(1,M-1);A];
A=[[1; zeros(M-1,1)] A]; The code will become clear by examining the matrices A
P=toeplitz(A*(p(:))); and the output P.
317
(c) The next step is to determine the effect of the mapping of bits to transmission vectors sj .
The matrix D with i, jth element dij that indicates the number of bit positions in which the
bit string assigned to si differs from the bit string assigned to sj . In this case, the integers
provide a compact representation of this mapping. For example the binary mapping is
s0 s1 s2 s3 s4 s5 s6 s7
000 001 010 011 100 101 110 111
0 1 2 3 4 5 6 7
The Gray mapping is
s0 s1 s2 s3 s4 s5 s6 s7
000 001 011 010 110 111 101 100
0 1 3 2 6 7 5 4

Thus the binary mapping can be represented by a vector c1 = 0 1 · · · 7 while the Gray

mapping is described by c2 = 0 1 3 2 6 7 5 4 .
function D=mpskdist(c); The function D=mpskdist(c) translates the mapping

L=length(c);m=log2(L); vector c into the matrix D with entries dij . The
[C1,C2]=ndgrid(c,c); method is to generate grids C1 and C2 for the pairs
B1=dec2bin(C1,m); of integers, convert each integer into a length log2 M
B2=dec2bin(C2,m);
binary string, and then to count the number of bit
D=reshape(sum((B1~=B2),2),L,L);
positions in which each pair differs.
Given matrices P and D, the rest is easy. We treat BER as as a finite random variable that
takes on value dij with probability Pij . the expected value of this finite random variable is
the expected number of bit errors. Note that the BER is a “rate” in that
1
BER = Pij dij . (4)
M
i j
is the expected number of bit errors per transmitted symbol.
function Pb=mpskmap(c,snr,n); Given the integer mapping vector c, we estimate the

M=length(c); BER of the a mapping using just one more function
D=mpskdist(c); Pb=mpskmap(c,snr,n). First we calculate the matrix
Pb=zeros(size(snr)); D with elements dij. Next, for each value of the vector
for i=1:length(snr),
snr, we use n transmissions to estimate the probabili-
p=mpskerr(M,snr(i),n);
P=mpskmatrix(p); ties Pij . Last, we calculate the expected number of bit
Pb(i)=finiteexp(D,P)/M; errors per transmission.
end
(d) We evaluate the binary mapping with the following commands:
>> c1=0:7;
>>snr=[4 8 16 32 64];
>>Pb=mpskmap(c1,snr,1000000);
>> Pb
Pb =
0.7640 0.4878 0.2198 0.0529 0.0038
318
(e) Here is the performance of the Gray mapping:
>> c2=[0 1 3 2 6 7 5 4];

>>snr=[4 8 16 32 64];
>>Pg=mpskmap(c2,snr,1000000);
>> Pg
Pg =
0.4943 0.2855 0.1262 0.0306 0.0023
Experimentally, we observe that the BER of the binary mapping is higher than the BER of
the Gray mapping by a factor in the neighborhood of 1.5 to 1.7
In fact, this approximate ratio can be derived by a quick and dirty analysis. For high SNR,
suppose that that si is decoded as si+1 or si−1 with probability q = Pi,i+1 = Pi,i−1 and all
other types of errors are negligible. In this case, the BER formula based on this approximation
corresponds to summing the matrix D for the first off-diagonals and the corner elements. Here
are the calculations:
>> D=mpskdist(c1);
>> sum(diag(D,1))+sum(diag(D,-1))+D(1,8)+D(8,1)
ans =
28
>> DG=mpskdist(c2);
>> sum(diag(DG,1))+sum(diag(DG,-1))+DG(1,8)+DG(8,1)
ans =
16
Thus in high SNR, we would expect
BER(binary) ≈ 28q/M, BER(Gray) ≈ 16q/M. (5)
The ratio of BERs is 28/16 = 1.75. Experimentally, we found at high SNR that the ratio of
BERs was 0.0038/0.0023 = 1.65, which seems to be in the right ballpark.

As this problem is a continuation of Problem 8.4.9, this solution is also a continuation. In this
problem, we want to determine the error probability for each bit k in a mapping of bits to the
M -PSK signal constellation. The bit error rate associated with bit k is
1
BER(k) = Pij dij (k) (1)
M
i j
where dij (k) indicates whether the bit strings mapped to si and sj differ in bit position k.
As in Problem 8.4.9, we describe the mapping by the vector of integers d. For example the
binary mapping is
s0 s1 s2 s3 s4 s5 s6 s7
000 001 010 011 100 101 110 111
0 1 2 3 4 5 6 7
The Gray mapping is
319
s0 s1 s2 s3 s4 s5 s6 s7
000 001 011 010 110 111 101 100
0 1 3 2 6 7 5 4

Thus the binary mapping can be represented by a vector c1 = 0 1 · · · 7 while the Gray

mapping is described by c2 = 0 1 3 2 6 7 5 4 .
function D=mpskdbit(c,k); The function D=mpskdbit(c,k) translates the mapping
%See Problem 8.4.10: For mapping vector c into the matrix D with entries dij that indicates
%c, calculate BER of bit k whether bit k is in error when transmitted symbol si is
L=length(c);m=log2(L); decoded by the receiver as sj . The method is to generate
[C1,C2]=ndgrid(c,c);
grids C1 and C2 for the pairs of integers, identify bit k in
B1=bitget(C1,k);
B2=bitget(C2,k); each integer, and then check if the integers differ in bit k.
D=(B1~=B2);
Thus, there is a matrix D associated with each bit position and we calculate the expected number
of bit errors associated with each bit position. For each bit, the rest of the solution is the same as
in Problem 8.4.9. We use the commands p=mpskerr(M,snr,n) and P=mpskmatrix(p) to calculate
the matrix P which holds an estimate of each probability Pij . Finally, using matrices P and D, we
treat BER(k) as a finite random variable that takes on value dij with probability Pij . the expected
value of this finite random variable is the expected number of bit errors.
function Pb=mpskbitmap(c,snr,n); Given the integer mapping vector c, we estimate the
%Problem 8.4.10: Calculate prob. of BER of the a mapping using just one more function
%bit error for each bit position for Pb=mpskmap(c,snr,n). First we calculate the matrix
%an MPSK bit to symbol mapping c D with elements dij. Next, for a given value of snr,
M=length(c);m=log2(M);
we use n transmissions to estimate the probabilities
p=mpskerr(M,snr,n);
P=mpskmatrix(p); Pij . Last, we calculate the expected number of bit k
Pb=zeros(1,m); errors per transmission.
for k=1:m,
D=mpskdbit(c,k);
Pb(k)=finiteexp(D,P)/M;
end
For an SNR of 10dB, we evaluate the two mappings with the following commands:
>> c1=0:7;
>> mpskbitmap(c1,10,100000)
ans =
0.2247 0.1149 0.0577
>> c2=[0 1 3 2 6 7 5 4];

>> mpskbitmap(c2,10,100000)
ans =
0.1140 0.0572 0.0572
We see that in the binary mapping, the 0.22 error rate of bit 1 is roughly double that of bit 2,
which is roughly double that of bit 3. For the Gray mapping, the error rate of bit 1 is cut in half
relative to the binary mapping. However, the bit error rates at each position a re still not identical
since the error rate of bit 1 is still double that for bit 2 or bit 3. One might surmise that careful
study of the matrix D might lead one to prove for the Gray map that the error rate for bit 1 is
exactly double that for bits 2 and 3 . . . but that would be some other homework problem.
320

First we note that the event T > t0 has probability
' ∞
P [T > t0 ] = λe−λt dt = e−λt0 . (1)
t0
Given T > t0 , the conditional PDF of T is

λe−λ(t−t0 ) t > t0 ,
fT (t)
P [T >t 0 ] t > t0
fT |T >t0 (t) = = (2)
0 otherwise 0 otherwise.
Given T > t0 , the minimum mean square error estimate of T is

' ∞ ' ∞
T̂ = E [T |T > t0 ] = tfT |T >t0 (t) dt = λte−λ(t−t0 ) dt. (3)
−∞ t0
With the substitution t = t − t0 , we obtain

' ∞

T̂ = λ(t0 + t )e−λt dt (4)
0
' ∞ ' ∞
−λt
= t0 λe dt + t λe−λt dt = t0 + E [T ] (5)
0 0
1 E[T ]
(a) For 0 ≤ x ≤ 1,
' ∞ ' 1
fX (x) = fX,Y (x, y) dy = 6(y − x) dy (1)
−∞ x
$y=1
= 3y 2 − 6xy $y=x (2)
= 3(1 − 2x + x ) = 3(1 − x)
2 2
(3)
The complete expression for the marginal PDF of X is

3(1 − x)2 0 ≤ x ≤ 1,
fX (x) = (4)
0 otherwise.
(b) The blind estimate of X is

' $
1
3 2 3 4 $$1 1
x̂B = E [X] = 3x(1 − x) dx = x − 2x + x $ = .
2 3
(5)
0 2 4 0 4
321
(c) First we calculate
' '
0.5 0.5 $0.5 7
P [X < 0.5] = fX (x) dx = 3(1 − x)2 dx = −(1 − x)3 $0 = . (6)
0 0 8
The conditional PDF of X given X < 0.5 is

24
fX (x)
7 (1 − x)
0 ≤ x < 0.5 2 0 ≤ x < 0.5,
fX|X<0.5 (x) = P [X<0.5] = (7)
0 otherwise 0 otherwise.
The minimum mean square estimate of X given X < 0.5 is

' ∞ ' 0.5
24x
E [X|X < 0.5] = xfX|X<0.5 (x) dx = (1 − x)2 dx (8)
−∞ 0 7
$0.5
12x2 16x3 6x4 $$ 11
= − + $ = . (9)
7 7 7 0 56
(d) For y < 0 or y > 1, fY (y) = 0. For 0 ≤ y ≤ 1,

' y
$y
fY (y) = 6(y − x) dx = 6xy − 3x2 $0 = 3y 2 . (10)
0

2
3y 0 ≤ y ≤ 1,
fY (y) = (11)
0 otherwise.
(e) The blind estimate of Y is

' ∞ ' 1
3
ŷB = E [Y ] = yfY (y) dy = 3y 3 dy = . (12)
−∞ 0 4
(f) First we calculate

' ∞ ' 1
7
P [Y > 0.5] = fY (y) dy = 3y 2 dy = . (13)
0.5 0.5 8
The conditional PDF of Y given Y > 0.5 is

fY (y)
P [Y >0.5] y > 0.5, 24y 2 /7 y > 0.5,
fY |Y >0.5 (y) = = (14)
The minimum mean square estimate of Y given Y > 0.5 is

' ' $1
∞ 1
24y 3 6y 4 $$ 45
E [Y |Y > 0.5] = yfY |Y >0.5 (y) dy = dy = $ = . (15)
−∞ 0.5 7 7 0.5 56
322
(a) For 0 ≤ x ≤ 1,
' ∞ ' 1
fX (x) = fX,Y (x, y) dy = 2 dy = 2(1 − x). (1)
−∞ x
The complete expression of the PDF of X is

2(1 − x) 0 ≤ x ≤ 1,
fX (x) = (2)
0 otherwise.
(b) The blind estimate of X is

' $1
1
2x3 $$ 1
X̂B = E [X] = 2x(1 − x) dx = x −
2
$ = . (3)
0 3 0 3
(c) First we calculate

' '
1 1 $1 1
P [X > 1/2] = fX (x) dx = 2(1 − x) dx = (2x − x2 )$1/2 = . (4)
1/2 1/2 4
Now we calculate the conditional PDF of X given X > 1/2.

fX (x)
P [X>1/2] x > 1/2, 8(1 − x) 1/2 < x ≤ 1,
fX|X>1/2 (x) = = (5)
The minimum mean square error estimate of X given X > 1/2 is

' ∞
E [X|X > 1/2] = xfX|X>1/2 (x) dx (6)
−∞
' 1 $1
8x3 $$ 2
= 8x(1 − x) dx = 4x − 2
$ = . (7)
1/2 3 1/2 3
(d) For 0 ≤ y ≤ 1, the marginal PDF of Y is

' ∞ ' y
fY (y) = fX,Y (x, y) dx = 2 dx = 2y. (8)
−∞ 0

2y 0 ≤ y ≤ 1,
fY (y) = (9)
0 otherwise.
(e) The blind estimate of Y is

' 1
2
ŷB = E [Y ] = 2y 2 dy = . (10)
0 3
323
(f) We already know that P [X > 1/2] = 1/4. However, this problem differs from the other
problems in this section because we will estimate Y based on the observation of X. In this
case, we need to calculate the conditional joint PDF

fX,Y (x,y)
P [X>1/2] x > 1/2, 8 1/2 < x ≤ y ≤ 1,
fX,Y |X>1/2 (=) = (11)
The MMSE estimate of Y given X > 1/2 is

' ∞' ∞
E [Y |X > 1/2] = yfX,Y |X>1/2 (x, y) dx dy (12)
−∞ −∞
' !' "
1 y
= y 8 dx dy (13)
1/2 1/2
' 1
5
= y(8y − 4) dy = . (14)
1/2 6


6(y − x) 0 ≤ x ≤ y ≤ 1
fX,Y (x, y) = (1)
0 otherwise
(a) The conditional PDF of X given Y is found by dividing the joint PDF by the marginal with
respect to Y . For y < 0 or y > 1, fY (y) = 0. For 0 ≤ y ≤ 1,
' y
$y
fY (y) = 6(y − x) dx = 6xy − 3x2 $0 = 3y 2 (2)
0

2
3y 0 ≤ y ≤ 1
fY (y) = (3)
0 otherwise
Thus for 0 < y ≤ 1,

6(y−x)
fX,Y (x, y) 3y 2
0≤x≤y
fX|Y (x|y) = = (4)
fY (y) 0 otherwise
(b) The minimum mean square estimator of X given Y = y is

' ∞
X̂M (y) = E [X|Y = y] = xfX|Y (x|y) dx (5)
−∞
' y $x=y
6x(y − x) 3x2 y − 2x3 $$
= dx = $ = y/3 (6)
0 3y 2 3y 2 x=0
324
(c) First we must find the marginal PDF for X. For 0 ≤ x ≤ 1,
' '
∞ 1 $y=1
fX (x) = fX,Y (x, y) dy = 6(y − x) dy = 3y 2 − 6xy $y=x (7)
−∞ x
= 3(1 − 2x + x2 ) = 3(1 − x)2 (8)
The conditional PDF of Y given X is

2(y−x)
fX,Y (x, y) 1−2x+x2
x≤y≤1
fY |X (y|x) = = (9)
fX (x) 0 otherwise
(d) The minimum mean square estimator of Y given X is

' ∞
ŶM (x) = E [Y |X = x] = yfY |X (y|x) dy (10)
−∞
' 1
2y(y − x)
= dy (11)
x 1 − 2x + x
2
$y=1
(2/3)y 3 − y 2 x $$ 2 − 3x + x3
= = . (12)
1 − 2x + x2 $y=x 3(1 − x)2
Perhaps surprisingly, this result can be simplified to

x 2
ŶM (x) = + . (13)
3 3
(a) First we find the marginal PDF fY (y). For' 0 ≤ y ≤ 2, '

y ∞ y
fY (y) = fX,Y (x, y) dx = 2 dx = 2y (1)
1 x=0 −∞ 0
Hence, for 0 ≤ y ≤ 2, the conditional PDF of X given Y is
x=y
x fX,Y (x, y) 1/y 0 ≤ x ≤ y
fX|Y (x|y) = = (2)
1 fY (y) 0 otherwise
(b) The optimum mean squared error estimate of X given Y = y is

' ∞ ' y
x
x̂M (y) = E [X|Y = y] = xfX|Y (x|y) dx = dx = y/2 (3)
−∞ 0 y
(c) The MMSE estimator of X given Y is X̂M (Y ) = E[X|Y ] = Y /2. The mean squared error is

e∗X,Y = E (X − X̂M (Y ))2 = E (X − Y /2)2 = E X 2 − XY + Y 2 /4 (4)
325
Of course, the integral must be evaluated.
' 1' y
∗
eX,Y = 2(x2 − xy + y 2 /4) dx dy (5)
0 0
' 1
3 $x=y
= 2x /3 − x2 y + xy 2 /2 $x=0 dy (6)
0
' 1 3
y
= dy = 1/24 (7)
0 6
Another approach to finding the mean square error is to recognize that the MMSE estimator
is a linear estimator and thus must be the optimal linear estimator. Hence, the mean square
error of the optimal linear estimator given by Theorem 9.4 must equal e∗X,Y . That is, e∗X,Y =
Var[X](1 − ρ2X,Y ). However, calculation of the correlation coefficient ρX,Y is at least as much
work as direct calculation of e∗X,Y .
(a) The marginal PMFs of X and Y are listed below

1/3 x = −1, 0, 1 1/4 y = −3, −1, 0, 1, 3
PX (x) = PY (y) = (1)
(b) No, the random variables X and Y are not independent since
PX,Y (1, −3) = 0 = PX (1) PY (−3) (2)
(c) Direct evaluation leads to
E [X] = 0 Var [X] = 2/3 (3)

E [Y ] = 0 Var [Y ] = 5 (4)
This implies
Cov [X, Y ] = Cov [X, Y ] = E [XY ] − E [X] E [Y ] = E [XY ] = 7/6 (5)
(d) From Theorem 9.4, the optimal linear estimate of X given Y is

σX 7
X̂L (Y ) = ρX,Y (Y − µY ) + µX = Y + 0. (6)
σY 30
Therefore, a∗ = 7/30 and b∗ = 0.
(e) From the previous part, X and Y have correlation coefficient

ρX,Y = Cov [X, Y ] / Var[X] Var[Y ] = 49/120. (7)
From Theorem 9.4, the minimum mean square error of the optimum linear estimate is
2 71 71
e∗L = σX
2
(1 − ρ2X,Y ) = = . (8)
3 120 180
326
(f) The conditional probability mass function is
⎧
⎪
⎨
1/6
1/4 = 2/3 x = −1,
PX,Y (x, −3)
PX|Y (x| − 3) = = 1/12 (9)
⎪ 1/4 = 1/3 x = 0.
PY (−3) ⎩
0 otherwise.
(g) The minimum mean square estimator of X given that Y = 3 is

x̂M (−3) = E [X|Y = −3] = xPX|Y (x| − 3) = −2/3. (10)
x
(h) The mean squared error of this estimator is

êM (−3) = E (X − x̂M (−3))2 |Y = −3 (11)

= (x + 2/3)2 PX|Y (x| − 3) (12)
x
= (−1/3)2 (2/3) + (2/3)2 (1/3) = 2/9. (13)

The problem statement tells us that

1/12 −6 ≤ v ≤ 6,
fV (v) = (1)
0 otherwise.
√
Furthermore, we are also told that R = V + X where X is a Gaussian (0, 3) random variable.
(a) The expected value of R is the expected value V plus the expected value of X. We already
know that X has zero expected value, and that V is uniformly distributed between -6 and 6
volts and therefore also has zero expected value. So
E [R] = E [V + X] = E [V ] + E [X] = 0. (2)
(b) Because X and V are independent random variables, the variance of R is the sum of the
variance of V and the variance of X.
Var[R] = Var[V ] + Var[X] = 12 + 3 = 15. (3)
(c) Since E[R] = E[V ] = 0,

Cov [V, R] = E [V R] = E [V (V + X)] = E V 2 = Var[V ]. (4)
(d) The correlation coefficient of V and R is

Cov [V, R] Var[V ] σV
ρV,R = = = . (5)
Var[V ] Var[R] Var[V ] Var[R] σR
The LMSE estimate of V given R is
σV σ2 12
V̂ (R) = ρV,R (R − E [R]) + E [V ] = V2 R = R. (6)
σR σR 15
Therefore a∗ = 12/15 = 4/5 and b∗ = 0.
327
(e) The minimum mean square error in the estimate is
e∗ = Var[V ](1 − ρ2V,R ) = 12(1 − 12/15) = 12/5 (7)

The solution to this problem is to simply calculate the various quantities required for the optimal
linear estimator given by Theorem 9.4. First we calculate the necessary moments of X and Y .
E [X] = −1(1/4) + 0(1/2) + 1(1/4) = 0 (1)

E X 2 = (−1)2 (1/4) + 02 (1/2) + 12 (1/4) = 1/2 (2)
E [Y ] = −1(17/48) + 0(17/48) + 1(14/48) = −1/16 (3)

E Y 2 = (−1)2 (17/48) + 02 (17/48) + 12 (14/48) = 31/48 (4)
E [XY ] = 3/16 − 0 − 0 + 1/8 = 5/16 (5)
The variances and covariance are

Var[X] = E X 2 − (E [X])2 = 1/2 (6)

Var[Y ] = E Y 2 − (E [Y ])2 = 493/768 (7)
Cov [X, Y ] = E [XY ] − E [X] E [Y ] = 5/16 (8)
√
Cov [X, Y ] 5 6
ρX,Y = =√ (9)
Var[X] Var[Y ] 493
By reversing the labels of X and Y in Theorem 9.4, we find that the optimal linear estimator of
Y given X is
σY 5 1
ŶL (X) = ρX,Y (X − E [X]) + E [Y ] = X − (10)
σX 8 16
The mean square estimation error is
e∗L = Var[Y ](1 − ρ2X,Y ) = 343/768 (11)

These four joint PMFs are actually related to each other. In particular, completing the row sums
and column sums shows that each random variable has the same marginal PMF. That is,
PX (x) = PY (x) = PU (x) = PV (x) = PS (x) = PT (x) = PQ (x) = PR (x) (1)

1/3 x = −1, 0, 1
= (2)
0 otherwise
This implies
E [X] = E [Y ] = E [U ] = E [V ] = E [S] = E [T ] = E [Q] = E [R] = 0 (3)
and that

E X2 = E Y 2 = E U 2 = E V 2

= E S 2 = E T 2 = E Q2 = E R2 = 2/3 (4)
Since each random variable has zero expected value,√the second moment equals the variance. Also,
the standard deviation of each random variable is 2/3. These common properties will make it
much easier to answer the questions.
328
(a) Random variables X and Y are independent since for all x and y,
PX,Y (x, y) = PX (x) PY (y) (5)
Since each other pair of random variables has the same marginal PMFs as X and Y but a
different joint PMF, all of the other pairs of random variables must be dependent. Since X
and Y are independent, ρX,Y = 0. For the other pairs, we must compute the covariances.
Cov [U, V ] = E [U V ] = (1/3)(−1) + (1/3)(−1) = −2/3 (6)

Cov [S, T ] = E [ST ] = 1/6 − 1/6 + 0 + −1/6 + 1/6 = 0 (7)
Cov [Q, R] = E [QR] = 1/12 − 1/6 − 1/6 + 1/12 = −1/6 (8)
The correlation coefficient of U and V is

Cov [U, V ] −2/3
ρU,V = √ √ =√ √ = −1 (9)
Var[U ] Var[V ] 2/3 2/3
In fact, since the marginal PMF’s are the same, the denominator of the correlation coefficient
will be 2/3 in each case. The other correlation coefficients are
Cov [S, T ] Cov [Q, R]
ρS,T = =0 ρQ,R = = −1/4 (10)
2/3 2/3
(b) From Theorem 9.4, the least mean square linear estimator of U given V is
σU
ÛL (V ) = ρU,V (V − E [V ]) + E [U ] = ρU,V V = −V (11)
σV
Similarly for the other pairs, all expected values are zero and the ratio of the standard
deviations is always 1. Hence,
X̂L (Y ) = ρX,Y Y = 0 (12)

ŜL (T ) = ρS,T T = 0 (13)
Q̂L (R) = ρQ,R R = −R/4 (14)
From Theorem 9.4, the mean square errors are
e∗L (X, Y ) = Var[X](1 − ρ2X,Y ) = 2/3 (15)

e∗L (U, V ) = Var[U ](1 − ρ2U,V ) = 0 (16)
e∗L (S, T ) = Var[S](1 − ρ2S,T ) = 2/3 (17)
e∗L (Q, R) = Var[Q](1 − ρ2Q,R ) = 5/8 (18)

To solve this problem, we use Theorem 9.4. The only difficulty is in computing E[X], E[Y ], Var[X],
Var[Y ], and ρX,Y . First we calculate the marginal PDFs
' 1
$y=1
fX (x) = 2(y + x) dy = y 2 + 2xy $y=x = 1 + 2x − 3x2 (1)
' y
x
$x=y
fY (y) = 2(y + x) dx = 2xy + x2 $x=0 = 3y 2 (2)
0
329
The first and second moments of X are
' 1
$1
E [X] = (x + 2x2 − 3x3 ) dx = x2 /2 + 2x3 /3 − 3x4 /4$0 = 5/12 (3)
0
' 1
2 $1
E X = (x2 + 2x3 − 3x4 ) dx = x3 /3 + x4 /2 − 3x5 /5$0 = 7/30 (4)
0
The first and second moments of Y are

' 1 '
1
E [Y ] = 3y 3 dy = 3/4, E Y 2
= 3y 4 dy = 3/5. (5)
0 0
Thus, X and Y each have variance

129 3
Var[X] = E X 2 − (E [X])2 = Var[Y ] = E Y 2 − (E [Y ])2 = (6)
2160 80
To calculate the correlation coefficient, we first must calculate the correlation
' 1' y
E [XY ] = 2xy(x + y) dx dy (7)
0 0
' 1 ' 1 4
3 $
2 2 $x=y 5y
= 2x y/3 + x y x=0 dy = dy = 1/3 (8)
0 0 3
Hence, the correlation coefficient is
Cov [X, Y ] E [XY ] − E [X] E [Y ] 5
ρX,Y = =√ (9)
Var[X] Var[Y ] Var[X] Var[Y ] 129
Finally, we use Theorem 9.4 to combine these quantities in the optimal linear estimator.
σX
X̂L (Y ) = ρX,Y (Y − E [Y ]) + E [X] (10)
σY
√
5 129 3 5 5
= √ Y − + = Y. (11)
129 9 4 12 9

The linear mean square estimator of X given Y is

E [XY ] − µX µY
X̂L (Y ) = (Y − µY ) + µX . (1)
Var[Y ]
To find the parameters of this estimator, we calculate
' y
$y
fY (y) = 6(y − x) dx = 6xy − 3x2 $0 = 3y 2 (0 ≤ y ≤ 1) (2)
0
' 1
3(1 + −2x + x2 ) 0 ≤ x ≤ 1,
fX (x) = 6(y − x) dy = (3)
x 0 otherwise.
The moments of X and Y are
' 1 ' 1
E [Y ] = 3y 3 dy = 3/4 E [X] = 3x(1 − 2x + x2 ) dx = 1/4 (4)
0 0
' 1 '
2 1
E Y = 3y 4 dy = 3/5 E X2 = 3x2 (1 + −2x + x2 ) dx = 1/10 (5)
0 0
330
The correlation between X and Y is
' 1' 1
E [XY ] = 6 xy(y − x) dy dx = 1/5 (6)
0 x
Putting these pieces together, the optimal linear estimate of X given Y is

1/5 − 3/16 3 1 Y
X̂L (Y ) = Y − + = (7)
3/5 − (3/4)2 4 4 3

We are told that random variable X has a second order Erlang distribution

λxe−λx x ≥ 0
fX (x) = (1)
0 otherwise
We also know that given X = x, random variable Y is uniform on [0, x] so that

1/x 0 ≤ y ≤ x
fY |X (y|x) = (2)
0 otherwise
(a) Given X = x, Y is uniform on [0, x]. Hence E[Y |X = x] = x/2. Thus the minimum mean
square estimate of Y given X is
ŶM (X) = E [Y |X] = X/2 (3)
(b) The minimum mean square estimate of X given Y can be found by finding the conditional
probability density function of X given Y . First we find the joint density function.

λe−λx 0 ≤ y ≤ x
fX,Y (x, y) = fY |X (y|x) · fX (x) = (4)
0 otherwise
Now we can find the marginal of Y

' ∞ −λy
−λx e y≥0
fY (y) = λe dx = (5)
y 0 otherwise
By dividing the joint density by the marginal density of Y we arrive at the conditional density
of X given Y .
fX,Y (x, y) λe−λ(x−y) x ≥ y
fX|Y (x|y) = = (6)
fY (y) 0 otherwise
Now we are in a position to find the minimum mean square estimate of X given Y . Given
Y = y, the conditional expected value of X is
' ∞
E [X|Y = y] = λxe−λ(x−y) dx (7)
y
Making the substitution u = x − y yields

' ∞
E [X|Y = y] = λ(u + y)e−λu du (8)
0
331
We observe that if U is an exponential random variable with parameter λ, then
1
E [X|Y = y] = E [U + y] = +y (9)
λ
The minimum mean square error estimate of X given Y is
1
X̂M (Y ) = E [X|Y ] = +Y (10)
λ
(c) Since the MMSE estimate of Y given X is the linear estimate ŶM (X) = X/2, the optimal
linear estimate of Y given X must also be the MMSE estimate. That is, ŶL (X) = X/2.
(d) Since the MMSE estimate of X given Y is the linear estimate X̂M (Y ) = Y + 1/λ, the optimal
linear estimate of X given Y must also be the MMSE estimate. That is, X̂L (Y ) = Y + 1/λ.

From the problem statement, we learn the following facts:
−r −rx
e r ≥ 0, re x ≥ 0,
fR (r) = fX|R (x|r) = (1)
Note that fX,R (x, r) > 0 for all non-negative X and R. Hence, for the remainder of the problem,
we assume both X and R are non-negative and we omit the usual “zero otherwise” considerations.
(a) To find r̂M (X), we need the conditional PDF
fX|R (x|r) fR (r)
fR|X (r|x) = . (2)
fX (x)
The marginal PDF of X is
' ∞ ' ∞
fX (x) = fX|R (x|r) fR (r) dr = re−(x+1)r dr (3)
0 0
( (
We use the integration by parts formula u dv = uv − v du by choosing u = r and dv =
e−(x+1)r dr. Thus v = −e−(x+1)r /(x + 1) and
$ ' ∞ $∞
−r −(x+1)r $$∞ 1 −(x+1)r −1 $
−(x+1)r $ 1
fX (x) = e $ + e dr = 2
e $ = (x + 1)2 (4)
x+1 0 x + 1 0 (x + 1) 0
Now we can find the conditional PDF of R given X.

fX|R (x|r) fR (r)
fR|X (r|x) = = (x + 1)2 re−(x+1)r (5)
fX (x)
By comparing, fR|X (r|x) to the Erlang PDF shown in Appendix A, we see that given X = x,
the conditional PDF of R is an Erlang PDF with parameters n = 1 and λ = x + 1. This
implies
1 1
E [R|X = x] = Var [R|X = x] = (6)
x+1 (x + 1)2
Hence, the MMSE estimator of R given X is
1
r̂M (X) = E [R|X] = (7)
X +1
332
(b) The MMSE estimate of X given R = r is E[X|R = r]. From the initial problem statement,
we know that given R = r, X is exponential with expectred value 1/r. That is, E[X|R = r] =
1/r. Another way of writing this statement is
x̂M (R) = E [X|R] = 1/R. (8)
(c) Note that the expected value of X is

' ∞ ' ∞
x
E [X] = xfX (x) dx = dx = ∞ (9)
0 0 (x + 1)2
Because E[X] doesn’t exist, the LMSE estimate of X given R doesn’t exist.
(d) Just as in part (c), because E[X] doesn’t exist, the LMSE estimate of R given X doesn’t
exist.
(a) As a function of a, the mean squared error is

e = E (aY − X)2 = a2 E Y 2 − 2aE [XY ] + E X 2 (1)
Setting de/da|a=a∗ = 0 yields

E [XY ]
a∗ = (2)
E [Y 2 ]
(b) Using a = a∗ , the mean squared error is
(E [XY ])2
e∗ = E X 2 − (3)
E [Y 2 ]
(c) We can write the LMSE estimator given in Theorem 9.4 in the form
σX
x̂L (()Y ) = ρX,Y Y −b (4)
σY
where
σX
b = ρX,Y E [Y ] − E [X] (5)
σY
When b = 0, X̂(Y ) is the LMSE estimate. Note that the typical way that b = 0 occurs when
E[X] = E[Y ] = 0. However, it is possible that the right combination of expected values,
variances, and correlation coefficent can also yield b = 0.
333
In this case, the joint PDF of X and R is
fX,R (x, r) = fX|R (x|r) fR (r) (1)

2
√1
r0 128π
e−(x+40+40 log10 r) /128 0 ≤ r ≤ r0 ,
= (2)
0 otherwise.
From Theorem 9.6, the MAP estimate of R given X = x is the value of r that maximizes
fX|R (x|r)fR (r). Since R has a uniform PDF over [0, 1000],
r̂MAP (x) = arg max fX|R (x|r) fR (r) = arg max fX|R (x|r) (3)
0≤r 0≤r≤1000
Hence, the maximizing value of r is the same as for the ML estimate in Quiz 9.3 unless the
maximizing r exceeds 1000 m. In this case, the maximizing value is r = 1000 m. From the solution
to Quiz 9.3, the resulting ML estimator is

1000 x < −160
r̂ML (x) = (4)
(0.1)10−x/40 x ≥ −160

From the problem statement we know that R is an exponential random variable with expected
value 1/µ. Therefore it has the following probability density function.

µe−µr r ≥ 0
fR (r) = (1)
0 otherwise
It is also known that, given R = r, the number of phone calls arriving at a telephone switch, N , is
a Poisson (α = rT ) random variable. So we can write the following conditional probability mass
function of N given R.
(rT )n e−rT
PN |R (n|r) = n! n = 0, 1, . . . (2)
0 otherwise
(a) The minimum mean square error estimate of N given R is the conditional expected value of
N given R = r. This is given directly in the problem statement as r.
N̂M (r) = E [N |R = r] = rT (3)
(b) The maximum a posteriori estimate of N given R is simply the value of n that will maximize
PN |R (n|r). That is,
n̂M AP (r) = arg max PN |R (n|r) = arg max(rT )n e−rT /n! (4)
n≥0 n≥0
Usually, we set a derivative to zero to solve for the maximizing value. In this case, that tech-
nique doesn’t work because n is discrete. Since e−rT is a common factor in the maximization,
we can define g(n) = (rT )n /n! so that n̂M AP = arg maxn g(n). We observe that
rT
g(n) = g(n − 1) (5)
n
this implies that for n ≤ rT , g(n) ≥ g(n − 1). Hence the maximizing value of n is the largest
n such that n ≤ rT . That is, n̂M AP = rT .
334
(c) The maximum likelihood estimate of N given R selects the value of n that maximizes
fR|N =n (r), the conditional PDF of R given N . When dealing with situations in which we
mix continuous and discrete random variables, its often helpful to start from first principles.
In this case,
fR|N (r|n) dr = P [r < R ≤ r + dr|N = n] (6)

P [r < R ≤ r + dr, N = n]
= (7)
P [N = n]
P [N = n|R = r] P [r < R ≤ r + dr]
= (8)
P [N = n]
In terms of PDFs and PMFs, we have
PN |R (n|r) fR (r)
fR|N (r|n) = (9)
PN (n)
To find the value of n that maximizes fR|N (r|n), we need to find the denominator PN (n).
' ∞
PN (n) = PN |R (n|r) fR (r) dr (10)
−∞
' ∞
(rT )n e−rT −µr
= µe dr (11)
n!
0
' ∞
µT n
= rn (µ + T )e−(µ+T )r dr (12)
n!(µ + T ) 0
µT n
= E [X n ] (13)
n!(µ + T )
where X is an exponential random variable with expected value 1/(µ + T ). There are several
ways to derive the nth moment of an exponential random variable including integration by
parts. In Example 6.5, the MGF is used to show that E[X n ] = n!/(µ + T )n . Hence, for n ≥ 0,
µT n
PN (n) = (14)
(µ + T )n+1
Finally, the conditional PDF of R given N is

(rT )n e−rT
PN |R (n|r) fR (r) n! µe−µr
fR|N (r|n) = = µT n (15)
PN (n)
(µ+T )n+1
[(µ + T )r]n e−(µ+T )r
= (µ + T ) (16)
n!
The ML estimate of N given R is
[(µ + T )r]n e−(µ+T )r

n̂M L (r) = arg max fR|N (r|n) = arg max(µ + T ) (17)
n≥0 n≥0 n!
This maximization is exactly the same as in the previous part except rT is replaced by
(µ + T )r. The maximizing value of n is n̂M L = (µ + T )r.
335
Both parts (a) and (b) rely on the conditional PDF of R given N = n. When dealing with
situations in which we mix continuous and discrete random variables, its often helpful to start from
first principles.
P [r < R ≤ r + dr, N = n]
fR|N (r|n) dr = P [r < R ≤ r + dr|N = n] = (1)
P [N = n]
P [N = n|R = r] P [r < R ≤ r + dr]
= (2)
P [N = n]
In terms of PDFs and PMFs, we have
PN |R (n|r) fR (r)
fR|N (r|n) = (3)
PN (n)
To find the value of n that maximizes fR|N (r|n), we need to find the denominator PN (n).
' ∞
PN (n) = PN |R (n|r) fR (r) dr (4)
−∞
' ∞
(rT )n e−rT −µr
= µe dr (5)
n!
0
' ∞
µT n µT n
= rn (µ + T )e−(µ+T )r dr = E [X n ] (6)
n!(µ + T ) 0 n!(µ + T )
where X is an exponential random variable with expected value 1/(µ + T ). There are several ways
to derive the nth moment of an exponential random variable including integration by parts. In
Example 6.5, the MGF is used to show that E[X n ] = n!/(µ + T )n . Hence, for n ≥ 0,
µT n
PN (n) = (7)
(µ + T )n+1
Finally, the conditional PDF of R given N is
(rT )n e−rT
PN |R (n|r) fR (r) n! µe−µr (µ + T )n+1 rn e−(µ+T )r
fR|N (r|n) = = µT n
= (8)
PN (n) n!
(µ+T )n+1
(a) The MMSE estimate of R given N = n is the conditional expected value E[R|N = n]. Given
N = n, the conditional PDF oF R is that of an Erlang random variable or order n + 1. From
Appendix A, we find that E[R|N = n] = (n + 1)/(µ + T ). The MMSE estimate of R given
N is
N +1
R̂M (N ) = E [R|N ] = (9)
µ+T
(b) The MAP estimate of R given N = n is the value of r that maximizes fR|N (r|n).
(µ + T )n+1 n −(µ+T )r
R̂MAP (n) = arg max fR|N (r|n) = arg max r e (10)
r≥0 r≥0 n!
By setting the derivative with respect to r to zero, we obtain the MAP estimate
n
R̂MAP (n) = (11)
µ+T
336
(c) The ML estimate of R given N = n is the value of R that maximizes PN |R (n|r). That is,
(rT )n e−rT
R̂ML (n) = arg max (12)
r≥0 n!
Seting the derivative with respect to r to zero yields
R̂ML (n) = n/T (13)

This problem is closely related to Example 9.7.
(a) Given Q = q, the conditional PMF of K is
n k
k q (1 − q)
n−k k = 0, 1, . . . , n
PK|Q (k|q) = (1)
0 otherwise
The ML estimate of Q given K = k is
q̂ML (k) = arg max PQ|K (q|k) (2)
0≤q≤1
Differentiating PQ|K (q|k) with respect to q and setting equal to zero yields

dPQ|K (q|k) n
= kq k−1 (1 − q)n−k − (n − k)q k (1 − q)n−k−1 = 0 (3)
dq k
T‘¡he maximizing value is q = k/n so that
K
Q̂ML (K) = (4)
n
(b) To find the PMF of K, we average over all q.
' ∞ ' 1
n k
PK (k) = PK|Q (k|q) fQ (q) dq = q (1 − q)n−k dq (5)
−∞ 0 k
We can evaluate this integral by expressing it in terms of the integral of a beta PDF. Since
(n+1)!
β(k + 1, n − k + 1) = k!(n−k)! , we can write
' 1
1 1
PK (k) = β(k + 1, n − k + 1)q k (1 − q)n−k dq = (6)
n+1 0 n+1
That is, K has the uniform PMF

1/(n + 1) k = 0, 1, . . . , n
PK (k) = (7)
0 otherwise
Hence, E[K] = n/2.
(c) The conditional PDF of Q given K is

(n+1)! k
PK|Q (k|q) fQ (q)
k!(n−k)! q (1 − q)n−k 0 ≤ q ≤ 1
fQ|K (q|k) = = (8)
PK (k) 0 otherwise
That is, given K = k, Q has a beta (k + 1, n − k + 1) PDF.
(d) The MMSE estimate of Q given K = k is the conditional expectation E[Q|K = k]. From the
beta PDF described in Appendix A, E[Q|K = k] = (k + 1)/(n + 2). The MMSE estimator is
K +1
Q̂M (K) = E [Q|K] = (9)
n+2
337
From the problem statement, we learn for vectors X = X1 X2 X3 and Y = Y1 Y2 that
⎡ ⎤
1 3/4 1/2
1 1 0
E [X] = 0, RX = ⎣3/4 1 3/4⎦ , Y = AX = X. (1)
0 1 1
1/2 3/4 1
(a) Since E[Y] = AE[X] = 0, we can apply Theorem 9.7(a) which states that the minimum
mean square error estimate of X1 is X̂1 (Y) = â Y where â = R−1 Y RYX1 . The rest of the
solution is just calculation. (We note that even in the case of a 3 × 3 matrix, its convenient
to use Matlab with format rat mode to perform the calculations and display the results
as nice fractions.) From Theorem 5.13,
⎡ ⎤⎡ ⎤
1 3/4 1/2 1 0
1 1 0 ⎣ ⎦ ⎣ ⎦ 7/2 3
RY = ARX A = 3/4 1 3/4 1 1 = . (2)
0 1 1 3 7/2
1/2 3/4 1 0 1
In addition, since RYX1 = E[YX1 ] = E[AXX1 ] = AE[XX1 ],

⎡ 2 ⎤ ⎡ ⎤ ⎡ ⎤
E X1 RX (1, 1) 1
⎣ ⎦ ⎣ ⎦ 1 1 0 ⎣ ⎦ 7/4
RYX1 = A E [X2 X1 ] = A RX (2, 1) = 3/4 = . (3)
0 1 1 5/4
E [X3 X1 ] RX (3, 1) 1/2
Finally,
14/13 −12/13 7/4 1 19
â = R−1
Y RYX1 = = . (4)
−12/13 14/13 5/4 26 −7
Thus the linear MMSE estimator of X1 given Y is
19 7
X̂1 (Y) = â Y = Y1 − Y2 = 0.7308Y1 − 0.2692Y2 . (5)
26 26
(b) By Theorem 9.7(c), the mean squared error of the optimal estimator is
e∗L = Var[X1 ] − â RYX1 (6)

= RX (1, 1) − RYX1 R−1
Y RYX1 (7)

14/13 −12/13 7/4 3
= 1 − 7/4 5/4 = (8)
−12/13 14/13 5/4 52
(c) We can estimate random variable X1 based on the observation of random variable Y1 using
Theorem 9.4. Note that Theorem 9.4 is just a special case of Theorem 9.8 in which the
observation is a random vector. In any case, from Theorem 9.4, the optimum linear estimate
is X̂1 (Y1 ) = a∗ Y1 + b∗ where
Cov [X1 , Y1 ]
a∗ = , b∗ = µX1 − a∗ µY1 . (9)
Var[Y1 ]
Since Y1 = X1 + X2 , we see that
µX1 = E [X1 ] = 0, µY1 = E [Y1 ] = E [X1 ] + E [X2 ] = 0. (10)
338
These facts, along with RX and RY from part (a), imply
Cov [X1 , Y1 ] = E [X1 Y1 ] (11)
= E [X1 (X1 + X2 )] = RX (1, 1) + RX (1, 2) = 7/4, (12)

Var[Y1 ] = E Y12 = RY (1, 1) = 7/2 (13)
Thus
Cov [X1 , Y1 ] 7/4 1
a∗ = = = , b∗ = µX1 − a∗ µY1 = 0. (14)
Var[Y1 ] 7/2 2
Thus the optimum linear estimate of X1 given Y1 is
1
X̂1 (Y1 ) = Y1 . (15)
2
From Theorem 9.4(b), the mean square error of this estimator is
e∗L = σX
2
1
(1 − ρ2X1 ,Y1 ) (16)
Since X1 and Y1 have zero expected value, σX 2 = R (1, 1) = 1 and σ 2 = R (1, 1) = 7/2.
1 X Y1 Y
Also, since Cov[X1 , Y1 ] = 7/4, we see that
)
Cov [X1 , Y1 ] 7/4 7
ρX1 ,Y1 = = = . (17)
σX1 σY1 7/2 8

Thus e∗L = 1 − ( 7/8)2 = 1/8. Note that 1/8 > 3/52. As we would expect, the estimate of
X1 based on just Y1 has larger mean square error than the estimate based on both Y1 and Y2 .

From the problem statement, we learn for vectors X = X1 X2 X3 and W = W1 W2 that
⎡ ⎤
1 3/4 1/2
0.1 0
E [X] = 0, RX = ⎣3/4 1 3/4⎦ , E [W] = 0, RW = (1)
0 0.1
1/2 3/4 1
In addition,
Y1 1 1 0
Y= = AX + W = X + W. (2)
Y2 0 1 1
(a) Since E[Y] = AE[X] = 0, we can apply Theorem 9.7(a) which states that the minimum
mean square error estimate of X1 is X̂1 (Y) = â Y where â = R−1
Y RYX1 . First we find RY .

RY = E YY = E (AX + W)(AX + W) (3)

= E (AX + W)(X A + W ) (4)

= E AXX A + E WX A + E AXW + E WW (5)
Since X and W are independent, E[WX ] = 0 and E[XW ] = 0. This implies

RY = AE XX A + E WW (6)

= ARX A + RW (7)
⎡ ⎤⎡ ⎤
1 3/4 1/2 1 0
1 1 0 ⎣ 0.1 0 3.6 3
= 3/4 1 3/4⎦ ⎣1 1⎦ + = . (8)
0 1 1 0 0.1 3 3.6
1/2 3/4 1 0 1
339
Once again, independence of W and X1 yields
RYX1 = E [YX1 ] = E [(AX + W)X1 ] = AE [XX1 ] . (9)
This implies
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
E X12 RX (1, 1) 1
⎣ ⎦ ⎣ ⎦ 1 1 0 ⎣ ⎦ 7/4
RYX1 = A E [X2 X1 ] = A RX (2, 1) = 3/4 = . (10)
0 1 1 5/4
E [X3 X1 ] RX (3, 1) 1/2
Putting these facts together, we find that

−1 10/11 −25/33 7/4 1 85
â = RY RYX1 = = . (11)
−25/33 10/11 5/4 132 −25

85 25
X̂1 (Y) = â Y = Y1 − Y2 = 0.6439Y1 − 0.1894Y2 . (12)
132 132
e∗L = Var[X1 ] − â RYX1 (13)

= RX (1, 1) − RYX1 R−1
Y RYX1 (14)

10/11 −25/33 7/4 29
= 1 − 7/4 5/4 = = 0.1098. (15)
−25/33 10/11 5/4 264
In Problem 9.4.1, we solved essentialy the same problem but the observations Y were not
subjected to the additive noise W. In comparing the estimators, we see that the additive noise
perturbs the estimator somewhat but not dramatically because the correaltion structure of
X and the mapping A from X to Y remains unchanged. On the other hand, in the noiseless
case, the resulting mean square error was about half as much, 3/52 = 0.0577 versus 0.1098.
Theorem 9.4. Note that Theorem 9.4 is a special case of Theorem 9.8 in which the observation
is a random vector. In any case, from Theorem 9.4, the optimum linear estimate is X̂1 (Y1 ) =
a∗ Y1 + b∗ where
Cov [X1 , Y1 ]
a∗ = , b∗ = µX1 − a∗ µY1 . (16)
Var[Y1 ]
Since E[Xi ] = µXi = 0 and Y1 = X1 + X2 + W1 , we see that
µY1 = E [Y1 ] = E [X1 ] + E [X2 ] + E [W1 ] = 0. (17)
These facts, along with independence of X1 and W1 , imply
Cov [X1 , Y1 ] = E [X1 Y1 ] = E [X1 (X1 + X2 + W1 )] (18)

= RX (1, 1) + RX (1, 2) = 7/4 (19)
In addition, using RY from part (a), we see that

Var[Y1 ] = E Y12 = RY (1, 1) = 3.6. (20)
340
Thus
Cov [X1 , Y1 ] 7/4 35
a∗ = = = , b∗ = µX1 − a∗ µY1 = 0. (21)
Var[Y1 ] 3.6 72
35
X̂1 (Y1 ) = Y1 . (22)
72
e∗L = σX
2
1
(1 − ρ2X1 ,Y1 ) (23)
Since X1 and Y1 have zero expected value, σX 2 = R (1, 1) = 1 and σ 2 = R (1, 1) = 3.6.
1 X Y1 Y
Also, since Cov[X1 , Y1 ] = 7/4, we see that
√
Cov [X1 , Y1 ] 7/4 490
ρX1 ,Y1 = =√ = . (24)
σX1 σY1 3.6 24
Thus e∗L = 1 − (490/242 ) = 0.1493. As we would expect, the estimate of X1 based on just Y1
has larger mean square error than the estimate based on both Y1 and Y2 .

From the problem statement, we learn for vectors X = X1 X2 X3 and W = W1 W2 that
⎡ ⎤ ⎡ ⎤
−1 1 3/4 1/2
0 0.1 0
E [X] = 0 ⎦ ,
⎣ RX = ⎣3/4 1 1/2⎦ , E [W] = , RW = (1)
0 0 0.1
1 1/2 3/4 1
In addition, from Theorem 5.12,
CX = RX − µX µX (2)
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
1 3/4 1/2 −0.1 0.99 0.75 0.51
= ⎣3/4 1 1/2⎦ − ⎣ 0 ⎦ −0.1 0 0.1 = ⎣0.75 1.0 0.75⎦ , (3)
1/2 3/4 1 0.1 0.51 0.75 0.99
and
Y1 1 1 0
Y= = AX + W = X + W. (4)
Y2 0 1 1
(a) Since E[X] is nonzero, we use Theorem 9.8 which states that the minimum mean square error
estimate of X1 is X̂1 (Y) = â Y + b̂ where â = C−1
Y CYX1 and b̂ = E[X1 ] − â E[Y].
First we find RY . Since E[Y] = AE[X] + E[W] = AE[X],

CY = E (Y − E [Y])(Y − E [Y]) (5)

= E (A(X − E [X]) + W)(A(X − E [X]) + W) (6)

= E (A(X − E [X]) + W)((X − E [X]) A + W ) (7)

= AE (X − E [X])(X − E [X]) A

+ E W(X − E [X]) A + AE (X − E [X])W + E WW (8)
341
Since X and W are independent, E[W(X − E[X]) ] = 0 and E[(X − E[X])W ] = 0. This
implies
CY = ACX A + RW (9)
⎡ ⎤⎡ ⎤
0.99 0.75 0.51 1 0
1 1 0 ⎣ 0.1 0 3.59 3.01
= 0.75 1.0 0.75⎦ ⎣1 1⎦ + = . (10)
0 1 1 0 0.1 3.01 3.59
0.51 0.75 0.99 0 1
Once again, independence of W and X1 yields E[W(X1 − E[X1 ])] = 0, implying
CYX1 = E [(Y − E [Y])(X1 − E [X1 ])] (11)

= E [(A(X − E [X]) + W)(X1 − E [X1 ])] (12)
= AE [(X − E [X])(X1 − E [X1 ])] (13)
⎡ ⎤ ⎡ ⎤
CX (1, 1) 0.99
⎣ ⎦ 1 1 0 ⎣ ⎦ 1.74
= A CX (2, 1) = 0.75 = . (14)
0 1 1 1.26
CX (3, 1) 0.51
Finally,
0.9378 −0.7863 1.74 0.6411
â = C−1
Y CYX1 = = (15)
−0.7863 0.9378 1.26 −0.1865
Next, we find that
⎡ ⎤
−0.1
1 1 0 ⎣ −0.1
E [Y] = AE [X] = 0 ⎦= . (16)
0 1 1 0.1
0.1
This implies

−0.1
b̂ = E [X1 ] − â E [Y] = −0.1 − 0.6411 −0.1865 = −0.0172 (17)
0.1
X̂1 (Y) = â Y + b̂ = 0.6411Y1 − 0.1865Y2 + 0.0172. (18)
We note that although E[X] = 0.1, the estimator’s offset is only 0.0172. This is because the
change in E[X] is also included in the change in E[Y ].
(b) By Theorem 9.8(c), the mean square error of the optimal estimator is
e∗L = Var[X1 ] − â CYX1 (19)

= CX (1, 1) − CYX1 C−1
Y CYX1 (20)

0.9378 −0.7863 1.7400
= 0.99 − 1.7400 1.2600 = 0.1096. (21)
−0.7863 0.9378 1.2600
Theorem 9.4. Note that Theorem 9.4 is a special case of Theorem 9.8 in which the observation
is a random vector. In any case, from Theorem 9.4, the optimum linear estimate is X̂1 (Y1 ) =
a∗ Y1 + b∗ where
Cov [X1 , Y1 ]
a∗ = , b∗ = µX1 − a∗ µY1 . (22)
Var[Y1 ]
342
First we note that E[X1 ] = µX1 = −0.1. Second, since Y1 = X1 + X2 + W1 and E[W1 ] = 0,
we see that
µY1 = E [Y1 ] = E [X1 ] + E [X2 ] + E [W1 ] (23)

= E [X1 ] + E [X2 ] (24)
= −0.1 (25)
These facts, along with independence of X1 and W1 , imply
Cov [X1 , Y1 ] = E [(X1 − E [X1 ])(Y1 − E [Y1 ])] (26)

= E [(X1 − E [X1 ])((X1 − E [X1 ]) + (X2 − EX2 ) + W1 )] (27)
= CX (1, 1) + CX (1, 2) = 1.74 (28)
In addition, using CY from part (a), we see that

Var[Y1 ] = E Y12 = CY (1, 1) = 3.59. (29)
Thus
Cov [X1 , Y1 ] 1.74
a∗ = = = 0.4847, b∗ = µX1 − a∗ µY1 = −0.0515. (30)
Var[Y1 ] 3.59
X̂1 (Y1 ) = 0.4847Y1 − 0.0515. (31)
e∗L = σX
2
1
(1 − ρ2X1 ,Y1 ) (32)
Note that σX2 = R (1, 1) = 0.99 and σ 2 = R (1, 1) = 3.59. Also, since Cov[X , Y ] = 1.74,
1 X Y1 Y 1 1
we see that
Cov [X1 , Y1 ] 1.74
ρX1 ,Y1 = = = 0.923. (33)
σX1 σY1 (0.99)(3.59)
Thus e∗L = 0.99(1 − (0.923)2 ) = 0.1466. As we would expect, the estimate of X1 based on
just Y1 has larger mean square error than the estimate based on both Y1 and Y2 .

From Theorem 9.7(a), we know that the minimum mean square error estimate of X given Y is
X̂L (Y) = â Y, where â = R−1
Y RYX . In this problem, Y is simply a scalar Y and â is a scalar â.
Since E[Y ] = 0,
RY = E YY = E Y 2 = σY2 . (1)
Similarly,
RYX = E [YX] = E [Y X] = Cov [X, Y ] . (2)
It follows that
2 −1 σX Cov [X, Y ] σX
â = R−1
Y RYX = σY Cov [X, Y ] = = ρX,Y . (3)
σY σX σY σY
343
In this problem, we view Y = X1 X2 as the observation and X = X3 as the variable we wish
to estimate. Since E[X] = 0, we can use Theorem 9.7(a) to find the minimum mean square error
estimate X̂L (Y) = â Y where â = R−1
Y RYX .
(a) In this case,

E X12 E [X
RY = E YY = 1X2]
(1)
E [X2 X1 ] E X22

r r 1 −0.8
= 11 12 = . (2)
r21 r22 −0.8 1
Similarly,

X1 E [X1 X3 ] r13 0.64
RYX = E [YX] = E X3 = = = . (3)
X2 E [X2 X3 ] r23 −0.8
Thus −1
25/9 20/9 0.64 0
â = R−1
Y RYX1 = = . (4)
20/9 25/9 −0.8 −0.8
The optimum linear estimator of X3 given X1 and X2 is
X̂3 = −0.8X2 . (5)
The fact that this estimator depends only on X2 while ignoring X1 is an example of a result
to be proven in Problem 9.4.6.

∗
0.64
eL = Var[X3 ] − â RYX3 = 1 − 0 −0.8 = 0.36 (6)
−0.8
(c) In the previous part, we found that the optimal linear estimate of X3 based on the observation
of random variables X1 and X2 employed only X2 . Hence this same estimate, X̂3 = −0.8X2 ,
is the optimal linear estimate of X3 just using X2 . (This can be derived using Theorem 9.4,
if you wish to do more algebra.)
(d) Since the estimator is the same, the mean square error is still e∗L = 0.36.

For this problem, let Y = X1 X2 · · · n−1 and let X = Xn . Since E[Y] = 0 and E[X] = 0,
Theorem 9.7(a) tells us that the minimum mean square linear estimate of X given Y is X̂n (Y) =
â Y, where â = R−1
Y RYX . This implies that â is the solution to
RY â = RYX . (1)
Note that
⎡ ⎤ ⎡⎡ ⎤ ⎤ ⎡ n−1 ⎤
1 c ··· cn−2 X1 c
⎢ .. .. ⎥ ⎢⎢ ⎥ ⎥ ⎢cn−2 ⎥
⎢ c c2 . . ⎥ ⎢⎢ X2 ⎥ ⎥ ⎢ ⎥
RY = E YY = ⎢

⎢
⎥,
⎥ RYX = E ⎢⎢ .. ⎥ Xn ⎥ = ⎢ .. ⎥ . (2)
⎣
..
.
..
.
..
. c ⎦ ⎣⎣ . ⎦ ⎦ ⎣ . ⎦
cn−2 · · · c 1 Xn−1 c
344

We see that the last column of cRY equals RYX . Equivalently, if â = 0 · · · 0 c , then
RY â = RYX . It follows that the optimal linear estimator of Xn given Y is
X̂n (Y) = â Y = cXn−1 , (3)
which completes the proof of the claim.
The mean square error of this estimate is

e∗L = E (Xn − cXn−1 )2 (4)
= RX (n, n) − cRX (n, n − 1) − cRX (n − 1, n) + c2 RX (n − 1, n − 1) (5)
= 1 − 2c + c = 1 − c
2 2 2
(6)
When c is close to 1, Xn−1 and Xn are highly correlated and the estimation error will be small.
Comment: We will see in Chapter 11 that correlation matrices with this structure arise frequently
in the study of wide sense stationary random sequences. In fact, if you read ahead, you will find
that the claim we just proved is the essentially the same as that made in Theorem 11.10.
(a) In this case, we use the observation Y to estimate each Xi . Since E[Xi ] = 0,

k
√
E [Y] = E [Xj ] pj Sj + E [N] = 0. (1)
j=1
Thus, Theorem 9.7(a) tells us that the MMSE linear estimate of Xi is X̂i (Y) = â Y where
â = R−1
Y RYXi . First we note that
⎡⎛ ⎞ ⎤

k
√
RYXi = E [YXi ] = E ⎣⎝ Xj pj Sj + N⎠ Xi ⎦ (2)
j=1
Since N and Xi are independent, E[NXi ] = E[N]E[Xi ] = 0. Because Xi and Xj are

independent for i = j, E[Xi Xj ] = E[Xi ]E[Xj ] = 0 for i = j. In addition, E[Xi2 ] = 1, and it
follows that
k
√ √
RYXi = E [Xj Xi ] pj Sj + E [NXi ] = pi Si . (3)
j=1
For the same reasons,
⎡⎛ ⎞! "⎤
k
√ k
√
RY = E YY = E ⎣⎝ pj Xj Sj + N⎠ pl Xl S l + N ⎦ (4)
j=1 l=1

k
k
√
= pj pl E [Xj Xl ] Sj S l
j=1 l=1

k
√
k
√
+ pj E [Xj N] Sj + pl E Xl N S l + E NN (5)

j=1 l=1
=0 =0

k
= pj Sj S j + σ 2 I (6)
j=1
345
Now we use a linear algebra identity. For a matrix S with columns S1 , S2 , . . . , Sk , and a
diagonal matrix P = diag[p1 , p2 , . . . , pk ],

k
pj Sj S j = SPS . (7)
j=1
Although this identity may be unfamiliar, it is handy in manipulating correlation matrices.

(Also, if this is unfamiliar, you may wish to work out an example with k = 2 vectors of length
2 or 3.) Thus,
RY = SPS + σ 2 I, (8)
and
2 −1 √
â = R−1
Y RYXi = SPS + σ I pi Si . (9)
Recall that if C is symmetric, then C−1 is also symmetric. This implies the MMSE estimate
of Xi given Y is
√ −1
X̂i (Y) = â Y = pi S i SPS + σ 2 I Y (10)
(b) We observe that V = (SPS + σ 2 I)−1 Y is a vector that does not depend on which bit Xi
√
that we want to estimate. Since X̂i = pi S i V, we can form the vector of estimates
⎡ ⎤ ⎡√ ⎤ ⎡√ ⎤⎡ ⎤
X̂1 p1 S 1 V p1 S 1
⎢ .. ⎥ ⎢ . ⎥ ⎢ .. ⎥ ⎢ .. ⎥
X̂ = ⎣ . ⎦ = ⎣ .. ⎦=⎣ . ⎦⎣ . ⎦V (11)
√ √
X̂k pk S k V pk S k
1/2
=P SV (12)
−1
= P1/2 S SPS + σ 2 I Y (13)

This problem can be solved using the function mse defined in Example
9.10. All we need to do is
define the correlation structure of the vector X = X1 · · · X21 . Just as in Example 9.10, we
do this by defining just the first row of the correlation matrix. Here are the commands we need,
and the resulting plot.
r1=sinc(0.1*(0:20)); mse(r1); 1
hold on;
r5=sinc(0.5*(0:20)); mse(r5); 0.8
r9=sinc(0.9*(0:20)); mse(r9); 0.6
0.4
0.2
0
0 5 10 15 20 25
Although the plot lacks labels, there are three curves for the mean square error MSE(n) corre-
sponding to φ0 ∈ {0.1, 0.5, 0.9}. Keep in mind that MSE(n) is the MSE of the linear estimate of
X21 using random variables X1 , . . . , Xn .
If you run the commands, you’ll find that the φ0 = 0.1 yields the lowest mean square error
while φ0 = 0.9 results in the highest mean square error. When φ0 = 0.1, random variables Xn
for n = 10, 11, . . . , 20 are increasingly correlated with X21 . The result is that the MSE starts to
346
decline rapidly for n > 10. As φ0 increases, fewer observations Xn are correlated with X21 . The
result is the MSE is simply worse as φ0 increases. For example, when φ0 = 0.9, even X20 has only
a small correlation with X21 . We only get a good estimate of X21 at time n = 21 when we observe
X21 + W21 .

This problem can be solved using the function mse defined in Example
9.10. All we need to do is
define the correlation structure of the vector X = X1 · · · X21 . Just as in Example 9.10, we
do this by defining just the first row r of the correlation matrix. Here are the commands we need,
and the resulting plots.
0.1 0.1 0.1
0.05 0.05 0.05
0 0 0
0 10 20 0 10 20 0 10 20
n=0:20; n=0:20; n=0:20;
r1=cos(0.1*pi*n); r5=cos(0.5*pi*n); r9=cos(0.9*pi*n);
mse(r1); mse(r5); mse(r9);
All three cases report similar results for the mean square error (MSE). the reason is that in all
three cases, X1 and X21 are completely correlated; that is, ρX1 ,X21 = 1. As a result, X1 = X21 so
that at time n = 1, the observation is
Y1 = X1 + W1 = X21 + W1 . (1)
The MSE at time n = 1 is 0.1, corresponding to the variance of the additive noise. Subsequent
improvements in the estimates are the result of making other measurements of the form Yn =
Xn + Wn where Xn is highly correlated with X21 . The result is a form of averaging of the additive
noise, which effectively reduces its variance.
The choice of φ0 changes the values of n for which Xn and X21 are highly correlated. However,
in all three cases, there are several such values of n and the result in all cases is an improvement
in the MSE due to averaging the additive noise.

The solution to this problem is almost the same as the solution to Example 9.10, except perhaps
the Matlab code is somewhat simpler. As in the example, let W(n) , X(n) , and Y(n) denote the
vectors, consisting of the first n components of W, X, and Y. Just as in Examples 9.8 and 9.10,
independence of X(n) and W(n) implies that the correlation matrix of Y(n) is

RY(n) = E (X(n) + W(n) )(X(n) + W(n) ) = RX(n) + RW(n) (1)
Note that RX(n) and RW(n) are the n × n upper-left submatrices of RX and RW . In addition,
⎡⎡ ⎤ ⎤ ⎡ ⎤
X 1 + W1 r0
⎢⎢ .. ⎥ ⎥ ⎢ .. ⎥
RY(n) X = E ⎣⎣ . ⎦ X1 ⎦ = ⎣ . ⎦ . (2)
X n + Wn rn−1
347
Compared to the solution of Example 9.10, the only difference in the solution is in the reversal of
the vector RY(n) X . The optimal filter based on the first n observations is â(n) = R−1 R (n) ,
Y (n) Y X
and the mean square error is
e∗L = Var[X1 ] − (â(n) ) RY(n) X . (3)
function e=mse953(r) The program mse953.m simply calculates the mean

N=length(r); square error e∗L . The
input is the
vector r correspond-
e=[]; ing to the vector r0 · · · r20 , which holds the first
for n=1:N, row of the Toeplitz correlation matrix RX . Note that
RYX=r(1:n)’;
RX(n) is the Toeplitz matrix whose first row is the first
RY=toeplitz(r(1:n))+0.1*eye(n);
a=RY\RYX; n elements of r.
en=r(1)-(a’)*RYX;
e=[e;en];
end
plot(1:N,e);
To plot the mean square error as a function of the number of observations, n, we generate the
vector r and then run mse953(r). For the requested cases (a) and (b), the necessary Matlab
commands and corresponding mean square estimation error output as a function of n are shown
here:
0.1 0.1
MSE
MSE
0.05 0.05
0 0
0 5 10 15 20 25 0 5 10 15 20 25
n n
ra=sinc(0.1*pi*(0:20)); rb=cos(0.5*pi*(0:20));
mse953(ra) mse953(rb)
(a) (b)
In comparing the results of cases (a) and (b), we see that the mean square estimation error depends
strongly on the correlation structure given by r|i−j| . For case (a), Y1 is a noisy observation of X1
and is highly correlated with X1 . The MSE at n = 1 is just the variance of W1 . Additional samples
of Yn mostly help to average the additive noise. Also, samples Xn for n ≥ 10 have very little
correlation with X1 . Thus for n ≥ 10, the samples of Yn result in almost no improvement in the
estimate of X1 .
In case (b), Y1 = X1 + W1 , just as in case (a), is simply a noisy copy of X1 and the estimation
error is due to the variance of W1 . On the other hand, for case (b), X5 , X9 , X13 and X17 and X21
are completely correlated with X1 . Other samples also have significant correlation with X1 . As a
result, the MSE continues to go down with increasing n.

When the transmitted bit vector is X, the received signal vector Y = SP1/2 X+N. In Problem 9.4.7,
we found that the linear MMSE estimate of the vector X is
−1
X̂ = P1/2 S SPS + σ 2 I Y. (1)
As indicated in the problem statement, the evaluation of the LMSE detector follows the approach
outlined in Problem 8.4.6. Thus the solutions are essentially the same.
348
function S=randomsignals(n,k); The transmitted data vector x belongs to the set Bk of
%S is an n by k matrix, columns are all binary ±1 vectors of length k. This short program
%random unit length signal vectors generates k random signals, each of length n. Each
S=(rand(n,k)>0.5); random signal is a binary ±1 sequence normalized to
S=((2*S)-1.0)/sqrt(n);
length 1.
The evaluation of the LMSE detector is most similar to evaluation of the matched filter (MF)
detector in Problem 8.4.6. We define a function err=lmsecdmasim(S,P,m) that simulates the
LMSE detector for m symbols for a given set of signal vectors S. In lmsecdmasim, there is no
need for looping. The mth transmitted symbol is represented by the mth column of X and the
corresponding received signal is given by the mth column of Y. The matched filter processing can
be applied to all m columns at once. A second function Pe=lmsecdma(n,k,snr,s,m) cycles through
all combination of users k and SNR snr and calculates the bit error rate for each pair of values.
Here are the functions:
function e=lmsecdmasim(S,P,m); function Pe=lmsecdma(n,k,snr,s,m);
%err=lmsecdmasim(P,S,m); %Pe=lmsecdma(n,k,snr,s,m);
%S= n x k matrix of signals %RCDMA, LMSE detector, users=k
%P= diag matrix of SNRs %proc gain=n, rand signals/frame
% SNR=power/var(noise) % s frames, m symbols/frame
k=size(S,2); %no. of users %See Problem 9.5.4 Solution
n=size(S,1); %proc. gain [K,SNR]=ndgrid(k,snr);
P2=sqrt(P); Pe=zeros(size(SNR));
X=randombinaryseqs(k,m); for j=1:prod(size(SNR)),
Y=S*P2*X+randn(n,m); p=SNR(j);kt=K(j); e=0;
L=P2*S’*inv((S*P*S’)+eye(n)); for i=1:s,
XR=sign(L*Y); S=randomsignals(n,kt);
e=sum(sum(XR ~= X)); e=e+lmsecdmasim(S,p*eye(kt),m);
end
Pe(j)=e/(s*m*kt);
disp([snr kt e]);
end
Here is a run of lmsecdma.
>> pelmse = lmsecdma(32,k,4,1000,1000);
4 2 48542
4 4 109203
4 8 278266
4 16 865358
4 32 3391488
>> pelmse’
ans =
0.0243 0.0273 0.0348 0.0541 0.1060
>>
For processing gain n = 32, the maximum likelihood detector is too complex for my version of
Matlab to run quickly. Instead we can compare the LMSE detector to the matched filter (MF)
detector of Problem 8.4.6 and the decorrelator of Problem 8.4.7 with the following script:
k=[2 4 8 16 32];
pemf = mfrcdma(32,k,4,1000,1000);
pedec=berdecorr(32,k,4,10000);
pelmse = lmsecdma(32,k,4,1000,1000);
plot(k,pemf,’-d’,k,pedec,’-s’,k,pelmse);
legend(’MF’,’DEC’,’LMSE’,2);
axis([0 32 0 0.5]);
349
The resulting plot resembles
0.5
MF
0.4 DEC
LMSE
0.3
0.2
0.1
0
0 5 10 15 20 25 30
Compared to the matched filter and the decorrelator, the linear pre-processing of the LMSE detector
offers an improvement in the bit error rate. Recall that for each bit Xi , the decorrelator zeroes
out the interference from all users j = i at the expense of enhancing the receiver noise. When the
number of users is small, the decorrelator works well because the cost of suppressing other users is
small. When the number of users equals the processing gain, the decorrelator owrks poorly because
the noise is gratly enhanced. By comparison, the LMSE detector applies linear processing that
results in an output that minimizes the mean square error between the output and the original
transmitted bit. It works about as well as the decorrelator when the number of users is small. For
a large number of users, it still works better than the matched filter detector.
350
• In Example 10.3, the daily noontime temperature at Newark Airport is a discrete time,
continuous value random process. However, if the temperature is recorded only in units of
one degree, then the process was would be discrete value.
• In Example 10.4, the number of active telephone calls is discrete time and discrete value.
• The dice rolling experiment of Example 10.5 yields a discrete time, discrete value random
process.
• The QPSK system of Example 10.6 is a continuous time and continuous value random process.

The sample space of the underlying experiment is S = {s0 , s1 , s2 , s3 }. The four elements in the
sample space are equally likely. The ensemble of sample functions is {x(t, si )|i = 0, 1, 2, 3} where
x(t, si ) = cos(2πf0 t + π/4 + iπ/2) (0 ≤ t ≤ T ) (1)
For f0 = 5/T , this ensemble is shown below.

1
0.5
x(t,s0)
0
−0.5
−1
0 0.2T 0.4T 0.6T 0.8T T
1
0.5
x(t,s1)
0
−0.5
−1
0 0.2T 0.4T 0.6T 0.8T T
1
0.5
x(t,s2)
0
−0.5
−1
0 0.2T 0.4T 0.6T 0.8T T
1
0.5
x(t,s3)
0
−0.5
−1
0 0.2T 0.4T 0.6T 0.8T T
t

The eight possible waveforms correspond to the bit sequences
{(0, 0, 0), (1, 0, 0), (1, 1, 0), . . . , (1, 1, 1)} (1)

The corresponding eight waveforms are:
351
1
0
−1
10 T 2T 3T
0
−1
10 T 2T 3T
0
−1
10 T 2T 3T
0
−1
10 T 2T 3T
0
−1
10 T 2T 3T
0
−1
10 T 2T 3T
0
−1
10 T 2T 3T
0
−1
0 T 2T 3T

The statement is false. As a counterexample, consider the rectified cosine waveform X(t) =
R| cos 2πf t| of Example 10.9. When t = π/2, then cos 2πf t = 0 so that X(π/2) = 0. Hence
X(π/2) has PDF
fX(π/2) (x) = δ(x) (1)
That is, X(π/2) is a discrete random variable.

In this problem, we start from first principles. What makes this problem fairly straightforward is
that the ramp is defined for all time. That is, the ramp doesn’t start at time t = W .
P [X(t) ≤ x] = P [t − W ≤ x] = P [W ≥ t − x] (1)
Since W ≥ 0, if x ≥ t then P [W ≥ t − x] = 1. When x < t,

' ∞
P [W ≥ t − x] = fW (w) dw = e−(t−x) (2)
t−x
Combining these facts, we have

e−(t−x) x < t
FX(t) (x) = P [W ≥ t − x] = (3)
1 t≤x
We note that the CDF contain no discontinuities. Taking the derivative of the CDF FX(t) (x) with
respect to x, we obtain the PDF
x−t
e x<t
fX(t) (x) = (4)
0 otherwise
352
(a) Each resistor has frequency W in Hertz with uniform PDF

0.025 9980 ≤ r ≤ 1020
fR (r) = (1)
0 otherwise
The probability that a test yields a one part in 104 oscillator is

' 10001
p = P [9999 ≤ W ≤ 10001] = (0.025) dr = 0.05 (2)
9999
(b) To find the PMF of T1 , we view each oscillator test as an independent trial. A success occurs
on a trial with probability p if we find a one part in 104 oscillator. The first one part in 104
oscillator is found at time T1 = t if we observe failures on trials 1, . . . , t − 1 followed by a
success on trial t. Hence, just as in Example 2.11, T1 has the geometric PMF

(1 − p)t−1 p t = 1, 2, . . .
PT1 (t) = (3)
9 otherwise
A geometric random variable with success probability p has mean 1/p. This is derived in
Theorem 2.5. The expected time to find the first good oscillator is E[T1 ] = 1/p = 20 minutes.
(c) Since p = 0.05, the probability the first one part in 104 oscillator is found in exactly 20
minutes is PT1 (20) = (0.95)19 (0.05) = 0.0189.
(d) The time T5 required to find the 5th one part in 104 oscillator is the number of trials needed
for 5 successes. T5 is a Pascal random variable. If this is not clear, see Example 2.15 where
the Pascal PMF is derived. When we are looking for 5 successes, the Pascal PMF is
t−1 5
4 p (1 − p)
t−5 t = 5, 6, . . .
PT5 (t) = (4)
0 otherwise
Looking up the Pascal PMF in Appendix A, we find that E[T5 ] = 5/p = 100 minutes. The
following argument is a second derivation of the mean of T5 . Once we find the first one part in
104 oscillator, the number of additional trials needed to find the next one part in 104 oscillator
once again has a geometric PMF with mean 1/p since each independent trial is a success with
probability p. Similarly, the time required to find 5 one part in 104 oscillators is the sum of
five independent geometric random variables. That is,
T5 = K1 + K2 + K3 + K4 + K5 (5)
where each Ki is identically distributed to T1 . Since the expectation of the sum equals the
sum of the expectations,
E [T5 ] = E [K1 + K2 + K3 + K4 + K5 ] = 5E [Ki ] = 5/p = 100 minutes (6)
353
Once we find the first one part in 104 oscillator, the number of additional tests needed to find
the next one part in 104 oscillator once again has a geometric PMF with mean 1/p since each
independent trial is a success with probability p. That is T2 = T1 + T where T is independent and
identically distributed to T1 . Thus,

E [T2 |T1 = 3] = E [T1 |T1 = 3] + E T |T1 = 3 (1)

= 3 + E T = 23 minutes. (2)

Since the problem states that the pulse is delayed, we will assume T ≥ 0. This problem is difficult
because the answer will depend on t. In particular, for t < 0, X(t) = 0 and fX(t) (x) = δ(x). Things
are more complicated when t > 0. For x < 0, P [X(t) > x] = 1. For x ≥ 1, P [X(t) > x] = 0.
Lastly, for 0 ≤ x < 1,

P [X(t) > x] = P e−(t−T ) u(t − T ) > x (1)
= P [t + ln x < T ≤ t] (2)
= FT (t) − FT (t + ln x) (3)
Note that condition T ≤ t is needed to make sure that the pulse doesn’t arrive after time t. The
other condition T > t + ln x ensures that the pulse didn’t arrrive too early and already decay too
much. We can express these facts in terms of the CDF of X(t).
⎧
⎨ 0 x<0
FX(t) (x) = 1 − P [X(t) > x] = 1 + FT (t + ln x) − FT (t) 0 ≤ x < 1 (4)
⎩
1 x≥1
We can take the derivative of the CDF to find the PDF. However, we need to keep in mind that
the CDF has a jump discontinuity at x = 0. In particular, since ln 0 = −∞,
FX(t) (0) = 1 + FT (−∞) − FT (t) = 1 − FT (t) (5)
Hence, when we take a derivative, we will see an impulse at x = 0. The PDF of X(t) is

(1 − FT (t))δ(x) + fT (t + ln x) /x 0 ≤ x < 1
fX(t) (x) = (6)
0 otherwise

Each Yk is the sum of two identical independent Gaussian random variables. Hence, each Yk must
have the same PDF. That is, the Yk are identically distributed. Next, we observe that the sequence
of Yk is independent. To see this, we observe that each Yk is composed of two samples of Xk that
are unused by any other Yj for j = k.

Each Wn is the sum of two identical independent Gaussian random variables. Hence, each Wn
must have the same PDF. That is, the Wn are identically distributed. However, since Wn−1 and
354
Wn both use Xn−1 in their averaging, Wn−1 and Wn are dependent. We can verify this observation
by calculating the covariance of Wn−1 and Wn . First, we observe that for all n,
E [Wn ] = (E [Xn ] + E [Xn−1 ])/2 = 30 (1)
Next, we observe that Wn−1 and Wn have covariance
Cov [Wn−1 , Wn ] = E [Wn−1 Wn ] − E [Wn ] E [Wn−1 ] (2)

1
= E [(Xn−1 + Xn−2 )(Xn + Xn−1 )] − 900 (3)
4
We observe that for n = m, E[Xn Xm ] = E[Xn ]E[Xm ] = 900 while

E Xn2 = Var[Xn ] + (E [Xn ])2 = 916 (4)
Thus,
900 + 916 + 900 + 900
Cov [Wn−1 , Wn ] = − 900 = 4 (5)
4
Since Cov[Wn−1 , Wn ] = 0, Wn and Wn−1 must be dependent.

The number Yk of failures between successes k − 1 and k is exactly y ≥ 0 iff after success k − 1,
there are y failures followed by a success. Since the Bernoulli trials are independent, the probability
of this event is (1 − p)y p. The complete PMF of Yk is

(1 − p)y p y = 0, 1, . . .
PYk (y) = (1)
0 otherwise
Since this argument is valid for all k including k = 1, we can conclude that Y1 , Y2 , . . . are identically
distributed. Moreover, since the trials are independent, the failures between successes k − 1 and k
and the number of failures between successes k − 1 and k are independent. Hence, Y1 , Y2 , . . . is an
iid sequence.

This is a very straightforward problem. The Poisson process has rate λ = 4 calls per second. When
t is measured in seconds, each N (t) is a Poisson random variable with mean 4t and thus has PMF
(4t)n −4t
PN (t) (n) = n! e n = 0, 1, 2, . . .
(1)
0 otherwise
Using the general expression for the PMF, we can write down the answer for each part.
(a) PN (1) (0) = 40 e−4 /0! = e−4 ≈ 0.0183.
(b) PN (1) (4) = 44 e−4 /4! = 32e−4 /3 ≈ 0.1954.
(c) PN (2) (2) = 82 e−8 /2! = 32e−8 ≈ 0.0107.
355
Following the instructions given, we express each answer in terms of N (m) which has PMF

(6m)n e−6m /n! n = 0, 1, 2, . . .
PN (m) (n) = (1)
0 otherwise
(a) The probability of no queries in a one minute interval is PN (1) (0) = 60 e−6 /0! = 0.00248.
(b) The probability of exactly 6 queries arriving in a one minute interval is PN (1) (6) = 66 e−6 /6! =
0.161.
(c) The probability of exactly three queries arriving in a one-half minute interval is PN (0.5) (3) =
33 e−3 /3! = 0.224.

Since there is always a backlog an the service times are iid exponential random variables, The time
between service completions are a sequence of iid exponential random variables. that is, the service
completions are a Poisson process. Since the expected service time is 30 minutes, the rate of the
Poisson process is λ = 1/30 per minute. Since t hours equals 60t minutes, the expected number
serviced is λ(60t) or 2t. Moreover, the number serviced in the first t hours has the Poisson PMF

(2t)n e−2t
n = 0, 1, 2, . . .
PN (t) (n) = n! (1)
0 otherwise

Since D(t) is a Poisson process with rate 0.1 drops/day, the random variable D(t) is a Poisson
random variable with parameter α = 0.1t. The PMF of D(t). the number of drops after t days, is

(0.1t)d e−0.1t /d! d = 0, 1, 2, . . .
PD(t) (d) = (1)
0 otherwise

Note that it matters whether t ≥ 2 minutes. If t ≤ 2, then any customers that have arrived must
still be in service. Since a Poisson number of arrivals occur during (0, t],

(λt)n e−λt /n! n = 0, 1, 2, . . .
PN (t) (n) = (0 ≤ t ≤ 2) (1)
0 otherwise
For t ≥ 2, the customers in service are precisely those customers that arrived in the interval (t−2, t].
The number of such customers has a Poisson PMF with mean λ[t − (t − 2)] = 2λ. The resulting
PMF of N (t) is
(2λ)n e−2λ /n! n = 0, 1, 2, . . .
PN (t) (n) = (t ≥ 2) (2)
0 otherwise
356
The time T between queries are independent exponential random variables with PDF

(1/8)e−t/8 t ≥ 0
fT (t) = (1)
0 otherwise
From the PDF, we can calculate for t > 0,

' t
P [T ≥ t] = fT t dt = e−t/8 (2)
0
Using this formula, each question can be easily answered.
(a) P [T ≥ 4] = e−4/8 ≈ 0.951.
(b)
P [T ≥ 13, T ≥ 5]
P [T ≥ 13|T ≥ 5] = (3)
P [T ≥ 5]
P [T ≥ 13] e−13/8
= = −5/8 = e−1 ≈ 0.368 (4)
P [T ≥ 5] e
(c) Although the time betwen queries are independent exponential random variables, N (t) is not
exactly a Poisson random process because the first query occurs at time t = 0. Recall that
in a Poisson process, the first arrival occurs some time after t = 0. However N (t) − 1 is a
Poisson process of rate 8. Hence, for n = 0, 1, 2, . . .,
P [N (t) − 1 = n] = (t/8)n e−t/8 /n! (5)
Thus, for n = 1, 2, . . ., the PMF of N (t) is
PN (t) (n) = P [N (t) − 1 = n − 1] = (t/8)n−1 e−t/8 /(n − 1)! (6)
The complete expression of the PMF of N (t) is

(t/8)n−1 e−t/8 /(n − 1)! n = 1, 2, . . .
PN (t) (n) = (7)
0 otherwise

This proof is just a simplified version of the proof given for Theorem 10.3. The first arrival occurs
at time X1 > x ≥ 0 iff there are no arrivals in the interval (0, x]. Hence, for x ≥ 0,
P [X1 > x] = P [N (x) = 0] = (λx)0 e−λx /0! = e−λx (1)
Since P [X1 ≤ x] = 0 for x < 0, the CDF of X1 is the exponential CDF

0 x<0
FX1 (x) = (2)
1 − e−λx x ≥ 0
357
(a) For Xi = − ln Ui , we can write

P [Xi > x] = P [− ln Ui > x] = P [ln Ui ≤ −x] = P Ui ≤ e−x (1)
When x < 0, e−x > 1 so that P [Ui ≤ e−x ] = 1. When x ≥ 0, we have 0 < e−x ≤ 1, implying
P [Ui ≤ e−x ] = e−x . Combining these facts, we have

1 x<0
P [Xi > x] = (2)
e−x x ≥ 0
This permits us to show that the CDF of Xi is

0 x<0
FXi (x) = 1 − P [Xi > x] = (3)
1 − e−x x > 0
We see that Xi has an exponential CDF with mean 1.
(b) Note that N = n iff

n
n+1
Ui ≥ e−t > Ui (4)
i=1 i=1
By taking the logarithm of both inequalities, we see that N = n iff

n
n+1
ln Ui ≥ −t > ln Ui (5)
i=1 i=1
Next, we multiply through by −1 and recall that Xi = − ln Ui is an exponential random

variable. This yields N = n iff

n
n+1
Xi ≤ t < Xi (6)
i=1 i=1
Now we recall that a Poisson process N (t) of rate 1 has independent

exponential interarrival
times X1 , X2 , . . .. That is, the ith arrival occurs at time ij=1 Xj . Moreover, N (t) = n iff
the first n arrivals occur by time t but arrival n + 1 occurs after time t. Since the random
variable N (t) has a Poisson distribution with mean t, we can write
0 n 1

n+1
tn e−t
P Xi ≤ t < Xi = P [N (t) = n] = . (7)
n!
i=1 i=1

Customers entering (or not entering) the casino is a Bernoulli decomposition of the Poisson process
of arrivals at the casino doors. By Theorem 10.6, customers entering the casino are a Poisson
process of rate 100/2 = 50 customers/hour. Thus in the two hours from 5 to 7 PM, the number, N ,
of customers entering the casino is a Poisson random variable with expected value α = 2 · 50 = 100.
The PMF of N is
100n e−100 /n! n = 0, 1, 2, . . .
PN (n) = (1)
0 otherwise
358
In an interval (t, t + ∆] with an infinitesimal ∆, let Ai denote the event of an arrival of the process
Ni (t). Also, let A = A1 ∪A2 denote the event of an arrival of either process. Since Ni (t) is a Poisson
process, the alternative model says that P [Ai ] = λi ∆. Also, since N1 (t)+N2 (t) is a Poisson process,
the proposed Poisson process model says
P [A] = (λ1 + λ2 )∆ (1)
Lastly, the conditional probability of a type 1 arrival given an arrival of either type is
P [A1 A] P [A1 ] λ1 ∆ λ1
P [A1 |A] = = = = (2)
P [A] P [A] (λ1 + λ2 )∆ λ1 + λ2
This solution is something of a cheat in that we have used the fact that the sum of Poisson processes
is a Poisson process without using the proposed model to derive this fact.

We start with the case when t ≥ 2. When each service time is equally likely to be either 1 minute
or 2 minutes, we have the following situation. Let M1 denote those customers that arrived in the
interval (t − 1, 1]. All M1 of these customers will be in the bank at time t and M1 is a Poisson
random variable with mean λ.
Let M2 denote the number of customers that arrived during (t − 2, t − 1]. Of course, M2 is
Poisson with expected value λ. We can view each of the M2 customers as flipping a coin to determine
whether to choose a 1 minute or a 2 minute service time. Only those customers that chooses a
2 minute service time will be in service at time t. Let M2 denote those customers choosing a 2
minute service time. It should be clear that M2 is a Poisson number of Bernoulli random variables.
Theorem 10.6 verifies that using Bernoulli trials to decide whether the arrivals of a rate λ Poisson
process should be counted yields a Poisson process of rate pλ. A consequence of this result is that a
Poisson number of Bernoulli (success probability p) random variables has Poisson PMF with mean
pλ. In this case, M2 is Poisson with mean λ/2. Moreover, the number of customers in service at
time t is N (t) = M1 + M2 . Since M1 and M2 are independent Poisson random variables, their sum
N (t) also has a Poisson PMF. This was verified in Theorem 6.9. Hence N (t) is Poisson with mean
E[N (t)] = E[M1 ] + E[M2 ] = 3λ/2. The PMF of N (t) is

(3λ/2)n e−3λ/2 /n! n = 0, 1, 2, . . .
PN (t) (n) = (t ≥ 2) (1)
0 otherwise
Now we can consider the special cases arising when t < 2. When 0 ≤ t < 1, every arrival is still in
service. Thus the number in service N (t) equals the number of arrivals and has the PMF

(λt)n e−λt /n! n = 0, 1, 2, . . .
PN (t) (n) = (0 ≤ t ≤ 1) (2)
0 otherwise
When 1 ≤ t < 2, let M1 denote the number of customers in the interval (t − 1, t]. All M1 customers
arriving in that interval will be in service at time t. The M2 customers arriving in the interval
(0, t − 1] must each flip a coin to decide one a 1 minute or two minute service time. Only those
customers choosing the two minute service time will be in service at time t. Since M2 has a Poisson
PMF with mean λ(t − 1), the number M2 of those customers in the system at time t has a Poisson
PMF with mean λ(t − 1)/2. Finally, the number of customers in service at time t has a Poisson
359
PMF with expected value E[N (t)] = E[M1 ] + E[M2 ] = λ + λ(t − 1)/2. Hence, the PMF of N (t)
becomes

(λ(t + 1)/2)n e−λ(t+1)/2 /n! n = 0, 1, 2, . . .
PN (t) (n) = (1 ≤ t ≤ 2) (3)
0 otherwise

Since the arrival times S1 , . . . , Sn are ordered in time and since a Poisson process cannot have two
simultaneous arrivals, the conditional PDF fS1 ,...,Sn |N (S1 , . . . , Sn |n) is nonzero only if s1 < s2 <
· · · < sn < T . In this case, consider an arbitrarily small ∆; in particular, ∆ < mini (si+1 − si )/2
implies that the intervals (si , si + ∆] are non-overlapping. We now find the joint probability
P [s1 < S1 ≤ s1 + ∆, . . . , sn < Sn ≤ sn + ∆, N = n]
that each Si is in the interval (si , si + ∆] and that N = n. This joint event implies that there
were zero arrivals in each interval (si + ∆, si+1 ]. That is, over the interval [0, T ], the Poisson
< has exactly one arrival in each interval (si , si + ∆] and zero arrivals in the time period
process
T − ni=1 (si , si + ∆]. The collection of intervals in which there was no arrival had a total duration
of T − n∆. Note that the probability of exactly one arrival in the interval (si , si + ∆] is λ∆e−λδ
and the probability of zero arrivals in a period of duration T − n∆ is e−λ(Tn −∆) . In addition, the
event of one arrival in each interval (si , si + ∆) and zero events in the period of length T − n∆ are
independent events because they consider non-overlapping periods of the Poisson process. Thus,
n
P [s1 < S1 ≤ s1 + ∆, . . . , sn < Sn ≤ sn + ∆, N = n] = λ∆e−λ∆ e−λ(T −n∆) (1)
= (λ∆)n e−λT (2)
Since P [N = n] = (λT )n e−λT /n!, we see that
P [s1 < S1 ≤ s1 + ∆, . . . , sn < Sn ≤ sn + ∆|N = n]

P [s1 < S1 ≤ s1 + ∆, . . . , sn < Sn ≤ sn + ∆, N = n]
= (3)
P [N = n]
n
(λ∆) e −λT
= (4)
(λT )n e−λT /n!
n!
= n ∆n (5)
T
Finally, for infinitesimal ∆, the conditional PDF of S1 , . . . , Sn given N = n satisfies
fS1 ,...,Sn |N (s1 , . . . , sn |n) ∆n = P [s1 < S1 ≤ s1 + ∆, . . . , sn < Sn ≤ sn + ∆|N = n] (6)

n!
= n ∆n (7)
T
Since the conditional PDF is zero unless s1 < s2 < · · · < sn ≤ T , it follows that

n!/T n 0 ≤ s1 < · · · < sn ≤ T,
fS1 ,...,Sn |N (s1 , . . . , sn |n) = (8)
0 otherwise.
If it seems that the above argument had some “hand-waving,” we now do the derivation of
P [s1 < S1 ≤ s1 + ∆, . . . , sn < Sn ≤ sn + ∆|N = n] in somewhat excruciating detail. (Feel free to
skip the following if you were satisfied with the earlier explanation.)
360
For the interval (s, t], we use the shorthand notation 0(s,t) and 1(s,t) to denote the events of 0
arrivals and 1 arrival respectively. This notation permits us to write
P [s1 < S1 ≤ s1 + ∆, . . . , sn < Sn ≤ sn + ∆, N = n]

= P 0(0,s1 ) 1(s1 ,s1 +∆) 0(s1 +∆,s2 ) 1(s2 ,s2 +∆) 0(s2 +∆,s3 ) · · · 1(sn ,sn +∆) 0(sn +∆,T ) (9)
The set of events 0(0,s1 ) , 0(sn +∆,T ) , and for i = 1, . . . , n−1, 0(si +∆,si+1 ) and 1(si ,si +∆) are independent
because each devent depend on the Poisson process in a time interval that overlaps none of the
other time intervals. In addition, since the Poisson process has rate λ, P [0(s,t) ] = e−λ(t−s) and
P [1(si ,si +∆) ] = (λ∆)e−λ∆ . Thus,
P [s1 < S1 ≤ s1 + ∆, . . . , sn < Sn ≤ sn + ∆, N = n]

= P 0(0,s1 ) P 1(s1 ,s1 +∆) P 0(s1 +∆,s2 ) · · · P 1(sn ,sn +∆) P 0(sn +∆,T ) (10)

= e−λs1 λ∆e−λ∆ e−λ(s2 −s1 −∆) · · · λ∆e−λ∆ e−λ(T −sn −∆) (11)
= (λ∆)n e−λT (12)

From the problem statement, the change in the stock price is X(8)−X(0) and the standard deviation
of X(8)−X(0) is 1/2 point. In other words, the variance of X(8)−X(0) is Var[X(8)−X(0)] = 1/4.
By the definition of Brownian motion. Var[X(8) − X(0)] = 8α. Hence α = 1/32.

We need to verify that Y (t) = X(ct) satisfies the conditions given in Definition 10.10. First
we observe that Y (0) = X(c · 0) = X(0) = 0. Second, we note that since X(t) is Brownian
motion process implies that Y (t) − Y (s) = X(ct) − X(cs) is a Gaussian random variable. Further,
X(ct) − X(cs) is independent of X(t ) for all t ≤ cs. Equivalently, we can say that X(ct) − X(cs)
is independent of X(cτ ) for all τ ≤ s. In other words, Y (t) − Y (s) is independent of Y (τ ) for all
τ ≤ s. Thus Y (t) is a Brownian motion process.

First we observe that Yn = Xn −Xn−1 = X(n)−X(n−1) is a Gaussian random variable with mean
zero and variance α. Since this fact is true for all n, we can conclude that Y1 , Y2 , . . . are identically
distributed. By Definition 10.10 for Brownian motion, Yn = X(n) − X(n − 1) is independent of
X(m) for any m ≤ n − 1. Hence Yn is independent of Ym = X(m) − X(m − 1) for any m ≤ n − 1.
Equivalently, Y1 , Y2 , . . . is a sequence of independent random variables.

Recall that the vector X of increments has independent components Xn = Wn − Wn−1 . Alterna-
tively, each Wn can be written as the sum
W1 = X 1 (1)
W2 = X 1 + X 2 (2)
..
.
Wk = X 1 + X 2 + · · · + X k . (3)
361
In terms of matrices, W = AX where A is the lower triangular matrix
⎡ ⎤
1
⎢1 1 ⎥
⎢ ⎥
A = ⎢. . ⎥. (4)
⎣.. . . ⎦
1 ··· ··· 1
Since E[W] = AE[X] = 0, it folows from Theorem 5.16 that
1
fW (w) = fX A−1 w . (5)
|det (A)|
Since A is a lower triangular matrix, det(A) = 1, the product of its diagonal entries. In addition,
reflecting the fact that each Xn = Wn − Wn−1 ,
⎡ ⎤ ⎡ ⎤
1 W1
⎢−1 1 ⎥ ⎢ W2 − W1 ⎥
⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥
A−1 = ⎢ 0 −1 1 ⎥ and A−1 W = ⎢ W3 − W2 ⎥ . (6)
⎢ .. .. .. .. ⎥ ⎢ .. ⎥
⎣ . . . . ⎦ ⎣ . ⎦
0 · · · 0 −1 1 Wk − Wk−1
2
Combining these facts with the observation that fX (x) = kn=1 fXn (xn ), we can write
k
fW (w) = fX A−1 w = fXn (wn − wn−1 ) , (7)
n=1
which completes the missing steps in the proof of Theorem 10.8.

The discrete time autocovariance function is
CX [m, k] = E [(Xm − µX )(Xm+k − µX )] (1)
2 . For k = 0, X and X
for k = 0, CX [m, 0] = Var[Xm ] = σX m m+k are independent so that
CX [m, k] = E [(Xm − µX )] E [(Xm+k − µX )] = 0 (2)

Thus the autocovariance of Xn is
2
σX k=0
CX [m, k] = (3)
0 k = 0

Recall that X(t) = t − W where E[W ] = 1 and E[W 2 ] = 2.
(a) The mean is µX (t) = E[t − W ] = t − E[W ] = t − 1.
(b) The autocovariance is
CX (t, τ ) = E [X(t)X(t + τ )] − µX (t)µX (t + τ ) (1)
= E [(t − W )(t + τ − W )] − (t − 1)(t + τ − 1) (2)

= t(t + τ ) − E [(2t + τ )W ] + E W 2 − t(t + τ ) + 2t + τ − 1 (3)
= −(2t + τ )E [W ] + 2 + 2t + τ − 1 (4)
=1 (5)
362
In this problem, the daily temperature process results from

2πn
Cn = 16 1 − cos + 4Xn (1)
365
where Xn is an iid random sequence of N [0, 1] random variables. The hardest part of this problem
is distinguishing between the process Cn and the covariance function CC [k].
(a) The expected value of the process is

2πn 2πn
E [Cn ] = 16E 1 − cos + 4E [Xn ] = 16 1 − cos (2)
365 365
(b) The autocovariance of Cn is

2πm 2π(m + k)
CC [m, k] = E Cm − 16 1 − cos Cm+k − 16 1 − cos (3)
365 365

16 k = 0
= 16E [Xm Xm+k ] = (4)
0 otherwise
(c) A model of this type may be able to capture the mean and variance of the daily temperature.
However, one reason this model is overly simple is because day to day temperatures are
uncorrelated. A more realistic model might incorporate the effects of “heat waves” or “cold
spells” through correlated daily temperatures.

By repeated application of the recursion Cn = Cn−1 /2 + 4Xn , we obtain

Cn−2 Xn−1
Cn = +4 + Xn (1)
4 2

Cn−3 Xn−2 Xn−1
= +4 + + Xn (2)
8 4 2
..
. (3)
n
C0 X1 X2 C0 Xi
= n + 4 n−1 + n−2 + · · · + Xn = n + 4 (4)
2 2 2 2 2n−i
i=1
(a) Since C0 , X1 , X2 , . . . all have zero mean,
E [C0 ] E [Xi ]
n
E [Cn ] = + 4 =0 (5)
2n 2n−i
i=1
(b) The autocovariance is

⎡! "⎛ ⎞⎤
C0
n
X i ⎝ C0
m+k
Xj ⎠⎦
CC [m, k] = E ⎣ n + 4 +4 (6)
2 2n−i 2m + k 2m+k−j
i=1 j=1
363
Since C0 , X1 , X2 , . . . are independent (and zero mean), E[C0 Xi ] = 0. This implies

E C02 E [Xi Xj ]
m m+k
CC [m, k] = 2m+k + 16 (7)
2 2m−i 2m+k−j
i=1 j=1
For i = j, E[Xi Xj ] = 0 so that only the i = j terms make any contribution to the double
sum. However, at this point, we must consider the cases k ≥ 0 and k < 0 separately. Since
each Xi has variance 1, the autocovariance for k ≥ 0 is
1
m
1
CC [m, k] = + 16 (8)
22m+k 22m+k−2i
i=1
1 16 m
= + k (1/4)m−i (9)
22m+k 2
i=1
1 16 1 − (1/4)m
= + k (10)
22m+k 2 3/4
For k < 0, we can write

E C02 E [Xi Xj ]
m m+k
CC [m, k] = 2m+k + 16 (11)
2 2m−i 2m+k−j
i=1 j=1
1
m+k
1
= + 16 (12)
22m+k 22m+k−2i
i=1
1 16
m+k
= + −k (1/4)m+k−i (13)
22m+k 2
i=1
1 16 1 − (1/4)m+k
= + k (14)
22m+k 2 3/4
A general expression that’s valid for all m and k is
1 16 1 − (1/4)min(m,m+k)
CC [m, k] = + (15)
22m+k 2|k| 3/4
(c) Since E[Ci ] = 0 for all i, our model has a mean daily temperature of zero degrees Celsius for
the entire year. This is not a reasonable model for a year.
(d) For the month of January, a mean temperature of zero degrees Celsius seems quite reasonable.
we can calculate the variance of Cn by evaluating the covariance at n = m. This yields
1 16 4(4n − 1)
Var[Cn ] = + (16)
4n 4n 3
Note that the variance is upper bounded by
Var[Cn ] ≤ 64/3 (17)

√
Hence the daily temperature has a standard deviation of 8/ 3 ≈ 4.6 degrees. Without actual
evidence of daily temperatures in January, this model is more difficult to discredit.
364
This derivation of the Poisson process covariance is almost identical to the derivation of the Brown-
ian motion autocovariance since both rely on the use of independent increments. From the definition
of the Poisson process, we know that µN (t) = λt. When τ ≥ 0, we can write
CN (t, τ ) = E [N (t)N (t + τ )] − (λt)[λ(t + τ )] (1)

= E [N (t)[(N (t + τ ) − N (t)) + N (t)]] − λ t(t + τ ) 2
(2)

= E [N (t)[N (t + τ ) − N (t)]] + E N 2 (t) − λ2 t(t + τ ) (3)
By the definition of the Poisson process, N (t + τ ) − N (t) is the number of arrivals in the interval
[t, t + τ ) and is independent of N (t) for τ > 0. This implies
E [N (t)[N (t + τ ) − N (t)]] = E [N (t)] E [N (t + τ ) − N (t)] = λt[λ(t + τ ) − λt] (4)
Note that since N (t) is a Poisson random variable, Var[N (t)] = λt. Hence

E N 2 (t) = Var[N (t)] + (E [N (t)]2 = λt + (λt)2 (5)
Therefore, for τ ≥ 0,
CN (t, τ ) = λt[λ(t + τ ) − λt) + λt + (λt)2 − λ2 t(t + τ ) = λt (6)
If τ < 0, then we can interchange the labels t and t+τ in the above steps to show CN (t, τ ) = λ(t+τ ).
For arbitrary t and τ , we can combine these facts to write
CN (t, τ ) = λ min(t, t + τ ) (7)

For an arbitrary set of samples Y (t1 ), . . . , Y (tk ), we observe that Y (tj ) = X(tj + a). This implies
fY (t1 ),...,Y (tk ) (y1 , . . . , yk ) = fX(t1 +a),...,X(tk +a) (y1 , . . . , yk ) (1)
Thus,
fY (t1 +τ ),...,Y (tk +τ ) (y1 , . . . , yk ) = fX(t1 +τ +a),...,X(tk +τ +a) (y1 , . . . , yk ) (2)
Since X(t) is a stationary process,
fX(t1 +τ +a),...,X(tk +τ +a) (y1 , . . . , yk ) = fX(t1 +a),...,X(tk +a) (y1 , . . . , yk ) (3)
This implies
fY (t1 +τ ),...,Y (tk +τ ) (y1 , . . . , yk ) = fX(t1 +a),...,X(tk +a) (y1 , . . . , yk ) (4)

= fY (t1 ),...,Y (tk ) (y1 , . . . , yk ) (5)
We can conclude that Y (t) is a stationary process.
365
For an arbitrary set of samples Y (t1 ), . . . , Y (tk ), we observe that Y (tj ) = X(atj ). This implies
fY (t1 ),...,Y (tk ) (y1 , . . . , yk ) = fX(at1 ),...,X(atk ) (y1 , . . . , yk ) (1)
Thus,
fY (t1 +τ ),...,Y (tk +τ ) (y1 , . . . , yk ) = fX(at1 +aτ ),...,X(atk +aτ ) (y1 , . . . , yk ) (2)
We see that a time offset of τ for the Y (t) process corresponds to an offset of time τ = aτ for the
X(t) process. Since X(t) is a stationary process,
fY (t1 +τ ),...,Y (tk +τ ) (y1 , . . . , yk ) = fX(at1 +τ ),...,X(atk +τ ) (y1 , . . . , yk ) (3)

= fX(at1 ),...,X(atk ) (y1 , . . . , yk ) (4)
= fY (t1 ),...,Y (tk ) (y1 , . . . , yk ) (5)

For a set of time samples n1 , . . . , nm and an offset k, we note that Yni +k = X((ni + k)∆). This
implies
fYn1 +k ,...,Ynm +k (y1 , . . . , ym ) = fX((n1 +k)∆),...,X((nm +k)∆) (y1 , . . . , ym ) (1)
Since X(t) is a stationary process,
fX((n1 +k)∆),...,X((nm +k)∆) (y1 , . . . , ym ) = fX(n1 ∆),...,X(nm ∆) (y1 , . . . , ym ) (2)
Since X(ni ∆) = Yni , we see that
fYn1 +k ,...,Ynm +k (y1 , . . . , ym ) = fYn1 ,...,Ynm (y1 , . . . , ym ) (3)
Hence Yn is a stationary random sequence.

Since Yn = Xkn ,
fYn1 +l ,...,Ynm +l (y1 , . . . , ym ) = fXkn1 +kl ,...,Xknm +kl (y1 , . . . , ym ) (1)
Stationarity of the Xn process implies
fXkn1 +kl ,...,Xknm +kl (y1 , . . . , ym ) = fXkn1 ,...,Xknm (y1 , . . . , ym ) (2)

= fYn1 ,...,Ynm (y1 , . . . , ym ) . (3)
We combine these steps to write
fYn1 +l ,...,Ynm +l (y1 , . . . , ym ) = fYn1 ,...,Ynm (y1 , . . . , ym ) . (4)
Thus Yn is a stationary process.

Comment: The first printing of the text asks whether Yn is wide stationary if Xn is wide sense
stationary. This fact is also true; however, since wide sense stationarity isn’t addressed until the
next section, the problem was corrected to ask about stationarity.
366
Given A = a, Y (t) = aX(t) which is a special case of Y (t) = aX(t) + b given in Theorem 10.10.
Applying the result of Theorem 10.10 with b = 0 yields
1 y yn
1
fY (t1 ),...,Y (tn )|A (y1 , . . . , yn |a) = n fX(t1 ),...,X(tn ) ,..., (1)
a a a
Integrating over the PDF fA (a) yields
' ∞
fY (t1 ),...,Y (tn ) (y1 , . . . , yn ) = fY (t1 ),...,Y (tn )|A (y1 , . . . , yn |a) fA (a) da (2)
'0 ∞ y
1 1 yn
= f , . . . , fA (a) da (3)
0 an X(t1 ),...,X(tn ) a a
This complicated expression can be used to find the joint PDF of Y (t1 + τ ), . . . , Y (tn + τ ):
' ∞ y
1 1 yn
fY (t1 +τ ),...,Y (tn +τ ) (y1 , . . . , yn ) = fX(t1 +τ ),...,X(tn +τ ) , . . . , fA (a) da (4)
0 an a a
Since X(t) is a stationary process, the joint PDF of X(t1 + τ ), . . . , X(tn + τ ) is the same as the
joint PDf of X(t1 ), . . . , X(tn ). Thus
' ∞ y
1 1 yn
fY (t1 +τ ),...,Y (tn +τ ) (y1 , . . . , yn ) = f , . . . , fA (a) da (5)
an X(t1 +τ ),...,X(tn +τ ) a a
'0 ∞ y
1 1 yn
= f X(t 1 ),...,X(t n ) , . . . , fA (a) da (6)
0 an a a
= fY (t1 ),...,Y (tn ) (y1 , . . . , yn ) (7)

Since g(·) is an unspecified function, we will work with the joint CDF of Y (t1 + τ ), . . . , Y (tn + τ ).
To show Y (t) is a stationary process, we will show that for all τ ,
FY (t1 +τ ),...,Y (tn +τ ) (y1 , . . . , yn ) = FY (t1 ),...,Y (tn ) (y1 , . . . , yn ) (1)
By taking partial derivatives with respect to y1 , . . . , yn , it should be apparent that this implies that
the joint PDF fY (t1 +τ ),...,Y (tn +τ ) (y1 , . . . , yn ) will not depend on τ . To proceed, we write
FY (t1 +τ ),...,Y (tn +τ ) (y1 , . . . , yn ) = P [Y (t1 + τ ) ≤ y1 , . . . , Y (tn + τ ) ≤ yn ] (2)
⎡ ⎤
⎢ ⎥
= P ⎣g(X(t1 + τ )) ≤ y1 , . . . , g(X(tn + τ )) ≤ yn ⎦ (3)

Aτ
In principle, we can calculate P [Aτ ] by integrating fX(t1 +τ ),...,X(tn +τ ) (x1 , . . . , xn ) over the region
corresponding to event Aτ . Since X(t) is a stationary process,
fX(t1 +τ ),...,X(tn +τ ) (x1 , . . . , xn ) = fX(t1 ),...,X(tn ) (x1 , . . . , xn ) (4)
This implies P [Aτ ] does not depend on τ . In particular,
FY (t1 +τ ),...,Y (tn +τ ) (y1 , . . . , yn ) = P [Aτ ] (5)
= P [g(X(t1 )) ≤ y1 , . . . , g(X(tn )) ≤ yn ] (6)
= FY (t1 ),...,Y (tn ) (y1 , . . . , yn ) (7)
367
The autocorrelation function RX (τ ) = δ(τ ) is mathematically valid in the sense that it meets the
conditions required in Theorem 10.12. That is,
RX (τ ) = δ(τ ) ≥ 0 (1)
RX (τ ) = δ(τ ) = δ(−τ ) = RX (−τ ) (2)
RX (τ ) ≤ RX (0) = δ(0) (3)
However, for a process X(t) with the autocorrelation RX (τ ) = δ(τ ), Definition 10.16 says that the
average power of the process is

E X 2 (t) = RX (0) = δ(0) = ∞ (4)
Processes with infinite average power cannot exist in practice.

Since Y (t) = A + X(t), the mean of Y (t) is
E [Y (t)] = E [A] + E [X(t)] = E [A] + µX (1)
The autocorrelation of Y (t) is
RY (t, τ ) = E [(A + X(t)) (A + X(t + τ ))] (2)

= E A2 + E [A] E [X(t)] + AE [X(t + τ )] + E [X(t)X(t + τ )] (3)

= E A2 + 2E [A] µX + RX (τ ) (4)
We see that neither E[Y (t)] nor RY (t, τ ) depend on t. Thus Y (t) is a wide sense stationary process.

In this problem, we find the autocorrelation RW (t, τ ) when
W (t) = X cos 2πf0 t + Y sin 2πf0 t, (1)
and X and Y are uncorrelated random variables with E[X] = E[Y ] = 0.

We start by writing
RW (t, τ ) = E [W (t)W (t + τ )] (2)

= E [(X cos 2πf0 t + Y sin 2πf0 t) (X cos 2πf0 (t + τ ) + Y sin 2πf0 (t + τ ))] . (3)
Since X and Y are uncorrelated, E[XY ] = E[X]E[Y ] = 0. Thus, when we expand E[W (t)W (t + τ )]
and take the expectation, all of the XY cross terms will be zero. This implies

RW (t, τ ) = E X 2 cos 2πf0 t cos 2πf0 (t + τ ) + E Y 2 sin 2πf0 t sin 2πf0 (t + τ ) (4)
Since E[X] = E[Y ] = 0,

E X 2 = Var[X] − (E [X])2 = σ 2 , E Y 2 = Var[Y ] − (E [Y ])2 = σ 2 . (5)
In addition, from Math Fact B.2, we use the formulas

1
cos A cos B = cos(A − B) + cos(A + B) (6)
2
1
sin A sin B = cos(A − B) − cos(A + B) (7)
2
368
to write
σ2 σ2
RW (t, τ ) = (cos 2πf0 τ + cos 2πf0 (2t + τ )) + (cos 2πf0 τ − cos 2πf0 (2t + τ )) (8)
2 2
= σ 2 cos 2πf0 τ (9)
Thus RW (t, τ ) = RW (τ ). Since
E [W (t)] = E [X] cos 2πf0 t + E [Y ] sin 2πf0 t = 0, (10)
we can conclude that W (t) is a wide sense stationary process. However, we note that if E[X 2 ] =
E[Y 2 ], then the cos 2πf0 (2t + τ ) terms in RW (t, τ ) would not cancel and W (t) would not be wide
sense stationary.
(a) In the problem statement, we are told that X(t) has average power equal to 1. By Defini-
tion 10.16, the average power of X(t) is E[X 2 (t)] = 1.
(b) Since Θ has a uniform PDF over [0, 2π],

1/(2π) 0 ≤ θ ≤ 2π
fΘ (θ) = (1)
0 otherwise
The expected value of the random phase cosine is
' ∞
E [cos(2πfc t + Θ)] = cos(2πfc t + θ)fΘ (θ) dθ (2)
−∞
' 2π
1
= cos(2πfc t + θ) dθ (3)
0 2π
1
= sin(2πfc t + θ)|2π
0 (4)
2π
1
= (sin(2πfc t + 2π) − sin(2πfc t)) = 0 (5)
2π
(c) Since X(t) and Θ are independent,
E [Y (t)] = E [X(t) cos(2πfc t + Θ)] = E [X(t)] E [cos(2πfc t + Θ)] = 0 (6)
Note that the mean of Y (t) is zero no matter what the mean of X(t) since the random phase
cosine has zero mean.
(d) Independence of X(t) and Θ results in the average power of Y (t) being

E Y 2 (t) = E X 2 (t) cos2 (2πfc t + Θ) (7)

= E X 2 (t) E cos2 (2πfc t + Θ) (8)

= E cos2 (2πfc t + Θ) (9)
Note that we have used the fact from part (a) that X(t) has unity average power. To finish
the problem, we use the trigonometric identity cos2 φ = (1 + cos 2φ)/2. This yields

2 1
E Y (t) = E (1 + cos(2π(2fc )t + Θ)) = 1/2 (10)
2
Note that E[cos(2π(2fc )t + Θ)] = 0 by the argument given in part (b) with 2fc replacing fc .
369
This proof simply parallels the proof of Theorem 10.12. For the first item, RX [0] = RX [m, 0] =
E[Xm 2 ]. Since X 2 ≥ 0, we must have E[X 2 ] ≥ 0. For the second item, Definition 10.13 implies
m m
that
RX [k] = RX [m, k] = E [Xm Xm+k ] = E [Xm+k Xm ] = RX [m + k, −k] (1)
Since Xm is wide sense stationary, RX [m + k, −k] = RX [−k]. The final item requires more effort.
First, we note that when Xm is wide sense stationary, Var[Xm ] = CX [0], a constant for all t.
Second, Theorem 4.17 says that
|CX [m, k]| ≤ σXm σXm+k = CX [0] . (2)
Note that CX [m, k] ≤ |CX [m, k]|, and thus it follows that
CX [m, k] ≤ σXm σXm+k = CX [0] , (3)
(This little step was unfortunately omitted from the proof of Theorem 10.12.) Now for any numbers
a, b, and c, if a ≤ b and c ≥ 0, then (a + c)2 ≤ (b + c)2 . Choosing a = CX [m, k], b = CX [0], and
c = µ2X yields
2 2
CX [m, m + k] + µ2X ≤ CX [0] + µ2X (4)
In the above expression, the left side equals (RX [k])2 while the right side is (RX [0])2 , which proves
the third part of the theorem.

The solution to this problem is essentially the same as the proof of Theorem 10.13 except integrals
are replaced by sums. First we verify that X m is unbiased:
0 m 1
1
E Xm = E Xn (1)
2m + 1 n=−m
1 m
1 m
= E [Xn ] = µX = µX (2)
2m + 1 n=−m 2m + 1 n=−m
To show consistency,

mit is sufficient to show that limm→∞ Var[X m ] = 0. First, we observe that
X m − µX = 2m+1
1
n=−m (Xn − µX ). This implies
⎡! "2 ⎤
1 m
Var[X(T )] = E ⎣ (Xn − µX ) ⎦ (3)
2m + 1 n=−m
0 ! m "! m "1
1
=E (Xn − µX ) (Xn − µX ) (4)
(2m + 1)2 n=−m n =−m
1
m
m
= E [(Xn − µX )(Xn − µX )] (5)
(2m + 1)2 n=−m n =−m
1 m
m

= CX n − n . (6)
(2m + 1)2 n=−m n =−m
370
We note that

m

m
$ $
CX n − n ≤ $CX n − n $ (7)
n =−m n =−m
∞ ∞

$ $
≤ $CX n − n $ = |CX (k)| < ∞. (8)
n =−∞ k=−∞
Hence there exists a constant K such that
1 m
K
Var[X m ] ≤ K= . (9)
(2m + 1)2 n=−m 2m + 1
Thus limm→∞ Var[X m ] ≤ limm→∞ K

2m+1 = 0.
(a) Since X(t) and Y (t) are independent processes,
E [W (t)] = E [X(t)Y (t)] = E [X(t)] E [Y (t)] = µX µY . (1)
In addition,
RW (t, τ ) = E [W (t)W (t + τ )] (2)

= E [X(t)Y (t)X(t + τ )Y (t + τ )] (3)
= E [X(t)X(t + τ )] E [Y (t)Y (t + τ )] (4)
= RX (τ )RY (τ ) (5)
We can conclude that W (t) is wide sense stationary.
(b) To examine whether X(t) and W (t) are jointly wide sense stationary, we calculate
RW X (t, τ ) = E [W (t)X(t + τ )] = E [X(t)Y (t)X(t + τ )] . (6)
By independence of X(t) and Y (t),
RW X (t, τ ) = E [X(t)X(t + τ )] E [Y (t)] = µY RX (τ ). (7)
Since W (t) and X(t) are both wide sense stationary and since RW X (t, τ ) depends only on
the time difference τ , we can conclude from Definition 10.18 that W (t) and X(t) are jointly
wide sense stationary.

To show that X(t) and Xi (t) are jointly wide sense stationary, we must first show that Xi (t) is wide
sense stationary and then we must show that the cross correlation RXXi (t, τ ) is only a function of
the time difference τ . For each Xi (t), we have to check whether these facts are implied by the fact
that X(t) is wide sense stationary.
371
(a) Since E[X1 (t)] = E[X(t + a)] = µX and
RX1 (t, τ ) = E [X1 (t)X1 (t + τ )] (1)
= E [X(t + a)X(t + τ + a)] (2)
= RX (τ ), (3)
we have verified that X1 (t) is wide sense stationary. Now we calculate the cross correlation
RXX1 (t, τ ) = E [X(t)X1 (t + τ )] (4)
= E [X(t)X(t + τ + a)] (5)
= RX (τ + a). (6)
Since RXX1 (t, τ ) depends on the time difference τ but not on the absolute time t, we conclude
that X(t) and X1 (t) are jointly wide sense stationary.
(b) Since E[X2 (t)] = E[X(at)] = µX and
RX2 (t, τ ) = E [X2 (t)X2 (t + τ )] (7)
= E [X(at)X(a(t + τ ))] (8)
= E [X(at)X(at + aτ )] = RX (aτ ), (9)
we have verified that X2 (t) is wide sense stationary. Now we calculate the cross correlation
RXX2 (t, τ ) = E [X(t)X2 (t + τ )] (10)
= E [X(t)X(a(t + τ ))] (11)
= RX ((a − 1)t + τ ). (12)
Except for the trivial case when a = 1 and X2 (t) = X(t), RXX2 (t, τ ) depends on both the
absolute time t and the time difference τ , we conclude that X(t) and X2 (t) are not jointly
wide sense stationary.
(a) Y (t) has autocorrelation function

RY (t, τ ) = E [Y (t)Y (t + τ )] (1)
= E [X(t − t0 )X(t + τ − t0 )] (2)
= RX (τ ). (3)
(b) The cross correlation of X(t) and Y (t) is

RXY (t, τ ) = E [X(t)Y (t + τ )] (4)
= E [X(t)X(t + τ − t0 )] (5)
= RX (τ − t0 ). (6)
(c) We have already verified that RY (t, τ ) depends only on the time difference τ . Since E[Y (t)] =
E[X(t − t0 )] = µX , we have verified that Y (t) is wide sense stationary.
(d) Since X(t) and Y (t) are wide sense stationary and since we have shown that RXY (t, τ )
depends only on τ , we know that X(t) and Y (t) are jointly wide sense stationary.
Comment: This problem is badly designed since the conclusions don’t depend on the specific
RX (τ ) given in the problem text. (Sorry about that!)
372
Problem 10.12.1(Solution
t+τ
Writing Y (t + τ ) = 0 N (v) dv permits us to write the autocorrelation of Y (t) as
' t ' t+τ
RY (t, τ ) = E [Y (t)Y (t + τ )] = E N (u)N (v) dv du (1)
0 0
' t ' t+τ
= E [N (u)N (v)] dv du (2)
0 0
' t ' t+τ
= αδ(u − v) dv du. (3)
0 0
At this point, it matters whether τ ≥ 0 or if τ < 0. When τ ≥ 0, then v ranges from 0 to t + τ and
at some point in the integral over v we will have v = u. That is, when τ ≥ 0,
' t
RY (t, τ ) = α du = αt. (4)
0
When τ < 0, then we must reverse the order of integration. In this case, when the inner integral
is over u, we will have u = v at some point so that
' t+τ ' t ' t+τ
RY (t, τ ) = αδ(u − v) du dv = α dv = α(t + τ ). (5)
0 0 0
Thus we see the autocorrelation of the output is
RY (t, τ ) = α min {t, t + τ } (6)
Perhaps surprisingly, RY (t, τ ) is what we found in Example 10.19 to be the autocorrelation of a

Brownian motion process. In fact, Brownian motion is the integral of the white noise process.

Let µi = E[X(ti )].
(a) Since CX (t1 , t2 − t1 ) = ρσ1 σ2 , the covariance matrix is
2
CX (t1 , 0) CX (t1 , t2 − t1 ) σ1 ρσ1 σ2
C= = (1)
CX (t2 , t1 − t2 ) CX (t2 , 0) ρσ1 σ2 σ22
Since C is a 2 × 2 matrix, it has determinant |C| = σ12 σ22 (1 − ρ2 ).
(b) Is is easy to verify that

⎡ 1 −ρ ⎤
1 ⎢ σ12 σ1 σ2 ⎥
C−1 = ⎣ 1 ⎦ (2)
1 − ρ2 −ρ
σ1 σ2 σ12
(c) The general form of the multivariate density for X(t1 ), X(t2 ) is
1 1 −1 (x−µ )
fX(t1 ),X(t2 ) (x1 , x2 ) = e− 2 (x−µX ) C X (3)
(2π)k/2 |C|1/2

where k = 2 and x = x1 x2 and µX = µ1 µ2 . Hence,
1 1
= . (4)
(2π)k/2 |C| 1/2
2π σ12 σ22 (1 − ρ2 )
373
Furthermore, the exponent is
1
− (x̄ − µ̄X ) C−1 (x̄ − µ̄X )
2
⎡ 1 −ρ ⎤

1 1 ⎢ σ2 σ1 σ2 ⎥ x1 − µ1
= − x1 − µ1 x2 − µ2 ⎣ 1 1 ⎦ x2 − µ2 (5)
2 1 − ρ2 −ρ
σ12
σ1 σ2
2
x1 − µ1 2ρ(x1 − µ1 )(x2 − µ2 ) x2 − µ2 2
− +
σ1 σ1 σ2 σ2
=− (6)
2(1 − ρ2 )
Plugging in each piece into the joint PDF fX(t1 ),X(t2 ) (x1 , x2 ) given above, we obtain the
bivariate Gaussian PDF.

Let W = W (t1 ) W (t2 ) · · · W (tn ) denote a vector of samples of a Brownian motion process.
To prove that W (t) is a Gaussian random process, we must show that W is a Gaussian random
vector. To do so, let

X = X1 · · · X n (1)

= W (t1 ) W (t2 ) − W (t1 ) W (t3 ) − W (t2 ) · · · W (tn ) − W (tn−1 ) (2)
denote the vector of increments. By the definition of Brownian motion, X1 , . . . , Xn is a sequence
of independent Gaussian random variables. Thus X is a Gaussian random vector. Finally,
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
W1 X1 1
⎢ W2 ⎥ ⎢ X1 + X2 ⎥ ⎢1 1 ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
W=⎢ . ⎥=⎢ .. ⎥ ⎢ ..
= .. ⎥ X. (3)
⎣ .. ⎦ ⎣ . ⎦ ⎣ . . ⎦
Wn X1 + · · · + Xn 1 ··· ··· 1

A
Since X is a Gaussian random vector and W = AX with A a rank n matrix, Theorem 5.16 implies
that W is a Gaussian random vector.

From the instructions given in the problem, the program noisycosine.m will generate the four
plots.
n=1000; t=0.001*(-n:n);
w=gaussrv(0,0.01,(2*n)+1);
%Continuous Time, Continuous Value
xcc=2*cos(2*pi*t) + w’;
plot(t,xcc);
xlabel(’\it t’);ylabel(’\it X_{cc}(t)’);
axis([-1 1 -3 3]);
figure; %Continuous Time, Discrete Value
xcd=round(xcc); plot(t,xcd);
xlabel(’\it t’);ylabel(’\it X_{cd}(t)’);
axis([-1 1 -3 3]);
figure; %Discrete time, Continuous Value
ts=subsample(t,100); xdc=subsample(xcc,100);
plot(ts,xdc,’b.’);
xlabel(’\it t’);ylabel(’\it X_{dc}(t)’);
axis([-1 1 -3 3]);
figure; %Discrete Time, Discrete Value
xdd=subsample(xcd,100); plot(ts,xdd,’b.’);
xlabel(’\it t’);ylabel(’\it X_{dd}(t)’);
axis([-1 1 -3 3]);
374
In noisycosine.m, we use a function subsample.m to obtain the discrete time sample functions.
In fact, subsample is hardly necessary since it’s such a simple one-line Matlab function:
function y=subsample(x,n)
%input x(1), x(2) ...
%output y(1)=x(1), y(2)=x(1+n), y(3)=x(2n+1)
y=x(1:n:length(x));
However, we use it just to make noisycosine.m a little more clear.
>> t=(1:600)’; These commands will simulate the switch for 600 minutes, pro-
>> M=simswitch(10,0.1,t); ducing the vector M of samples of M (t) each minute, the vector
>> Mavg=cumsum(M)./t; Mavg which is the sequence of time average estimates, and a plot
>> plot(t,M,t,Mavg); resembling this one:
150
100
M(t)
50
M(t)
Time Average M(t)
0
0 100 200 300 400 500 600
t
From the figure, it appears that the time average is converging to a value in th neighborhood of 100.
In particular, because the switch is initially empty with M (0) = 0, it takes a few hundred minutes
for the time average to climb to something close to 100. Following the problem instructions, we
can write the following short program to examine ten simulation runs:
function Mavg=simswitchavg(T,k)
%Usage: Mavg=simswitchavg(T,k)
%simulate k runs of duration T of the
%telephone switch in Chapter 10
%and plot the time average of each run
t=(1:k)’;
%each column of Mavg is a time average sample run
Mavg=zeros(T,k);
for n=1:k,
M=simswitch(10,0.1,t);
Mavg(:,n)=cumsum(M)./t;
end
plot(t,Mavg);
The command simswitchavg(600,10) produced this graph:
375
120
100
M(t) Time Average

80
60
40
20
0
0 100 200 300 400 500 600
t
From the graph, one can see that even after T = 600 minutes, each sample run produces a time
average M 600 around 100. Note that in Chapter 12, we will able Markov chains to prove that the
expected number of calls in the switch is in fact 100. However, note that even if T is large, M T is
still a random variable. From the above plot, one might guess that M 600 has a standard deviation
of perhaps σ = 2 or σ = 3. An exact calculation of the variance of M 600 is fairly difficult because
it is a sum of dependent random variables, each of which has a PDF that is in itself reasonably
difficult to calculate.

In this problem, our goal is to find out the average number of ongoing calls in the switch. Before
we use the approach of Problem 10.13.2, its worth a moment to consider the physical situation. In
particular, calls arrive as a Poisson process of rate λ = 100 call/minute and each call has duration
of exactly one minute. As a result, if we inspect the system at an arbitrary time t at least one
minute past initialization, the number of calls at the switch will be exactly the number of calls
N1 that arrived in the previous minute. Since calls arrive as a Poisson proces of rate λ = 100
calls/minute. N1 is a Poisson random variable with E[N1 ] = 100.
In fact, this should be true for every inspection time t. Hence it should surprising if we compute
the time average and find the time average number in the queue to be something other than
100. To check out this quickie analysis, we use the method of Problem 10.13.2. However, unlike
Problem 10.13.2, we cannot directly use the function simswitch.m because the call duration are no
longer exponential random variables. Instead, we must modify simswitch.m for the deterministic
one minute call durations, yielding the function simswitchd.m:
function M=simswitchd(lambda,T,t) Note that if you compare simswitch.m in the text
%Poisson arrivals, rate lambda with simswitchd.m here, two changes occurred. The
%Deterministic (T) call duration first is that the exponential call durations are replaced
%For vector t of times by the deterministic time T . The other change is
%M(i) = no. of calls at time t(i)
that count(s,t) is replaced by countup(s,t). In
s=poissonarrivals(lambda,max(t));
y=s+T; fact, n=countup(x,y) does exactly the same thing as
A=countup(s,t); n=count(x,y); in both cases, n(i) is the number of el-
D=countup(y,t); ements less than or equal to y(i). The difference is that
M=A-D; countup requires that the vectors x and y be nondecreas-
ing.
Now we use the same procedure as in Problem 10.13.2 and form the time average
1
T
M (T ) = M (t). (1)
T
t=1
376
>> t=(1:600)’; We form and plot the time average using these commands will
>> M=simswitchd(100,1,t); yield a plot vaguely similar to that shown below.
>> Mavg=cumsum(M)./t;
>> plot(t,Mavg);
105
M(t) time average
100
95
0 100 200 300 400 500 600
t
We used the word “vaguely” because at t = 1, the time average is simply the number of arrivals in
the first minute, which is a Poisson (α = 100) random variable which has not been averaged. Thus,
the left side of the graph will be random for each run. As expected, the time average appears to
be converging to 100.

The random variable Sn is the sum of n exponential (λ) random variables. That is, Sn is an Erlang
(n, λ) random variable. Since K = 1 if and only if Sn > T , P [K = 1] = P [Sn > T ]. Typically,
P [K = 1] is fairly high because
n 1.1λT
E [Sn ] = = ≈ 1.1T. (1)
λ λ
Increasing n increases P [K = 1]; however, poissonarrivals then does more work generating expo-
nential random variables. Although we don’t want to generate more exponential random variables
than necessary, if we need to generate a lot of arrivals (ie a lot of exponential interarrival times),
then Matlab is typically faster generating a vector of them all at once rather than generating
them one at a time. Choosing n = 1.1λT generates about 10 percent more exponential random
variables than we typically need. However, as long as P [K = 1] is high, a ten percent penalty won’t
be too costly.
When n is small, it doesn’t much matter if we are efficient because the amount of calculation is
small. The question that must be addressed is to estimate P [K = 1] when n is large. In this case,
we can use the central limit theorem because Sn is the sum of n exponential random variables.
Since E[Sn ] = n/λ and Var[Sn ] = n/λ2 ,
0 1
Sn − n/λ T − n/λ λT − n
P [Sn > T ] = P > ≈Q √ (2)
n/λ2 n/λ2 n
To simplify our algebra, we assume for large n that 0.1λT is an integer. In this case, n = 1.1λT
and
!) "
0.1λT λT
P [Sn > T ] ≈ Q − √ =Φ (3)
1.1λT 110
Thus for large λT , P [K = 1] is very small. For example, if λT = 1,000, P [Sn > T ] ≈ Φ(3.01) =
0.9987. If λT = 10,000, P [Sn > T ] ≈ Φ(9.5).
377
Following the problem instructions, we can write the function newarrivals.m. For convenience,
here are newarrivals and poissonarrivals side by side.
function s=newarrivals(lam,T) function s=poissonarrivals(lam,T)
%Usage s=newarrivals(lam,T) %arrival times s=[s(1) ... s(n)]
%Returns Poisson arrival times % s(n)<= T < s(n+1)
%s=[s(1) ... s(n)] over [0,T] n=ceil(1.1*lam*T);
n=poissonrv(lam*T,1); s=cumsum(exponentialrv(lam,n));
s=sort(T*rand(n,1)); while (s(length(s))< T),
s_new=s(length(s))+ ...
cumsum(exponentialrv(lam,n));
s=[s; s_new];
end
s=s(s<=T);
Clearly the code for newarrivals is shorter, more readable, and perhaps, with the help of
Problem 10.6.4, more logical than poissonarrivals. Unfortunately this doesn’t mean the code
runs better. Here are some cputime comparisons:
>> t=cputime;s=poissonarrivals(1,100000);t=cputime-t
t =
0.1110
>> t=cputime;s=newarrivals(1,100000);t=cputime-t
t =
0.5310
>> t=cputime;poissonrv(100000,1);t=cputime-t
t =
0.5200
>>
Unfortunately, these results were highly repeatable. The function poissonarrivals generated
100,000 arrivals of a rate 1 Poisson process required roughly 0.1 seconds of cpu time. The same
task took newarrivals about 0.5 seconds, or roughly 5 times as long! In the newarrivals code,
the culprit is the way poissonrv generates a single Poisson random variable with expected value
100,000. In this case, poissonrv generates the first 200,000 terms of the Poisson PMF! This
required calculation is so large that it dominates the work need to generate 100,000 uniform random
numbers. In fact, this suggests that a more efficient way to generate a Poisson (α) random variable
N is to generate arrivals of a rate α Poisson process until the N th arrival is after time 1.

We start with brownian.m to simulate the Brownian motion process with barriers, Since the goal
is to estimate the barrier probability P [|X(t)| = b], we don’t keep track of the value of the process
over all time. Also, we simply assume that a unit time step τ = 1 for the process. Thus, the process
starts at n = 0 at position W0 = 0 at each step n, the position, if we haven’t reached a barrier, is
√
Wn = Wn−1 + Xn , where X1 , . . . , XT are iid Gaussian (0, α) random variables. Accounting for
the effect of barriers,
Wn = max(min(Wn−1 + Xn , b), −b). (1)
To implement the simulation, we can generate the vector x of increments all at once. However to
check at each time step whether we are crossing a barrier, we need to proceed sequentially. (This
is analogous to the problem in Quiz 10.13.)
378
In brownbarrier shown below, pb(1) tracks how often the process touches the left barrier at
−b while pb(2) tracks how often the right side barrier at b is reached. By symmetry, P [X(t) = b] =
P [X(t) = −b]. Thus if T is chosen very large, we should expect pb(1)=pb(2). The extent to which
this is not the case gives an indication of the extent to which we are merely estimating the barrier
probability. Here is the code and for each T ∈ {10,000, 100,000, 1,000,000}, here two sample runs:
function pb=brownwall(alpha,b,T) >> pb=brownwall(0.01,1,1e4)

%pb=brownwall(alpha,b,T) pb =
%Brownian motion, param. alpha 0.0301 0.0353
%walls at [-b, b], sampled >> pb=brownwall(0.01,1,1e4)
%unit of time until time T pb =
%each Returns vector pb: 0.0417 0.0299
%pb(1)=fraction of time at -b >> pb=brownwall(0.01,1,1e5)
%pb(2)=fraction of time at b pb =
T=ceil(T); 0.0333 0.0360
x=sqrt(alpha).*gaussrv(0,1,T); >> pb=brownwall(0.01,1,1e5)
w=0;pb=zeros(1,2); pb =
for k=1:T, 0.0341 0.0305
w=w+x(k); >> pb=brownwall(0.01,1,1e6)
if (w <= -b) pb =
w=-b; 0.0323 0.0342
pb(1)=pb(1)+1; >> pb=brownwall(0.01,1,1e6)
elseif (w >= b) pb =
w=b; 0.0333 0.0324
pb(2)=pb(2)+1; >>
end
end
pb=pb/T;
The sample runs show that for α = 0.1 and b = 1 that the
P [X(t) = −b] ≈ P [X(t) = b] ≈ 0.03. (2)
Otherwise, the numerical simulations are not particularly instructive. Perhaps the most important
thing to understand is that the Brownian motion process with barriers is very different from the
ordinary Brownian motion process. Remember that for ordinary Brownian motion, the variance
of X(t) always increases linearly with t. For the process with barriers, X 2 (t) ≤ b2 and thus
Var[X(t)] ≤ b2 . In fact, for the process with barriers, the PDF of X(t) converges to a limit as t
becomes large. If you’re curious, you shouldn’t have much trouble digging in the library to find out
more.

In this problem, we start with the simswitch.m code to generate the vector of departure times y.
We then construct the vector I of inter-departure times. The command hist,20 will generate a
20 bin histogram of the departure times. The fact that this histogram resembles an exponential
PDF suggests that perhaps it is reasonable to try to match the PDF of an exponential (µ) random
variable against the histogram.
In most problems in which one wants to fit a PDF to measured data, a key issue is how to
choose the parameters of the PDF. In this problem, choosing µ is simple. Recall that the switch
has a Poisson arrival process of rate λ so interarrival times are exponential (λ) random variables.
If 1/µ < 1/λ, then the average time between departures from the switch is less than the average
379
time between arrivals to the switch. In this case, calls depart the switch faster than they arrive
which is impossible because each departing call was an arriving call at an earlier time. Similarly,
if 1/µ > 1/λ , then calls would be departing from the switch more slowly than they arrived. This
can happen to an overloaded switch; however, it’s impossible in this system because each arrival
departs after an exponential time. Thus the only possibility is that 1/µ = 1/λ. In the program
simswitchdepart.m, we plot a histogram of departure times for a switch with arrival rate λ against
the scaled exponential (λ) PDF λe−λx b where b is the histogram bin size. Here is the code:
function I=simswitchdepart(lambda,mu,T)
%Usage: I=simswitchdepart(lambda,mu,T)
%Poisson arrivals, rate lambda
%Exponential (mu) call duration
%Over time [0,T], returns I,
%the vector of inter-departure times
%M(i) = no. of calls at time t(i)
s=poissonarrivals(lambda,T);
y=s+exponentialrv(mu,length(s));
y=sort(y);
n=length(y);
I=y-[0; y(1:n-1)]; %interdeparture times
imax=max(I);b=ceil(n/100);
id=imax/b; x=id/2:id:imax;
pd=hist(I,x); pd=pd/sum(pd);
px=exponentialpdf(lambda,x)*id;
plot(x,px,x,pd);
xlabel(’\it x’);ylabel(’Probability’);
legend(’Exponential PDF’,’Relative Frequency’);
Here is an example of the output corresponding to simswitchdepart(10,1,1000).
0.1
Exponential PDF
0.08 Relative Frequency
Probability
0.06
0.04
0.02
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
As seen in the figure, the match is quite good. Although this is not a carefully designed statistical
test of whether the inter-departure times are exponential random variables, it is enough evidence
that one may want to pursue whether such a result can be proven.
In fact, the switch in this problem is an example of an M/M/∞ queuing system for which
it has been shown that not only do the inter-departure have an exponential distribution, but the
steady-state departure process is a Poisson process. For the curious reader, details can be found,
for example, in the text Discrete Stochastic Processes by Gallager.
380

For this problem, it is easiest to work with the expectation operator. The mean function of the
output is
E [Y (t)] = 2 + E [X(t)] = 2 (1)
The autocorrelation of the output is
RY (t, τ ) = E [(2 + X(t)) (2 + X(t + τ ))] (2)

= E [4 + 2X(t) + 2X(t + τ ) + X(t)X(t + τ )] (3)
= 4 + 2E [X(t)] + 2E [X(t + τ )] + E [X(t)X(t + τ )] (4)
= 4 + RX (τ ) (5)
We see that RY (t, τ ) only depends on the time difference τ . Thus Y (t) is wide sense stationary.

By Theorem 11.2, the mean of the output is
' ∞
µY = µX h(t) dt (1)
−∞
' 10−3
= −3 (1 − 106 t2 ) dt (2)
0
$10−3
= −3 t − (106 /3)t3 $0 (3)
= −2 × 10−3 volts (4)

By Theorem 11.2, the mean of the output is
' ∞ ' ∞ $∞
$
µY = µX h(t) dt = 4 e−t/a dt = −4ae−t/a $ = 4a. (1)
−∞ 0 0
Since µY = 1 = 4a, we must have a = 1/4.

Since E[Y 2 (t)] = RY (0), we use Theorem 11.2(a) to evaluate RY (τ ) at τ = 0. That is,
' ∞ ' ∞
RY (0) = h(u) h(v)RX (u − v) dv du (1)
'−∞∞ ' −∞
∞
= h(u) h(v)η0 δ(u − v) dv du (2)
−∞ −∞
' ∞
= η0 h2 (u) du, (3)
−∞
by the sifting property of the delta function.
381
(a) Note that

∞
1 1 1
Yi = hn Xi−n = Xi+1 + Xi + Xi−1 (1)
n=−∞
3 3 3
By matching coefficients, we see that

1/3 n = −1, 0, 1
hn = (2)
0 otherwise
(b) By Theorem 11.5, the output autocorrelation is

∞
∞

RY [n] = hi hj RX [n + i − j] (3)
i=−∞ j=−∞
1
1 1
= RX [n + i − j] (4)
9
i=−1 j=−1
1
= (RX [n + 2] + 2RX [n + 1] + 3RX [n] + 2RX [n − 1] + RX [n − 2]) (5)
9
Substituting in RX [n] yields
⎧
⎪
⎪ 1/3 n=0
⎨
2/9 |n| = 1
RY [n] = (6)
⎪
⎪ 1/9 |n| = 2
⎩
0 otherwise

Applying Theorem 11.4 with sampling period Ts = 1/4000 s yields
sin(2000πkTs ) + sin(1000πkTs )
RX [k] = RX (kTs ) = 10 (1)
2000πkTs
sin(0.5πk) + sin(0.25πk)
= 20 (2)
πk
= 10 sinc(0.5k) + 5 sinc(0.25k) (3)
(a) By Theorem 11.5, the expected value of the output is

∞

µW = µY hn = 2µY = 2 (1)
n=−∞
382
(b) Theorem 11.5 also says that the output autocorrelation is
∞
∞

RW [n] = hi hj RY [n + i − j] (2)
i=−∞ j=−∞

1
1
= RY [n + i − j] (3)
i=0 j=0
= RY [n − 1] + 2RY [n] + RY [n + 1] (4)
For n = −3,
RW [−3] = RY [−4] + 2RY [−3] + RY [−2] = RY [−2] = 0.5 (5)
Following the same procedure, its easy to show that RW [n] is nonzero for |n| = 0, 1, 2.
Specifically, ⎧
⎪
⎪ 0.5 |n| = 3
⎪
⎪
⎨ 3 |n| = 2
RW [n] = 7.5 |n| = 1 (6)
⎪
⎪
⎪
⎪ 10 n = 0
⎩
0 otherwise
(c) The second moment of the output is E[Wn2 ] = RW [0] = 10. The variance of Wn is

Var[Wn ] = E Wn2 − (E [Wn ])2 = 10 − 22 = 6 (7)
(d) This part doesn’t require any probability. It just checks your knowledge of linear systems
and convolution. There is a bit of confusion because hn is used to denote both the filter
that transforms Xn to Yn as well as the filter that transforms Yn to Wn . To avoid confusion,
we will use ĥn to denote the filter that transforms Xn to Yn . Using Equation (11.25) for
discrete-time convolution, we can write
∞
∞

Wn = hj Yn−j , Yn−j = ĥi Xn−j−i . (8)
j=−∞ i=−∞
Combining these equations yields

∞
∞

Wn = hj ĥi Xn−j−i . (9)
j=−∞ i=−∞
For each j, we make the substitution k = i + j, and then reverse the order of summation to
obtain
⎛ ⎞
∞ ∞ ∞ ∞
Wn = hj ĥk−j Xn−k = ⎝ hj ĥk−j ⎠ Xn−k . (10)
j=−∞ k=−∞ k=−∞ j=−∞
Thus we see that

∞

gk = hj ĥk−j . (11)
j=−∞
That is, the filter gn is the convolution of the filters ĥn and hn .
383
In the context of our particular problem, the filter ĥn that transforms Xn to Yn is given in
Example 11.5. The two filters are

ĥ = ĥ0 ĥ1 = 1/2 1/2 h = h0 h1 = 1 1 (12)
Keep in mind that hn = ĥn = 0 for n < 0 or n > 1. From Equation (11),
⎧
⎪
⎪ 1/2 k = 0
⎨
1 k=1
gk = ĥk + ĥk−1 = (13)
⎪
⎪ 1/2 k=2
⎩
0 otherwise
(a) By Theorem 11.5, the mean output is

∞

µV = µY hn = (−1 + 1)µY = 0 (1)
n=−∞
(b) Theorem 11.5 also says that the output autocorrelation is

∞
∞

RV [n] = hi hj RY [n + i − j] (2)
i=−∞ j=−∞

1
1
= hi hj RY [n + i − j] (3)
i=0 j=0
= −RY [n − 1] + 2RY [n] − RY [n + 1] (4)
For n = −3,
RV [−3] = −RY [−4] + 2RY [−3] − RY [−2] = RY [−2] = −0.5 (5)
Following the same procedure, it’s easy to show that RV [n] is nonzero for |n| = 0, 1, 2.
Specifically, ⎧
⎪
⎪ −0.5 |n| = 3
⎪
⎪
⎨ −1 |n| = 2
RV [n] = 0.5 |n| = 1 (6)
⎪
⎪
⎪
⎪ 2 n = 0
⎩
0 otherwise
(c) Since E[Vn ] = 0, the variance of the output is E[Vn2 ] = RV [0] = 2. The variance of Wn is

Var[Vn ] = E Wn2 RV [0] = 2 (7)
(d) This part doesn’t require any probability. It just checks your knowledge of linear systems
and convolution. There is a bit of confusion because hn is used to denote both the filter that
transforms Xn to Yn as well as the filter that transforms Yn to Vn . To avoid confusion, we will
384
use ĥn to denote the filter that transforms Xn to Yn . Using Equation (11.25) for discrete-time
convolution, we can write
∞
∞

Vn = hj Yn−j , Yn−j = ĥi Xn−j−i . (8)
j=−∞ i=−∞
Combining these equations yields

∞
∞

Vn = hj ĥi Xn−j−i . (9)
j=−∞ i=−∞
For the inner sum, we make the substitution k = i+j and then reverse the order of summation
to obtain
⎛ ⎞
∞
∞ ∞
∞
Vn = hj ĥk−j Xn−k = ⎝ hj ĥk−j ⎠ Xn−k . (10)
j=−∞ k=−∞ k=−∞ j=−∞
Thus we see that

∞

fk = hj ĥk−j . (11)
j=−∞
That is, the filter fn is the convolution of the filters ĥn and hn . In the context of our particular
problem, the filter ĥn that transforms Xn to Yn is given in Example 11.5. The two filters are

ĥ = ĥ0 ĥ1 = 1/2 1/2 h = h0 h1 = 1 −1 (12)
Keep in mind that hn = ĥn = 0 for n < 0 or n > 1. From Equation (11),
⎧
⎪
⎪ 1/2 k=0
⎨
0 k=1
fk = ĥk − ĥk−1 = (13)
⎪
⎪ −1/2 k = 2
⎩
0 otherwise

We start with Theorem 11.5:
∞
∞

RY [n] = hi hj RX [n + i − j] (1)
i=−∞ j=−∞
= RX [n − 1] + 2RX [n] + RX [n + 1] (2)

First we observe that for n ≤ −2 or n ≥ 2,
RY [n] = RX [n − 1] + 2RX [n] + RX [n + 1] = 0 (3)
This suggests that RX [n] = 0 for |n| > 1. In addition, we have the following facts:
RY [0] = RX [−1] + 2RX [0] + RX [1] = 2 (4)
RY [−1] = RX [−2] + 2RX [−1] + RX [0] = 1 (5)
RY [1] = RX [0] + 2RX [1] + RX [2] = 1 (6)
A simple solution to this set of equations is RX [0] = 1 and RX [n] = 0 for n = 0.
385
The mean of Yn = (Xn + Yn−1 )/2 can be found by realizing that Yn is an infinite sum of the Xi ’s.

1 1 1
Yn = Xn + Xn−1 + Xn−2 + . . . (1)
2 4 8
Since the Xi ’s are each of zero mean, the mean of Yn is also 0. The variance of Yn can be expressed
as
∞

1 1 1 1 1
Var[Yn ] = + + + . . . Var[X] = ( )i σ 2 = ( − 1)σ 2 = σ 2 /3 (2)
4 16 64 4 1 − 1/4
i=1
The above infinite sum converges to 1

1−1/4 − 1 = 1/3, implying
Var [Yn ] = (1/3) Var [X] = 1/3 (3)
The covariance of Yi+1 Yi can be found by the same method.

1 1 1 1 1 1
Cov[Yi+1 , Yi ] = [ Xn + Xn−1 + Xn−2 + . . .][ Xn−1 + Xn−2 + Xn−3 + . . .] (4)
2 4 8 2 4 8
Since E[Xi Xj ] = 0 for all i = j, the only terms that are left are
∞
∞
1 1 2 1 1
Cov[Yi+1 , Yi ] = E[X i ] = E[Xi2 ] (5)
2i 2i−1 2 4i
i=1 i=1
Since E[Xi2 ] = σ 2 , we can solve the above equation, yielding
Cov [Yi+1 , Yi ] = σ 2 /6 (6)
Finally the correlation coefficient of Yi+1 and Yi is

Cov[Yi+1 , Yi ] σ 2 /6 1
ρYi+1 Yi = = 2 = (7)
Var[Yi+1 ] Var[Yi ] σ /3 2

There is a technical difficulty with this problem since Xn is not defined for n < 0. This implies
CX [n, k] is not defined for k < −n and thus CX [n, k] cannot be completely independent of k. When
n is large, corresponding to a process that has been running for a long time, this is a technical
issue, and not a practical concern. Instead, we will find σ̄ 2 such that CX [n, k] = CX [k] for all n
and k for which the covariance function is defined. To do so, we need to express Xn in terms of
Z0 , Z1 , . . . , Zn1 . We do this in the following way:
Xn = cXn−1 + Zn−1 (1)

= c[cXn−2 + Zn−2 ] + Zn−1 (2)
2
= c [cXn−3 + Zn−3 ] + cZn−2 + Zn−1 (3)
..
. (4)
n
= c X0 + c n−1
Z0 + c n−2
Z2 + · · · + Zn−1 (5)

n−1
= cn X0 + cn−1−i Zi (6)
i=0
386
Since E[Zi ] = 0, the mean function of the Xn process is

n−1
E [Xn ] = cn E [X0 ] + cn−1−i E [Zi ] = E [X0 ] (7)
i=0
Thus, for Xn to be a zero mean process, we require that E[X0 ] = 0. The autocorrelation function
can be written as
⎡! "⎛ ⎞⎤

n−1
n+k−1
RX [n, k] = E [Xn Xn+k ] = E ⎣ cn X0 + cn−1−i Zi ⎝cn+k X0 + cn+k−1−j Zj ⎠⎦ (8)
i=0 j=0
Although it was unstated in the problem, we will assume that X0 is independent of Z0 , Z1 , . . . so

that E[X0 Zi ] = 0. Since E[Zi ] = 0 and E[Zi Zj ] = 0 for i = j, most of the cross terms will drop
out. For k ≥ 0, autocorrelation simplifies to

n−1
1 − c2n
RX [n, k] = c2n+k Var[X0 ] + c2(n−1)+k−2i) σ̄ 2 = c2n+k Var[X0 ] + σ̄ 2 ck (9)
1 − c2
i=0
Since E[Xn ] = 0, Var[X0 ] = RX [n, 0] = σ 2 and we can write for k ≥ 0,

k

2 c σ̄ 2
RX [n, k] = σ̄ +c2n+k
σ −
2
(10)
1 − c2 1 − c2
For k < 0, we have

⎡! "⎛ ⎞⎤

n−1
n+k−1
RX [n, k] = E ⎣ cn X0 + cn−1−i Zi ⎝cn+k X0 + cn+k−1−j Zj ⎠⎦ (11)
i=0 j=0

n+k−1
= c2n+k Var[X0 ] + c−k c2(n+k−1−j) σ̄ 2 (12)
j=0
1 − c2(n+k)
= c2n+k σ 2 + σ̄ 2 c−k (13)
1 − c2
2

σ̄ −k σ̄ 2
= c +c2n+k
σ −
2
(14)
1 − c2 1 − c2
We see that RX [n, k] = σ 2 c|k| by choosing
σ̄ 2 = (1 − c2 )σ 2 (15)

We can recusively solve for Yn as follows.
Yn = aXn + aYn−1 (1)

= aXn + a[aXn−1 + aYn−2 ] (2)
2 2
= aXn + a Xn−1 + a [aXn−2 + aYn−3 ] (3)
387
By continuing the same procedure, we can conclude that

n
Yn = aj+1 Xn−j + an Y0 (4)
j=0
Since Y0 = 0, the substitution i = n − j yields

n
Yn = an−i+1 Xi (5)
i=0
Now we can calculate the mean
0 1

n
n
n−i+1
E [Yn ] = E a Xi = an−i+1 E [Xi ] = 0 (6)
i=0 i=0
To calculate the autocorrelation RY [m, k], we consider first the case when k ≥ 0.
⎡ ⎤

m
m+k
m m+k
CY [m, k] = E ⎣ a m−i+1
Xi am+k−j+1 ⎦
Xj = am−i+1 am+k−j+1 E [Xi Xj ] (7)
i=0 j=0 i=0 j=0
Since the Xi is a sequence of iid standard normal random variables,

1 i=j
E [Xi Xj ] = (8)
0 otherwise
Thus, only the i = j terms make a nonzero contribution. This implies

m
CY [m, k] = am−i+1 am+k−i+1 (9)
i=0

m
k
=a a2(m−i+1) (10)
i=0
2 m+1
=a k
+ (a2 )m + · · · + a2
(a ) (11)
a2
= ak 1 − (a2 )m+1 (12)
1−a 2
For k ≤ 0, we start from

m m+k
CY [m, k] = am−i+1 am+k−j+1 E [Xi Xj ] (13)
i=0 j=0
As in the case of k ≥ 0, only the i = j terms make a contribution. Also, since m + k ≤ m,

m+k
m+k
m−j+1 m+k−j+1 −k
CY [m, k] = a a =a am+k−j+1 am+k−j+1 (14)
j=0 j=0
By steps quite similar to those for k ≥ 0, we can show that

a2
−k
CY [m, k] = a 1 − (a2 m+k+1
) (15)
1 − a2
A general expression that is valid for all m and k would be
a2
|k|
CY [m, k] = a 1 − (a 2 min(m,m+k)+1
) (16)
1 − a2
Since CY [m, k] depends on m, the Yn process is not wide sense stationary.
388
Sincethe process X n has expected value E[Xn ] = 0, we know that CX (k) = RX (k) = 2−|k| . Thus

X = X1 X2 X3 has covariance matrix
⎡ ⎤ ⎡ ⎤
20 2−1 2−2 1 1/2 1/4
CX = ⎣2−1 20 2−1 ⎦ = ⎣1/2 1 1/2⎦ . (1)
2−2 2−1 20 1/4 1/2 1
From Definition 5.17, the PDF of X is

1 1 −1
fX (x) = exp − x CX x . (2)
(2π)n/2 [det (CX )]1/2 2
If we are using Matlab for calculations, it is best to decalre the problem solved at this point.
However, if you like algebra, we can write out the PDF in terms of the variables x1 , x2 and x3 . To
do so we find that the inverse covariance matrix is
⎡ ⎤
4/3 −2/3 0
C−1 ⎣ 5/3 −2/3⎦
X = −2/3 (3)
0 −2/3 4/3
A little bit of algebra will show that det(CX ) = 9/16 and that
1 −1 2x2 5x2 2x2 2x1 x2 2x2 x3

x CX x = 1 + 2 + 3 − − . (4)
2 3 6 3 3 3
It follows that

4 2x21 5x22 2x23 2x1 x2 2x2 x3
fX (x) = exp − − − + + . (5)
3(2π)3/2 3 6 3 3 3

The sequence Xn is passed through the filter

h = h0 h1 h2 = 1 −1 1 (1)
The output sequence is Yn .

(a) Following the approach of Equation (11.58), we can write the output Y3 = Y1 Y2 Y3 as
⎡ ⎤ ⎡ ⎤
⎡ ⎤⎡ ⎤ X0 ⎡ ⎤ X0
Y1 h1 h0 0 0 ⎢ ⎥ −1 1 0 0 ⎢ ⎥
X1 ⎥ ⎣ X1 ⎥
Y3 = ⎣Y2 ⎦ = ⎣h2 h1 h0 0 ⎦ ⎢
⎣X2 ⎦ = 1 −1 1 0⎦ ⎢
⎣X2 ⎦ . (2)
Y3 0 h2 h1 h0 0 1 −1 1
X3 X3
H
X
We note that the components of X are iid Gaussian (0, 1) random variables. Hence X has
covariance matrix CX = I, the identity matrix. Since Y3 = HX,
⎡ ⎤
2 −2 1
CY3 = HCX H = HH = ⎣−2 3 −2⎦ . (3)
1 −2 3
389
Some calculation (by hand or by Matlab) will show that det(CY3 ) = 3 and that
⎡ ⎤
5 4 1
1⎣
C−1
Y3 = 4 5 2⎦ . (4)
3
1 2 2
Some algebra will show that
5y12 + 5y22 + 2y32 + 8y1 y2 + 2y1 y3 + 4y2 y3
y C−1
Y3 y = . (5)
3
This implies Y3 has PDF

1 1 −1
fY3 (y) = exp − y CY3 y (6)
(2π)3/2 [det (CY3 )]1/2 2

1 5y1 + 5y2 + 2y32 + 8y1 y2 + 2y1 y3 + 4y2 y3
2 2
= √ exp − . (7)
(2π)3/2 3 6

(b) To find the PDF of Y2 = Y1 Y2 , we start by observing that the covariance matrix of Y2
is just the upper left 2 × 2 submatrix of CY3 . That is,

2 −2 −1 3/2 1
CY2 = and CY2 = . (8)
−2 3 1 1
Since det(CY2 ) = 2, it follows that

1 1 −1
fY2 (y) = exp − y CY2 y (9)
(2π)3/2 [det (CY2 )]1/2 2

1 3 2
= √ exp − y1 − 2y1 y2 − y2 .
2
(10)
(2π)3/2 2 2


h = h0 h1 h2 = 1 −1 1 (1)
The output
is Yn . Following the approach of Equation (11.58), we can write the output
sequence
Y = Y1 Y2 Y3 as
⎡ ⎤ ⎡ ⎤
⎡ ⎤ ⎡ ⎤ X−1 ⎡ ⎤ X−1
Y1 h2 h1 h0 0 0 ⎢ ⎢ X0 ⎥
⎥ 1 −1 1 0 0 ⎢ ⎢ X0 ⎥
⎥
Y = ⎣Y2 ⎦ = ⎣ 0 h2 h1 h0 0 ⎦ ⎢ ⎥ ⎣
⎢ X1 ⎥ = 0 1 −1 1 0 ⎢ X1 ⎥ .
⎦⎢ ⎥ (2)
Y3 0 0 h2 h1 h0 ⎣ X2 ⎦ 0 0 1 −1 1 ⎣ X2 ⎦
X3 X
H 3

X
Since Xn has autocovariance function CX (k) = 2−|k| ,

X has covariance matrix
⎡ ⎤
1 1/2 1/4 1/8 1/16
⎢ 1/2 1 1/2 1/4 1/8 ⎥
⎢ ⎥
CX = ⎢⎢ 1/4 1/2 1 1/2 1/4 ⎥ ⎥. (3)
⎣ 1/8 1/4 1/2 1 1/2 ⎦
1/16 1/8 1/4 1/2 1
390
Since Y = HX, ⎡ ⎤
3/2 −3/8 9/16
CY = HCX H = ⎣−3/8 3/2 −3/8⎦ . (4)
9/16 −3/8 3/2
Some calculation (by hand or preferably by Matlab) will show that det(CY ) = 675/256 and that
⎡ ⎤
12 2 −4
1 ⎣ 2 11 2 ⎦ .
C−1
Y = (5)
15
−4 2 12
12y12 + 11y22 + 12y32 + 4y1 y2 + −8y1 y3 + 4y2 y3
y C−1
Y y = . (6)
15
This implies Y has PDF

1 1 −1
fY (y) = exp − y CY y (7)
(2π)3/2 [det (CY )]1/2 2

16 12y1 + 11y22 + 12y32 + 4y1 y2 + −8y1 y3 + 4y2 y3
2
= √ exp − . (8)
(2π)3/2 15 3 30
This solution is another demonstration of why the PDF of a Gaussian random vector should be
left in vector form.
Comment: We know from Theorem 11.5 that Yn is a stationary Gaussian process. As a result,
the random variables Y1 , Y2 and Y3 are identically distributed and CY is a symmetric Toeplitz
matrix. This might make on think that the PDF fY (y) should be symmetric in the variables y1 ,
y2 and y3 . However, because Y2 is in the middle of Y1 and Y3 , the information provided by Y1 and
Y3 about Y2 is different than the information Y1 and Y2 convey about Y3 . This fact appears as
asymmetry in fY (y).


h = h0 h1 h2 = 1 0 −1 (1)
The output
is Yn . Following the approach of Equation (11.58), we can write the output
sequence
Y = Y1 Y2 Y3 as
⎡ ⎤ ⎡ ⎤
⎡ ⎤ ⎡ ⎤ X −1 ⎡ ⎤ X −1
Y1 h2 h1 h0 0 0 ⎢ ⎢ X0 ⎥
⎥ 1 0 −1 0 0 ⎢ ⎢ X0 ⎥
⎥
Y = ⎣Y2 ⎦ = ⎣ 0 h2 h1 h0 0 ⎦ ⎢ ⎥ ⎣
⎢ X1 ⎥ = 0 1 0 −1 0 ⎢ X1 ⎥ .
⎦⎢ ⎥ (2)
Y3 0 0 h2 h1 h0 ⎣ X2 ⎦ 0 0 1 0 −1 ⎣ X2 ⎦
X3 X
H 3

X
Since Xn has autocovariance function CX (k) = 2−|k| ,

X has the Toeplitz covariance matrix
⎡ ⎤
1 1/2 1/4 1/8 1/16
⎢ 1/2 1 1/2 1/4 1/8 ⎥
⎢ ⎥
CX = ⎢⎢ 1/4 1/2 1 1/2 1/4 ⎥ ⎥. (3)
⎣ 1/8 1/4 1/2 1 1/2 ⎦
1/16 1/8 1/4 1/2 1
391
Since Y = HX, ⎡ ⎤
3/2 3/8 −9/16
CY = HCX H = ⎣ 3/8 3/2 3/8 ⎦ . (4)
−9/16 3/8 3/2
Some calculation (preferably by Matlab) will show that det(CY ) = 297/128 and that
⎡ ⎤
10/11 −1/3 14/33
C−1 ⎣ 5/6 −1/3 ⎦ .
Y = −1/3 (5)
14/33 −1/3 10/11
10 2 5 2 10 2 2 28 2
y C−1
Y y = y + y + y − y1 y2 + y1 y3 − y2 y3 . (6)
11 1 6 2 11 3 3 33 3
This implies Y has PDF

1 1 −1
fY (y) = exp − y CY y (7)
(2π)3/2 [det (CY )]1/2 2
√
8 2 5 2 5 2 5 2 1 14 1
= √ exp − y1 − y2 − y3 + y1 y2 − y1 y3 + y2 y3 . (8)
(2π)3/2 3 33 11 12 11 3 33 3
This solution is yet another demonstration of why the PDF of a Gaussian random vector should
be left in vector form.

This problem
is solved
using Theorem 11.9
with M =
2 and k = 1. The optimum linear predictor
filter h = h0 h1 of Xn+1 given Xn = Xn−1 Xn is given by

←
− h
h = 1 = R−1 Xn RXn Xn+k , (1)
h0
where
RX [0] RX [1] 1 3/4
RXn = = (2)
RX [1] RX [0] 3/4 1
and
Xn−1 RX [2] 1/2
RXn Xn+1 =E Xn+1 = = . (3)
Xn RX [1] 3/4
Thus the filter vector h satisfies
−1
←
− h1 1 3/4 1/2 −1/7
h = = = . (4)
h0 3/4 1 3/4 6/7

Thus h = 6/7 −1/7 and the optimum linear predictor of Xn+1 given Xn and Xn−1 is

←
− 1 6 Xn−1 1 6
X̂n+1 = h Xn = − 7 7 = − Xn−1 + Xn . (5)
Xn 7 7
To find the mean square error of this predictor, we can calculate it directly as
0 2 1
1 6
e∗L = E (X̂n+1 − Xn+1 )2 = E − Xn−1 + Xn − Xn+1 . (6)
7 7
392
By expanding the square and taking the expectation, each cross-term is of the form E[Xi Xj ] =
RX [i − j], so that
0 2 1
1 6 1 6
e∗L = E − Xn−1 + Xn − Xn+1 − Xn−1 + Xn − Xn+1 (7)
7 7 7 7
1 12 2 36 12
= RX [0] − RX [1] + RX [2] + RX [0] − RX [1] + RX [0] (8)
49 49 7 49 7
86 96 2 3
= RX [0] − RX [1] + RX [2] = . (9)
49 49 7 7
This direct method is already tedious, even for a simple filter of order M = 2. A better way to
calculate the mean square error is to recall that Theorem 11.9 is just of Theorem 9.7 expressed in
the terminology of filters. Expressing part (c) of Theorem 9.7 in terms of the linear prediction filter
h, the mean square error of the predictor is
←−
e∗L = Var[Xn+1 ] − h RXn Xn+k (10)

←
− RX [2]
= RX [0] − h (11)
RX [1]

1/2 3
= 1 − −1/7 6/7 = . (12)
3/4 7
For an arbitrary filter order M , Equation (10) is a much simpler way to compute the mean square
error.

This problem
is solved using Theorem
11.9 with
k = 1. The optimum linear predictor filter
h = h0 h1 of Xn+1 given Xn = Xn−1 Xn is given by

←− h
h = 1 = R−1 Xn RXn Xn+k , (1)
h0
where
RX [0] RX [1] 1.1 0.75
RXn = = (2)
RX [1] RX [0] 0.75 1.1
and
Xn−1 RX [2] 0.5
RXn Xn+1 =E Xn+1 = = . (3)
Xn RX [1] 0.75
Thus the filter vector h satisfies
−1
←− h1 1.1 0.75 0.5 −0.0193
h = = = . (4)
h0 0.75 1.1 0.75 0.6950

Thus h = 0.6950 −0.0193 and the optimum linear predictor of Xn+1 given Xn and Xn−1 is

←
− Xn−1
X̂n+1 = h Xn = −0.0193 0.6950 = −0.0193Xn−1 + 0.6950Xn . (5)
Xn
To find the mean square error of this predictor, we can calculate it directly as

e∗L = E (X̂n+1 − Xn+1 )2 = E (−0.0193Xn−1 + 0.6950Xn − Xn+1 )2 . (6)
393
We can expand the square and take the expectation term by term since each cross-term is of the
form E[Xi Xj ] = RX [i − j]. This approach is followed in the solution of Problem 11.4.1 and it is
quite tedious. A better way to calculate the mean square error is to recall that Theorem 11.9 is
just Theorem 9.7 expressed in the terminology of filters. Expressing part (c) of Theorem 9.7 in
terms of the linear prediction filter h, the mean square error of the predictor is
←−
e∗L = Var[Xn+1 ] − h RXn Xn+k (7)

←
− RX [2]
= RX [0] − h (8)
RX [1]

1/2
= 1.1 − −0.0193 0.6950 = 0.5884. (9)
3/4
Comment: It is instructive to compare this solution to the solution of Problem 11.4.1 where the
random process, denoted X̃n here to distinguish it from Xn in this problem, has autocorrelation
function
1 − 0.25 |k| |k| ≤ 4,
RX̃ [k] = (10)
0 otherwise.
The difference is simply that RX̃ [0] = 1, rather than RX [0] = 1.1 as in this problem. This difference
corresponds to adding an iid noise sequence to X̃n to create Xn . That is,
Xn = X̃n + Zn (11)
where Zn is an iid additive noise sequence with autocorrelation function RZ [k] = 0.1δ[k] that is
independent of the Xn process. Thus Xn in this problem can be viewed as a noisy version of
X̃n in Problem 11.4.1. Because
the X̃n process
is less noisy, the optimal predictor filter of X̃n+1

given X̃n−1 and X̃n is h̃ = 6/7 −1/7 = 0.8571 −0.1429 , which places more emphasis on the
current value X̃n in predicting the next value.
In addition, the mean squared error of the predictor of X̃n+1 is only 3/7 = 0.4285, which is less
than 0.5884. Not surprisingly, the noise in the Xn process reduces the performance of the predictor.

This problem generalizes Example 11.14 in that −0.9 is replaced by the parameter c and the noise

variance 0.2 is replaced by η 2 . Because we are only finding the first order filter h = h0 h1 , it is
relatively simple to generalize the solution
of Example 11.14 to the parameter values c and η 2 .

Based on the observation Y = Yn−1 Yn , Theorem 11.11 states that the linear MMSE esti-
←
−
mate of X = Xn is h Y where
←
−
h = R−1 −1
Y RYXn = (RXn + RWn ) RXn Xn . (1)

From Equation (11.82), RXn Xn = RX [1] RX [0] = c 1 . From the problem statement,
2
1 c η 0 1 + η2 c
RXn + RWn = + = . (2)
c 1 0 η2 c 1 + η2
394
This implies
−1
←
− 1 + η2 c c
h = (3)
c 1 + η2 1

1 1 + η2 −c c
= (4)
(1 + η 2 )2 − c2 −c 1 + η2 1

1 cη 2
= . (5)
(1 + η 2 )2 − c2 1 + η 2 − c2
The optimal filter is

1 1 + η 2 − c2
h= . (6)
(1 + η 2 )2 − c2 cη 2
To find the mean square error of this predictor, we recall that Theorem 11.11 is just Theorem 9.7
expressed in the terminology of filters. Expressing part (c) of Theorem 9.7 in terms of the linear
estimation filter h, the mean square error of the estimator is
←
−
e∗L = Var[Xn ] − h RYn Xn (7)
←
−
= Var[Xn ] − h RXn Xn (8)

←− c
= RX [0] − h (9)
1
c2 η 2 + η 2 + 1 − c2
=1− . (10)
(1 + η 2 )2 − c2
Note that we always find that e∗L < Var[Xn ] = 1 simply because the optimal estimator cannot be
worse than the blind estimator that ignores the observation Yn .

In this problem, we find the mean square estimation error of the optimal first order filter in Prob-
lem 11.4.3. This problem highlights a shortcoming of Theorem 11.11 in that the theorem doesn’t
explicitly provide the mean square error associated with the optimal filter h. We recall from the
discussion at the start of Section 11.4 that Theorem 11.11 is derived from Theorem 9.7 with
←
−
h = a = R−1 −1
Y RYX = (RXn + RWn ) RXn Xn (1)
From Theorem 9.7, the mean square error of the filter output is
e∗L = Var[X] − a RYX (2)

←
−
= RX [0] − h RXn Xn (3)
= RX [0] − RXn Xn (RXn + RWn ) −1
RXn Xn (4)
Equations (3) and (4) are general expressions for the means square error of the optimal linear filter
that can be applied to any situation described by Theorem 11.11.
To apply this result to the problem at hand, we observe that RX [0] = c0 = 1 and that

←− 1 cη 2 RX [1] c
h = , RXn Xn = = (5)
(1 + η 2 )2 − c2 1 + η 2 − c2 RX [0] 1
395
This implies
←
−
e∗L = RX [0] − h RXn Xn (6)

1 2 c
=1− cη 1 + η − c
2 2 (7)
(1 + η 2 )2 − c2 1
c2 η 2 + 1 + η 2 − c2
=1− (8)
(1 + η 2 )2 − c2

1 + η 2 − c2
= η2 (9)
(1 + η 2 )2 − c2
The remaining question is what value of c minimizes the mean square error e∗L . The usual
de∗
approach is to set the derivative dcL to zero. This would $yield the incorrect answer c = 0. In fact,
d2 e ∗ $
evaluating the second derivative at c = 0 shows that dc2L $ < 0. Thus the mean square error e∗L
c=0
is maximum at c = 0. For a more careful analysis, we observe that e∗L = η 2 f (x) where
a−x
f (x) = , (10)
a2 − x
with x = c2 , and a = 1 + η 2 . In this case, minimizing f (x) is equivalent to minimizing the mean
square error. Note that for RX [k] to be a respectable autocorrelation function, we must have
|c| ≤ 1. Thus we consider only values of x in the interval 0 ≤ x ≤ 1. We observe that
df (x) a2 − a
=− 2 (11)
dx (a − x)2
Since a > 1, the derivative is negative for 0 ≤ x ≤ 1. This implies the mean square error is
minimized by making x as large as possible, i.e., x = 1. Thus c = 1 minimizes the mean square
error. In fact c = 1 corresponds to the autocorrelation function RX [k] = 1 for all k. Since each Xn
has zero expected value, every pair of sample Xn and Xm has correlation coefficient
Cov [Xn , Xm ] RX [n − m]
ρXn ,Xm = = = 1. (12)
Var[Xn ] Var[Xm ] RX [0]
That is, c = 1 corresponds to a degenerate process in which every pair of samples Xn and Xm are
perfectly correlated. Physically, this corresponds to the case where where the random process Xn
is generated by generating a sample of a random variable X and setting Xn = X for all n. The
observations are then of the form Yn = X +Zn . That is, each observation is just a noisy observation
of the random variable X. For c = 1, the optimal filter

1 1
h= 2
. (13)
2+η 1
is just an equally weighted average of the past two samples.

This problem involves both linear estimation and prediction because we are using Yn−1 , the noisy
observation of Xn1 to estimate the future value Xn . Thus we can’t follow the cookbook recipes of
Theorem 11.9 or Theorem 11.11. Instead we go back to Theorem 9.4 to find the minimum mean
396
square error linear estimator. In that theorem, Xn and Yn−1 play the roles of X and Y . That is,
our estimate X̂n of Xn is

Var[Xn ] 1/2
X̂n = X̂L (Yn−1 ) = ρXn ,Yn−1 (Yn−1 − E [Yn−1 ]) + E [Xn ] (1)
Var[Yn−1 ]
By recursive application of Xn = cXn−1 + Zn−1 , we obtain

n
n
Xn = a X0 + aj−1 Zn−j (2)
j=1

n
The expected value of Xn is E[Xn ] = an E[X0 ] + j=1 a
j−1 E[Z
n−j ] = 0. The variance of Xn is

n
n
Var[Xn ] = a2n Var[X0 ] + [aj−1 ]2 Var[Zn−j ] = a2n Var[X0 ] + σ 2 [a2 ]j−1 (3)
j=1 j=1
Since Var[X0 ] = σ 2 /(1 − c2 ), we obtain

c2n σ 2 σ 2 (1 − c2n ) σ2
Var[Xn ] = + = (4)
1 − c2 1 − c2 1 − c2
Note that E[Yn−1 ] = dE[Xn−1 ] + E[Wn ] = 0. The variance of Yn−1 is
d2 σ 2
Var[Yn−1 ] = d2 Var[Xn−1 ] + Var[Wn ] = + η2 (5)
1 − c2
Since Xn and Yn−1 have zero mean, the covariance of Xn and Yn−1 is
Cov [Xn , Yn−1 ] = E [Xn Yn−1 ] = E [(cXn−1 + Zn−1 ) (dXn−1 + Wn−1 )] (6)
From the problem statement, we learn that
E[Xn−1 Wn−1 ] = 0 E[Xn−1 ]E[Wn−1 ] = 0
E[Zn−1 Xn−1 ] = 0 E[Zn−1 Wn−1 ] = 0
Hence, the covariance of Xn and Yn−1 is
Cov [Xn , Yn−1 ] = cd Var[Xn−1 ] (7)
The correlation coefficient of Xn and Yn−1 is
Cov [Xn , Yn−1 ]
ρXn ,Yn−1 = (8)
Var[Xn ] Var[Yn−1 ]
Since E[Yn−1 ] and E[Xn ] are zero, the linear predictor for Xn becomes

Var[Xn ] 1/2 Cov [Xn , Yn−1 ] cd Var[Xn−1 ]
X̂n = ρXn ,Yn−1 Yn−1 = Yn−1 = Yn−1 (9)
Var[Yn−1 ] Var[Yn−1 ] Var[Yn−1 ]
Substituting the above result for Var[Xn ], we obtain the optimal linear predictor of Xn given Yn−1 .
c 1
X̂n = Yn−1 (10)
d 1 + β (1 − c2 )
2
where β 2 = η 2 /(d2 σ 2 ). From Theorem 9.4, the mean square estimation error at step n
1 + β2
e∗L (n) = E[(Xn − X̂n )2 ] = Var[Xn ](1 − ρ2Xn ,Yn−1 ) = σ 2 (11)
1 + β 2 (1 − c2 )
We see that mean square estimation error e∗L (n) = e∗L , a constant for all n. In addition, e∗L is an
increasing function β.
397
To use Table 11.1, we write RX (τ ) in terms of the autocorrelation
sin(πx)
sinc(x) = . (1)
πx
In terms of the sinc(·) function, we obtain
RX (τ ) = 10 sinc(2000τ ) + 5 sinc(1000τ ). (2)
From Table 11.1,
10 f 5 f
SX (f ) = rect + rect (3)
2,000 2000 1,000 1,000
Here is a graph of the PSD.
0.012
0.01
0.008
SX(f)
0.006
0.004
0.002
0
−1500 −1000 −500 0 500 1000 1500
f

The process Y (t) has expected value E[Y (t)] = 0. The autocorrelation of Y (t) is
RY (t, τ ) = E [Y (t)Y (t + τ )] = E [X(αt)X(α(t + τ ))] = RX (ατ ) (1)
Thus Y (t) is wide sense stationary. The power spectral density is
' ∞
SY (f ) = RX (ατ )e−j2πf τ dτ. (2)
−∞
At this point, we consider the cases α > 0 and α < 0 separately. For α > 0, the substitution
τ = ατ yields
'
1 ∞ SX (f /α)
SY (f ) = RX (τ )e−j2π(f /α)τ dτ = (3)
α −∞ α
When α < 0, we start with Equation (2) and make the substitution τ = −ατ , yielding
' ∞
1 f
SY (f ) = RX (−τ )e−j2π −α τ dτ . (4)
−α −∞
Since RX (−τ ) = RX (τ ),
' ∞
1 f 1 f
−j2π −α τ
SY (f ) = RX (τ )e dτ . = SX (5)
−α −∞ −α −α
For −α = |α| for α < 0, we can combine the α > 0 and α < 0 cases in the expression

1 f
SY (f ) = SX . (6)
|α| |α|
398
Since the random sequence Xn has autocorrelation function
RX [k] = δk + (0.1)|k| , (1)
We can find the PSD directly from Table 11.2 with 0.1|k| corresponding to a|k| . The table yields
1 − (0.1)2 2 − 0.2 cos 2πφ

SX (φ) = 1 + = . (2)
1 + (0.1) − 2(0.1) cos 2πφ
2 1.01 − 0.2 cos 2πφ

First we show that SY X (f ) = SXY (−f ). From the definition of the cross spectral density,
' ∞
SY X (f ) = RY X (τ )e−j2πf τ dτ (1)
−∞
Making the subsitution τ = −τ yields

' ∞

SY X (f ) = RY X (−τ )ej2πf τ dτ (2)
−∞
By Theorem 10.14, RY X (−τ ) = RXY (τ ). This implies

' ∞

SY X (f ) = RXY (τ )e−j2π(−f )τ dτ = SXY (−f ) (3)
−∞
To complete the problem, we need to show that SXY (−f ) = [SXY (f )]∗ . First we note that since
RXY (τ ) is real valued, [RXY (τ )]∗ = RXY (τ ). This implies
' ∞
∗
[SXY (f )] = [RXY (τ )]∗ [e−j2πf τ ]∗ dτ (4)
−∞
' ∞
= RXY (τ )e−j2π(−f )τ dτ (5)
−∞
= SXY (−f ) (6)

Let a = 1/RC. The solution to this problem parallels Example 11.22.
(a) From Table 11.1, we observe that
2 · 104 1
SX (f ) = H(f ) = (1)
(2πf )2 + 104 a + j2πf
By Theorem 11.16,
2 · 104
SY (f ) = |H(f )|2 SX (f ) = (2)
[(2πf )2 + a2 ][(2πf )2 + 104 ]
To find RY (τ ), we use a form of partial fractions expansion to write

A B
SY (f ) = + (3)
(2πf )2 + a2 (2πf )2 + 104
399
Note that this method will work only if a = 100. This same method was also used in
Example 11.22. The values of A and B can be found by
$ $
2 · 104 $ −2 · 104 2 · 104 $ 2 · 104
A= $ = B = $ = (4)
$
(2πf )2 + 104 f = ja a2 − 104 $
a2 + 104 f = j100 a2 − 104
2π 2π
This implies the output power spectral density is
−104 /a 2a 1 200
SY (f ) = + (5)
a2 − 104 (2πf )2 + a2 a2 − 104 (2πf )2 + 104
Since e−c|τ | and 2c/((2πf )2 + c2 ) are Fourier transform pairs for any constant c > 0, we see
that
−104 /a −a|τ | 100
RY (τ ) = 2 e + 2 e−100|τ | (6)
a − 10 4 a − 104
(b) To find a = 1/(RC), we use the fact that
−104 /a 100
E Y 2 (t) = 100 = RY (0) = 2 + 2 (7)
a − 10 4 a − 104
Rearranging, we find that a must satisfy
a3 − (104 + 1)a + 100 = 0 (8)
This cubic polynomial has three roots:

√ √
a = 100 a = −50 + 2501 a = −50 − 2501 (9)
Recall that a = 100 is not a valid solution because our expansion of SY (f ) was not valid
√ Also, we require a > 0 in order to take the inverse transform of SY (f ). Thus
for a = 100.
a = −50 + 2501 ≈ 0.01 and RC ≈ 100.
(a) RW (τ ) = δ(τ ) is the autocorrelation function whose Fourier transform is SW (f ) = 1.
(b) The output Y (t) has power spectral density
SY (f ) = |H(f )|2 SW (f ) = |H(f )|2 (1)
(c) Since |H(f )| = 1 for f ∈ [−B, B], the average power of Y (t) is
' ∞ '
B
E Y 2 (t) = SY (f ) df = df = 2B (2)
−∞ −B
(d) Since the white noise W (t) has zero mean, the mean value of the filter output is
E [Y (t)] = E [W (t)] H(0) = 0 (3)
400
Since SY (f ) = |H(f )|2 SX (f ), we first find
|H(f )|2 = H(f )H ∗ (f ) (1)

= a1 e−j2πf t1 + a2 e−j2πf t2 a1 ej2πf t1 + a2 ej2πf t2 (2)

= a21 + a22 + a1 a2 e−j2πf (t2 −t1 ) + e−j2πf (t1 −t2 ) (3)
It follows that the output power spectral density is
SY (f ) = (a21 + a22 )SX (f ) + a1 a2 SX (f ) e−j2πf (t2 −t1 ) + a1 a2 SX (f ) e−j2πf (t1 −t2 ) (4)
Using Table 11.1, the autocorrelation of the output is
RY (τ ) = (a21 + a22 )RX (τ ) + a1 a2 (RX (τ − (t1 − t2 )) + RX (τ + (t1 − t2 ))) (5)
(a) The average power of the input is

E X 2 (t) = RX (0) = 1 (1)
(b) From Table 11.1, the input has power spectral density
1 2
SX (f ) = e−πf /4 (2)
2
The output power spectral density is
1
2
e−πf /4 |f | ≤ 2
SY (f ) = |H(f )| SX (f ) =
2
2 (3)
0 otherwise
(c) The average output power is

' ∞ '
1 2
2 /4
2
E Y (t) = SY (f ) df = e−πf df (4)
−∞ 2 −2
This integral cannot be expressed in closed form. However, we can express it inthe form of
the integral of a standardized Gaussian PDF by making the substitution f = z 2/π. With
this subsitution,
' √2π
2 1 −z 2 /2
E Y (t) = √ √ e dz (5)
2π − 2π
√ √
= Φ( 2π) − Φ(− 2π) (6)
√
= 2Φ( 2π) − 1 = 0.9876 (7)
The output power almost equals the input power because the filter bandwidth is sufficiently
wide to pass through nearly all of the power of the input.
401
(a) From Theorem 11.13(b),

' ∞ '
100
E X 2 (t) = SX (f ) df 10−4 df = 0.02 (1)
−∞ −100
(b) From Theorem 11.17

10−4 H(f ) |f | ≤ 100
SXY (f ) = H(f )SX (f ) = (2)
0 otherwise
(c) From Theorem 10.14,

RY X (τ ) = RXY (−τ ) (3)
From Table 11.1, if g(τ ) and G(f ) are a Fourier transform pair, then g(−τ ) and G∗ (f ) are a
Fourier transform pair. This implies
−4 ∗
∗ 10 H (f ) |f | ≤ 100
SY X (f ) = SXY (f ) = (4)
0 otherwise
(d) By Theorem 11.17,
SY (f ) = H ∗ (f )SXY (f ) = |H(f )|2 SX (f ) (5)

−4
10 /[104 π 2 + (2πf )2 ] |f | ≤ 100
= (6)
0 otherwise

' ∞ '
2
100
10−4
E Y (t) = SY (f ) df = 4 2 2 2
df (7)
−∞ −100 10 π + 4π f
' 100
2 df
= 8 2 (8)
10 π 0 1 + (f /50)2
By making the substitution, f = 50 tan θ, we have df = 50 sec2 θ dθ. Using the identity
1 + tan2 θ = sec2 θ, we have
' tan−1 (2)
100 tan−1 (2)
E Y 2 (t) = 8 2 dθ = = 1.12 × 10−7 (9)
10 π 0 106 π 2

The easy way to do this problem is to use Theorem 11.17 which states
SXY (f ) = H(f )SX (f ) (1)
402
(a) From Table 11.1, we observe that
8 1
SX (f ) = H(f ) = (2)
16 + (2πf )2 7 + j2πf
From Theorem 11.17,
8
SXY (f ) = H(f )SX (f ) = (3)
[7 + j2πf ][16 + (2πf )2 ]
(b) To find the cross correlation, we need to find the inverse Fourier transform of SXY (f ). A
straightforward way to do this is to use a partial fraction expansion of SXY (f ). That is, by
defining s = j2πf , we observe that
8 −8/33 1/3 1/11
= + + (4)
(7 + s)(4 + s)(4 − s) 7+s 4+s 4−s
Hence, we can write the cross spectral density as
−8/33 1/3 1/11
SXY (f ) = + + (5)
7 + j2πf 4 + j2πf 4 − jπf
Unfortunately, terms like 1/(a − j2πf ) do not have an inverse transforms. The solution is to
write SXY (f ) in the following way:
−8/33 8/33 1/11 1/11
SXY (f ) = + + + (6)
7 + j2πf 4 + j2πf 4 + j2πf 4 − j2πf
−8/33 8/33 8/11
= + + (7)
7 + j2πf 4 + j2πf 16 + (2πf )2
(8)
Now, we see from Table 11.1 that the inverse transform is

8 −7τ 8 1
RXY (τ ) = − e u(τ ) + e−4τ u(τ ) + e−4|τ | (9)
33 33 11
(a) Since E[N (t)] = µN = 0, the expected value of the output is µY = µN H(0) = 0.
(b) The output power spectral density is

6 |f |
SY (f ) = |H(f )|2 SN (f ) = 10−3 e−2×10 (1)
(c) The average power is

' ∞ ' ∞
6
E Y 2 (t) = SY (f ) df = 10−3 e−2×10 |f | df (2)
−∞ −∞
' ∞
−3 6
= 2 × 10 e−2×10 f df (3)
0
= 10−3 (4)
403
(d) Since N (t) is a Gaussian process, Theorem 11.3 says Y (t) is a Gaussian process. Thus the
random variable Y (t) is Gaussian with

E [Y (t)] = 0 Var[Y (t)] = E Y 2 (t) = 10−3 (5)
Thus we can use Table 3.1 to calculate

0 1
Y (t) 0.01
P [Y (t) > 0.01] = P > (6)
Var[Y (t)] Var[Y (t)]

0.01
1−Φ √ (7)
0.001
= 1 − Φ(0.32) = 0.3745 (8)

Suppose we assume that N (t) and Y (t) are the input and output of a linear time invariant filter
h(u). In that case,
' t ' ∞
Y (t) = N (u) du = h(t − u)N (u) du (1)
0 −∞
For the above two integrals to be the same, we must have

1 0≤t−u≤t
h(t − u) = (2)
0 otherwise
Making the substitution v = t − u, we have

1 0≤v≤t
h(v) = (3)
0 otherwise
Thus the impulse response h(v) depends on t. That is, the filter response is linear but not time
invariant. Since Theorem 11.2 requires that h(t) be time invariant, this example does not violate
the theorem.
(a) Note that |H(f )| = 1. This implies SM̂ (f ) = SM (f ). Thus the average power of M̂ (t) is
' ∞ ' ∞
q̂ = SM̂ (f ) df = SM (f ) df = q (1)
−∞ −∞
(b) The average power of the upper sideband signal is

E U 2 (t) = E M 2 (t) cos2 (2πfc t + Θ) (2)

− E 2M (t)M̂ (t) cos(2πfc t + Θ) sin(2πfc t + Θ) (3)

+ E M̂ 2 (t) sin2 (2πfc t + Θ) (4)
404
To find the expected value of the random phase cosine, for an integer n = 0, we evaluate
' 2π
1
E [cos(2πfc t + nΘ)] = cos(2πfc t + nθ) dθ (5)
0 2π
1
= sin(2πfc t + nθ)|2π
0 (6)
2nπ
1
= (sin(2πfc t + 2nπ) − sin(2πfc t)) = 0 (7)
2nπ
Similar steps will show that for any integer n = 0, the random phase sine also has expected
value
E [sin(2πfc t + nΘ)] = 0 (8)
Using the trigonometric identity cos2 φ = (1 + cos 2φ)/2, we can show

2 1
E cos (2πfc t + Θ) = E (1 + cos(2π(2fc )t + 2Θ)) = 1/2 (9)
2
Similarly,

1
2
E sin (2πfc t + Θ) = E (1 − cos(2π(2fc )t + 2Θ)) = 1/2 (10)
2
In addition, the identity 2 sin φ cos φ = sin 2φ implies
E [2 sin(2πfc t + Θ) cos(2πfc t + Θ)] = E [cos(4πfc t + 2Θ)] = 0 (11)
Since M (t) and M̂ (t) are independent of Θ, the average power of the upper sideband signal
is

E U 2 (t) = E M 2 (t) E cos2 (2πfc t + Θ) + E M̂ 2 (t) E sin2 (2πfc t + Θ) (12)

− E M (t)M̂ (t) E [2 cos(2πfc t + Θ) sin(2πfc t + Θ)] (13)
= q/2 + q/2 + 0 = q (14)
(a) Since SW (f ) = 10−15 for all f , RW (τ ) = 10−15 δ(τ ).
(b) Since Θ is independent of W (t),
E [V (t)] = E [W (t) cos(2πfc t + Θ)] = E [W (t)] E [cos(2πfc t + Θ)] = 0 (1)
(c) We cannot initially assume V (t) is WSS so we first find
RV (t, τ ) = E[V (t)V (t + τ )] (2)

= E[W (t) cos(2πfc t + Θ)W (t + τ ) cos(2πfc (t + τ ) + Θ)] (3)
= E[W (t)W (t + τ )]E[cos(2πfc t + Θ) cos(2πfc (t + τ ) + Θ)] (4)
−15
= 10 δ(τ )E[cos(2πfc t + Θ) cos(2πfc (t + τ ) + Θ)] (5)
405
We see that for all τ = 0, RV (t, t + τ ) = 0. Thus we need to find the expected value of
E [cos(2πfc t + Θ) cos(2πfc (t + τ ) + Θ)] (6)
only at τ = 0. However, its good practice to solve for arbitrary τ :
E[cos(2πfc t + Θ) cos(2πfc (t + τ ) + Θ)] (7)

1
= E[cos(2πfc τ ) + cos(2πfc (2t + τ ) + 2Θ)] (8)
2
'
1 1 2π 1
= cos(2πfc τ ) + cos(2πfc (2t + τ ) + 2θ) dθ (9)
2 2 0 2π
$2π
1 1 $
= cos(2πfc τ ) + sin(2πfc (2t + τ ) + 2θ)$$ (10)
2 2 0
1 1 1
= cos(2πfc τ ) + sin(2πfc (2t + τ ) + 4π) − sin(2πfc (2t + τ )) (11)
2 2 2
1
= cos(2πfc τ ) (12)
2
Consequently,
1 1
RV (t, τ ) = 10−15 δ(τ ) cos(2πfc τ ) = 10−15 δ(τ ) (13)
2 2
(d) Since E[V (t)] = 0 and since RV (t, τ ) = RV (τ ), we see that V (t) is a wide sense stationary
process. Since L(f ) is a linear time invariant filter, the filter output Y (t) is also a wide sense
stationary process.
(e) The filter input V (t) has power spectral density SV (f ) = 12 10−15 . The filter output has power
spectral density −15
10 /2 |f | ≤ B
SY (f ) = |L(f )| SV (f ) =
2
(14)
0 otherwise
The average power of Y (t) is
' ∞ '
B
1 −15
2
E Y (t) = SY (f ) df = 10 df = 10−15 B (15)
−∞ −B 2

The system described in this problem corresponds exactly to the system in the text that yielded
Equation (11.146).
(a) From Equation (11.146), the optimal linear filter is
SX (f )
Ĥ(f ) = (1)
SX (f ) + SN (f )
In this problem, RX (τ ) = sinc(2W τ ) so that

1 f
SX (f ) = rect . (2)
2W 2W
406
It follows that the optimal filter is

1 f
rect 2W
2W 105 f
Ĥ(f ) = = 5 rect . (3)
1
rect 2Wf
+ 10−5 10 + 2W 2W
2W
(b) From Equation (11.147), the minimum mean square error is

' ∞ ' ∞
SX (f ) SN (f )
e∗L = df = Ĥ(f )SN (f ) df (4)
−∞ SX (f ) + SN (f ) −∞
' W
105
= 5 10−5 df (5)
10 + 2W −W
2W
= 5 . (6)
10 + 2W
It follows that the mean square error satisfies e∗L ≤ 0.04 if and only if W ≤ 2,083.3 Hz. What
is occurring in this problem is the optimal filter is simply an ideal lowpass filter of bandwidth
W . Increasing W increases the bandwidth of the signal and the bandwidth of the filter Ĥ(f ).
This allows more noise to pass through the filter and decreases the quality of our estimator.

The system described in this problem corresponds exactly to the system in the text that yielded
Equation (11.146).
(a) From Equation (11.146), the optimal linear filter is
SX (f )
Ĥ(f ) = (1)
SX (f ) + SN (f )
In this problem, RX (τ ) = e−5000|τ | so that
104
SX (f ) = . (2)
(5,000)2 + (2πf )2
It follows that the optimal filter is
104
(5,000)2 +(2πf )2 109
Ĥ(f ) = 104
= . (3)
+ 10−5 1.025 × 109 + (2πf )2
(5,000)2 +(2πf )2
From Table 11.2, we see that the filter Ĥ(f ) has impulse response
109 −α|τ |
ĥ(τ ) = e (4)
2α
√
where α = 1.025 × 109 = 3.20 × 104 .
(b) From Equation (11.147), the minimum mean square error is
' ∞ ' ∞
SX (f ) SN (f )
e∗L = df = Ĥ(f )SN (f ) df (5)
−∞ SX (f ) + SN (f ) −∞
' ∞
−5
= 10 Ĥ(f ) df (6)
−∞
104
= 10−5 ĥ(0) = = 0.1562. (7)
2α
407
Although it is straightforward to calculate sample paths of Yn using the filter response Yn =
1 1
2 Yn−1 + 2 Xn directly, the necessary loops makes for a slow program. A solution using vectors and
matrices tends to run faster. From the filter response, we can write
1
Y1 = X1 (1)
2
1 1
Y2 = X1 + X2 (2)
4 2
1 1 1
Y3 = X1 + X2 + X3 (3)
8 4 2
..
. (4)
1 1 1
Yn = n
X1 + n−1 X2 + · · · + Xn (5)
2 2 2
In vector notation, these equations become
⎡ ⎤ ⎡ ⎤⎡ ⎤
Y1 1/2 0 ··· 0 X1
⎢ Y2 ⎥ ⎢ . .. ⎥ ⎢ X ⎥
⎢ ⎥ ⎢ 1/4 1/2 . . . ⎥
⎥⎢ 2⎥
⎢ .. ⎥ = ⎢
⎢ ⎥ ⎢ .. ⎥ . (6)
⎣ . ⎦ ⎣ .. . . .. . .. 0 ⎦⎣ . ⎦
Yn 1/2n · · · 1/4 1/2 Xn

Y H X
When X is a column of iid Gaussian (0, 1) random variables, the column vector Y = HX is a single
sample path of Y1 , . . . , Yn . When X is an n × m matrix of iid Gaussian (0, 1) random variables,
each column of Y = HX is a sample path of Y1 , . . . , Yn . In this case, let matrix entry Yi,j denote
a sample Yi of the jth sample path. The samples Yi,1 , Yi,2 , . . . , Yi,m are iid samples of Yi . We can
estimate the mean and variance of Yi using the sample mean Mn (Yi ) and sample variance Vm (Yi )
of Section 7.3. These estimates are
1 1
m m
Mn (Yi ) = Yi,j , V (Yi ) = (Yi,j − Mn (Yi ))2 (7)
m m−1
j=1 j=1
This is the approach of the following program.
function ymv=yfilter(m);
%ymv(i) is the mean and var (over m paths) of y(i),
%the filter output of 11.2.6 and 11.10.1
X=randn(500,m);
H=toeplitz([(0.5).^(1:500)],[0.5 zeros(1,499)]);
Y=H*X;
yav=sum(Y,2)/m;
yavmat=yav*ones(1,m);
yvar=sum((Y-yavmat).^2,2)/(m-1);
ymv=[yav yvar];
The commands ymv=yfilter(100);plot(ymv) will generate a plot similar to this:
408
0.6
Sample mean
0.4 Sample variance
0.2
−0.2
−0.4
0 50 100 150 200 250 300 350 400 450 500
n
We see that each sample mean is small, on the other order of 0.1. Note that E[Yi ] = 0. For m = 100
samples, the sample mean has variance 1/m = 0.01 and standard deviation 0.1. Thus it is to be
expected that we observe sample mean values around 0.1.
Also, it can be shown (in the solution to Problem 11.2.6 for example) that as i becomes large,
Var[Yi ] converges to 1/3. Thus our sample variance results are also not surprising.
Comment: Although within each sample path, Yi and Yi+1 are quite correlated, the sample
means of Yi and Yi+1 are not very correlated when a large number of sample paths are averaged.
Exact calculation of the covariance of the sample means of Yi and Yi+1 might be an interesting
exercise. The same observations apply to the sample variance as well.
This is just a Matlab question that has nothing to do with probability. In the Matlab oper-
ation R=fft(r,N), the shape of the output R is the same as the shape of the input r. If r is
a column vector, then R is a column vector. If r is a row vector, then R is a row vector. For
fftc to work the same way, the shape of n must be the same as the shape of R. The instruction
n=reshape(0:(N-1),size(R)) does this.
The program cospaths.m generates Gaussian sample paths with the desired autocorrelation func-
tion RX (k) = cos(0.04 ∗ pi ∗ k). Here is the code:
function x=cospaths(n,m);
%Generate m sample paths of length n of a
%Gaussian process with ACF R[k]=cos(0.04*pi*k)
k=0:n-1;
rx=cos(0.04*pi*k)’;
x=gaussvector(0,rx,m);
The program is simple because if the second input parameter to gaussvector is a length m vector
rx, then rx is assumed to be the first row of a symmetric Toeplitz covariance matrix. The commands
x=cospaths(100,10);plot(x) will produce a graph like this one:
2
1
Xn
−1
−2
0 10 20 30 40 50 60 70 80 90 100
n
409
We note that every sample path of the process is Gaussian random sequence. However, it would
also appear from the graph that every sample path is a perfect sinusoid. this may seem strange if
you are used to seeing Gaussian processes simply as noisy processes or fluctating Brownian motion
processes. However, in this case, the amplitude and phase of each sample path is random such
that over the ensemble of sinusoidal sample functions, each sample Xn is a Gaussian (0, 1) random
variable.
Finally, to confirm that that each sample path is a perfect sinusoid, rather than just resembling
a sinusoid, we calculate the DFT of each sample path. The commands
>> x=cospaths(100,10);
>> X=fft(x);
>> stem((0:99)/100,abs(X));
will produce a plot similar to this:
120
100
80
Xk
60
40
20
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
k/100
The above plot consists of ten overlaid 100-point DFT magnitude stem plots, one for each Gaussian
sample function. Each plot has exactly two nonzero components at frequencies k/100 = 0.02 and
(100 − k)/100 = 0.98 corresponding to each sample path sinusoid having frequency 0.02. Note that
the magnitude of each 0.02 frequency component depends on the magnitude of the corresponding
sinusoidal sample path.

Searching the Matlab full product help for inv yields this bit of advice:
In practice, it is seldom necessary to form the explicit inverse of a matrix. A frequent

misuse of inv arises when solving the system of linear equations . One way to solve
this is with x = inv(A) ∗ b. A better way, from both an execution time and numerical
accuracy standpoint, is to use the matrix division operator x = A\b. This produces
the solution using Gaussian elimination, without forming the inverse. See \ and / for
further information.
The same discussion goes on to give an example where x = A\b is both faster and more accurate.

The function lmsepredictor.m is designed so that if the sequence Xn has a finite duration autocor-
relation function such that RX [k] = 0 for |k| ≥ m, but the LMSE filter of order M − 1 for M ≥ m
is supposed to be returned, then the lmsepredictor automatically pads the autocorrelation vector
rx with a sufficient number of zeros so that the output is the order M − 1 filter. Conversely, if rx
specifies more values RX [k] than are needed, then the operation rx(1:M) extracts the M values
RX [0], . . . , RX [M − 1] that are needed.
410
However, in this problem RX [k] = (−0.9)|k| has infinite duration. When we pass the truncated
representation rx of length m = 6 and request lmsepredictor(rx,M) for M ≥ 6, the result is that
rx is incorrectly padded with zeros. The resulting filter output will be the LMSE filter for the filter
response
(−0.9)|k| |k| ≤ 5,
RX [k] = (1)
0 otherwise,
rather than the LMSE filter for the true autocorrelation function.

Applying Theorem 11.4 with sampling period Ts = 1/4000 s yields
RX [k] = RX (kTs ) = 10 sinc(0.5k) + 5 sinc(0.25k). (1)
To find the power spectral density SX (φ), we need to find the DTFT of sinc(φ0 k) Unfortunately,
this was omitted from Table 11.2 so we now take a detour and derive it here. As with any derivation
of the transform of a sinc function, we guess the answer and calculate the inverse transform. In
this case, suppose
1 1 |φ| ≤ φ0 /2,
SX (φ) = rect(φ/φ0 ) = (2)
φ0 0 otherwise.
We find RX [k] from the inverse DTFT. For |φ0 | ≤ 1,
' '
1/2
j2πφk 1 φ0 /2
1 ejπφ0 k − e−jπφ0 k
RX [k] = SX (φ) e dφ = ej2πφk dφ = = sinc(φ0 k) (3)
−1/2 φ0 −φ0 /2 φ0 j2πk
Now we apply this result to take the transform of RX [k] in Equation (1). This yields
10 5
SX (φ) = rect(φ/0.5) + rect(φ/0.25). (4)
0.5 0.25
Ideally, an 2N + 1-point DFT would yield a sampled version of the DTFT SX (φ). However, the
truncation of the autocorrelation RX [k] to 201 points results in a difference. For N = 100, the DFT
will be a sampled version of the DTFT of RX [k] rect(k/(2N + 1)). Here is a Matlab program that
shows the difference when the autocorrelation is truncated to 2N + 1 terms.
function DFT=twosincsdft(N);
%Usage: SX=twosincsdft(N);
%Returns and plots the 2N+1
%point DFT of R(-N) ... R(0) ... R(N)
%for ACF R[k] in Problem 11.2.2
k=-N:N;
rx=10*sinc(0.5*k) + 5*sinc(0.25*k);
DFT=fftc(rx);
M=ceil(0.6*N);
phi=(0:M)/(2*N+1);
stem(phi,abs(DFT(1:(M+1))));
xlabel(’\it \phi’);
ylabel(’\it S_X(\phi)’);
Here is the output of twosincsdft(100).
411
50
40
30
SX(φ)
20
10
0
0 0.05 0.1 0.15 0.2 0.25 0.3
φ
From the stem plot of the DFT, it is easy to see the deviations from the two rectangles that make
up the DTFT SX (φ). We see that the effects of windowing are particularly pronounced at the
break points.
Comment: In twosincsdft, DFT must be real-valued since it is the DFT of an autocorrelation
function. Hence the command stem(DFT) should be sufficient. However, due to numerical precision
issues, the actual DFT tends to have a tiny imaginary hence we use the abs operator.

In this problem, we generalize the solution of Problem 11.4.1 using Theorem 11.9 with k = 1 for

filter order M > 2 The optimum linear predictor filter h = h0 h1 · · · hM −1 of Xn+1 given

Xn = Xn−(M −1) · · · Xn−1 Xn is given by
⎡ ⎤
hM −1
←
− ⎢ ⎥
h = ⎣ ... ⎦ = R−1 X RXn Xn+1 ,
n
(1)
h0
where RXn is given by Theorem 11.6 and RXn Xn+1 is given by Equation (11.66). In this problem,
⎡⎡ ⎤ ⎤ ⎡ ⎤
Xn−M +1 RX [M ]
⎢⎢ .. ⎥ ⎥ ⎢ .. ⎥
RXn Xn+1 = E ⎣⎣ . ⎦ Xn+1 ⎦ = ⎣ . ⎦. (2)
Xn RX [1]
Going back to Theorem 9.7(a), the mean square error is

←− ←
−
e∗L = E (Xn+1 − h Xn )2 = Var[Xn+1 ] − h RXn Xn+1 . (3)
For M > 2, we use the Matlab function onesteppredictor(r,M) to perform the calculations.
function [h,e]=onesteppredictor(r,M);
%usage: h=onesteppredictor(r,M);
%input: r=[R_X(0) R_X(1) .. R_X(m-1)]
%assumes R_X(n)==0 for n >=m
%output=vector h for lmse predictor
% xx=h’[X(n),X(n-1),..,X(n-M+1)] for X(n+1)
m=length(r);
r=[r(:);zeros(M-m+1,1)];%append zeros if needed
RY=toeplitz(r(1:M));
RYX=r(M+1:-1:2);
h=flipud(RY\RYX);
e=r(1)-(flipud(h))’*RYX;
412
The code is pretty straightforward. Here are two examples just to show it works.
>> [h2,e2]=onesteppredictor(r,2)
h2 =
0.8571
-0.1429
e2 =
0.4286
>> [h4,e4]=onesteppredictor(r,4)
h4 =
0.8000
0.0000
-0.0000
-0.2000
e4 =
0.4000
The problem also requested that we calculate the mean square error as a function of the filter order
M . Here is a script and the resulting plot of the MSE.
%onestepmse.m 0.44
r=1-0.25*(0:3); 0.42
ee=[ ]; 0.4
for M=2:10,
MSE
0.38
[h,e]=onesteppredictor(r,M); 0.36
ee=[ee,e];
0.34
end
0.32
plot(2:10,ee,’-d’);
xlabel(’\itM’); 0.3
2 4 6 8 10
ylabel(’\it MSE’); M

This problem generalizes
the solution of Problem
11.10.7. to a k-step
predictor. The optimum linear

predictor filter h = h0 h1 · · · hM −1 of Xn+1 given Xn = Xn−(M −1) · · · Xn−1 Xn is
given by ⎡ ⎤
hM −1
←
− ⎢ ⎥
h = ⎣ ... ⎦ = R−1 X RXn Xn+k ,
n
(1)
h0
where RXn is given by Theorem 11.6 and RXn Xn+k is given by Equation (11.66). In this problem,
⎡⎡ ⎤ ⎤ ⎡ ⎤
Xn−M +1 RX [M + k − 1]
⎢⎢ . ⎥ ⎥ ⎢ .. ⎥
RXn Xn+k = E ⎣⎣ .. ⎦ Xn+k ⎦ = ⎣ . ⎦. (2)
Xn RX [k]
Going back to Theorem 9.7(a), the mean square error is

←− ←
−
e∗L = E (Xn+k − h Xn )2 = Var[Xn+k ] − h RXn Xn+k . (3)
The Matlab function kpredictor(r,M) implements this solution.
413
function h=kpredictor(r,M,k);
%usage: h=kpredictor(r,M,k);
%input: r=[R_X(0) R_X(1) .. R_X(m-1)]
%assumes R_X(n)==0 for n >=m
%output=vector a
% for lmse predictor xx=h’[X(n),X(n-1),..,X(n-N+1)] for X(n+k)
m=length(r);
r=[r(:);zeros(M-m+1,1)]; %appends zeros if needed
RY=toeplitz(r(1:M));
RYX=r(1+k:M+k);
h=flipud(RY\RYX);
The code is pretty straightforward.

To generate the sample paths Xn and Yn is relatively straightforward. For σ = η = 1, the solution
to Problem 11.4.5 showed that the optimal linear predictor X̂n (Yn−1 ) of Xn given Yn−1 is
cd
X̂n = Yn−1 . (1)
d2 + (1 − c2 )
In the following, we plot sample paths of Xn and X̂n . To show how sample paths are similar
for different values of c and d, we construct all samples paths using
√ the same sequence of noise
samples Wn and Zn . In addition, given c, we choose X0 = X00 / 1 − c2 where X00 is a Gaussian
(0, 1) random variable that is the same for all sample paths. To do this, we write a function
xpathplot(c,d,x00,z,w) which constructs a sample path given the parameters c and d, and the
noise vectors z and w. A simple script predictpaths.m calls xpathplot to generate the requested
sample paths. Here are the two programs:
function [x,xhat]=xpaths(c,d,x00,z,w) %predictpaths.m
n=length(z); w=randn(50,1);z=randn(50,1);
n0=(0:(n-1))’; n1=(1:n)’; x00=randn(1);
vx=1/(1-c^2); xpaths(0.9,10,x00,z,w);
x0=sqrt(vx)*x00; pause;
x=(c.^(n1)*x0)+... xpaths(0.9,1,x00,z,w);
toeplitz(c.^(n0),eye(1,n))*z; pause;
y=d*[x0;x(1:n-1)]+w; xpaths(0.9,0.1,x00,z,w);
vy=((d^2)*vx) + 1; pause;
xhat=((c*d*vx)/vy)*y; xpaths(0.6,10,x00,z,w);
plot(n1,x,’b-’,n1,xhat,’k:’); pause;
axis([0 50 -3 3]); xpaths(0.6,1,x00,z,w);
xlabel(’\itn’); ylabel(’\itX_n’); pause;
xpaths(0.6,0.1,x00,z,w);
Some sample paths for Xn and X̂n for the requested parameters are shown below. In each pair,
the one-step prediction X̂n is marked by dots.
2 2
Xn
0 0
X
−2 −2
0 10 20 30 40 50 0 10 20 30 40 50
n n
(a) c = 0.9, d = 10 (d) c = 0.6, d = 10
414
2 2
Xn
n
0 0
X
−2 −2
0 10 20 30 40 50 0 10 20 30 40 50
n n
(b) c = 0.9, d = 1 (e) c = 0.6, d = 1
2 2
Xn
n
0 0
X
−2 −2
0 10 20 30 40 50 0 10 20 30 40 50
n n
(c) c = 0.9, d = 0.1 (f ) c = 0.6, d = 0.1
The mean square estimation error at step n was found to be
d2 + 1
e∗L (n) = e∗L = σ 2 (2)
d2 + (1 − c2 )
We see that the mean square estimation error is e∗L (n) = e∗L , a constant for all n. In addition, e∗L
is a decreasing function of d. In graphs (a) through (c), we see that the predictor tracks Xn less
well as d decreases because decreasing d corresponds to decreasing the contribution of Xn−1 to the
measurement Yn−1 . Effectively, the impact of the measurement noise is increased. As d decreases,
the predictor places less emphasis on the measurement Yn and instead makes predictions closer to
E[X] = 0. That is, when d is small in graphs (c) and (f), the predictor stays close to zero. With
respect to c, the performance of the predictor is less easy to understand. In Equation (11), the
mean square error e∗L is the product of
σ2 (d2 + 1)(1 − c2 )
Var[Xn ] = 1 − ρ2Xn ,Yn−1 = (3)
1 − c2 d2 + (1 − c2 )
As a function of increasing c2 , Var[Xn ] increases while 1 − ρ2Xn ,Yn−1 decreases. Overall, the mean
square error e∗L is an increasing function of c2 . However, Var[X] is the mean square error obtained
using a blind estimator that always predicts E[X] while 1 − ρ2Xn ,Yn−1 characterizes the extent to
which the optimal linear predictor is better than the blind predictor. When we compare graphs
(a)-(c) with c = 0.9 to graphs (d)-(f) with c = 0.6, we see greater variation in Xn for larger a but
in both cases, the predictor worked well when d was large.
415

From the given Markov chain, the state transition matrix is
⎡ ⎤ ⎡ ⎤
P00 P01 P02 0.5 0.5 0
P = ⎣P10 P11 P12 ⎦ = ⎣ 0.5 0.5 0⎦ (1)
P20 P21 P22 0.25 0.25 0.5

This problem is very straightforward if we keep in mind that Pij is the probability that we transition
from state i to state j. From Example 12.1, the state transition matrix is

P P01 1−p p
P = 00 = (1)
P10 P11 q 1−q

In addition to the normal OFF and ON states for packetized voice, we add state 2, the “mini-OFF”
state. The Markov chain is
P00 P11 P22
P01 P12
0 1 2
P10 P21
P20
The only difference between this chain and an arbitrary 3 state chain is that transitions from 0,
the OFF state, to state 2, the mini-OFF state, are not allowed. From the problem statement, the
corresponding Markov chain is
⎡ ⎤ ⎡ ⎤
P00 P01 P02 0.999929 0.000071 0
P = ⎣P10 P11 P12 ⎦ = ⎣0.000100 0.899900 0.1⎦ . (1)
P20 P21 P22 0.000100 0.699900 0.3

Based on the problem statement, the state of the wireless LAN is given by the following Markov
chain:
0.5 0.9 0.9 0.9
0.5 0.06 0.06
0 1 2 3
0.04 0.04
0.04 0.02
0.04
The Markov chain has state transition matrix
⎡ ⎤
0.5 0.5 0 0
⎢0.04 0.9 0.06 0 ⎥
P=⎢ ⎣0.04
⎥. (1)
0 0.9 0.06⎦
0.04 0.02 0.04 0.9
416
In this problem, it is helpful to go fact by fact to identify the information given.
• “. . . each read or write operation reads or writes an entire file and that files contain a geometric
number of sectors with mean 50.”
This statement says that the length L of a file has PMF

(1 − p)l−1 p l = 1, 2, . . .
PL (l) = (1)
0 otherwise
with p = 1/50 = 0.02. This says that when we write a sector, we will write another sector
with probability 49/50 = 0.98. In terms of our Markov chain, if we are in the write state, we
write another sector and stay in the write state with probability P22 = 0.98. This fact also
implies P20 + P21 = 0.02.
Also, since files that are read obey the same length distribution,
P11 = 0.98 P10 + P12 = 0.02 (2)
• “Further, suppose idle periods last for a geometric time with mean 500.”
This statement simply says that given the system is idle, it remains idle for another unit of
time with probability P00 = 499/500 = 0.998. This also says that P01 + P02 = 0.002.
• “After an idle period, the system is equally likely to read or write a file.”
Given that at time n, Xn = 0, this statement says that the conditional probability that
P01
P [Xn+1 = 1|Xn = 0, Xn+1 = 0] = = 0.5 (3)
P01 + P02
Combined with the earlier fact that P01 + P02 = 0.002, we learn that
P01 = P02 = 0.001 (4)
• “Following the completion of a read, a write follows with probability 0.8.”

Here we learn that given that at time n, Xn = 1, the conditional probability that
P12
P [Xn+1 = 2|Xn = 1, Xn+1 = 1] = = 0.8 (5)
P10 + P12
P10 = 0.004 P12 = 0.016 (6)
• “However, on completion of a write operation, a read operation follows with probability 0.6.”
Now we find that given that at time n, Xn = 2, the conditional probability that
P21
P [Xn+1 = 1|Xn = 2, Xn+1 = 2] = = 0.6 (7)
P20 + P21
P20 = 0.008 P21 = 0.012 (8)
417
The complete tree is
0.998 0.98 0.98
0.001 0.016
0 1 2
0.004 0.012
0.001
0.008

To determine whether the sequence X̂n forms a Markov chain, we write

P X̂n+1 = j|X̂n = i, X̂n−1 = in−1 , . . . , X̂0 = i0

= P Xm(n+1) = j|Xmn = i, Xm(n−1) = in−1 , . . . , X0 = i0 (1)

= P Xm(n+1) = j|Xmn = i (2)
= Pij (m) (3)
The key step is in observing that the Markov property of Xn implies that Xmn summarizes the
past history of the Xn process. That is, given Xmn , Xm(n+1) is independent of Xmk for all k < n.
Finally, this implies that the state X̂n has one-step state transition probabilities equal to the
m-step transition probabilities for the Markov chain Xn . That is, P̂ = Pm .

Under construction.

Under construction.

From Example 12.1, the state transition matrix is

P00 P01 1−p p 0.2 0.8
P= = = . (1)
P10 P11 q 1−q 0.9 0.1
This chain is a special case of the two-state chain in Example 12.5 and Example 12.6 with p = 0.8
and q = 0.9. You may wish to derive the eigenvalues and eigenvectors of P in order to diagonalize
and then find Pn . Or, you may wish just to refer to Example 12.6 which showed that the chain
has eigenvalues λ1 = 1 and λ2 = 1 − (p + q) = −0.7 and that the n-step transition matrix is

n P00 (n) P01 (n) 1 q p λn2 p −p
P = = + (2)
P10 (n) P11 (n) p+q q p p + q −q q

1 0.9 0.8 (−0.7)n 0.8 −0.8
= + . (3)
1.7 0.9 0.8 1.7 −0.9 0.9
418
From the given Markov chain, the state transition matrix is
⎡ ⎤ ⎡ ⎤
P00 P01 P02 0.5 0.5 0
P = ⎣P10 P11 P12 ⎦ = ⎣ 0.5 0.5 0⎦ (1)
P20 P21 P22 0.25 0.25 0.5
The way to find Pn is to make the decomposition P = SDS−1 where the columns of S are the
eigenvectors of P and D is a diagonal matrix containing the eigenvalues of P. The eigenvalues are
λ1 = 1 λ2 = 0 λ3 = 1/2 (2)
The corresponding eigenvectors are
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
1 −1 0
s1 = ⎣1⎦ s2 = ⎣ 1 ⎦ s3 = ⎣0⎦ (3)
1 0 1
The decomposition of P is
⎡ ⎤⎡ ⎤⎡ ⎤
1 −1 0 1 0 0 0.5 0.5 0
P = SDS−1 = ⎣1 1 0⎦ ⎣0 0 0 ⎦ ⎣−0.5 0.5 0⎦ (4)
1 0 1 0 0 0.5 −0.5 −0.5 1
Finally, Pn is
⎡ ⎤⎡ ⎤⎡ ⎤
1 −1 0 1 0 0 0.5 0.5 0
Pn = SDn S−1 = ⎣1 1 0⎦ ⎣0 0 0 ⎦ ⎣−0.5 0.5 0⎦ (5)
1 0 1 0 0 (0.5)n −0.5 −0.5 1
⎡ ⎤
0.5 0.5 0
= ⎣ 0.5 0.5 0 ⎦ (6)
0.5 − (0.5)n+1 0.5 − (0.5)n+1 (0.5)n

From Example 12.8, the state probabilities at time n are
7 −5

12 + λ2 12 p0 − 12 p1
5 n 5 7 7
p(n) = 12 12 p0
+ 12 p1 (1)

with λ2 = 1 − (p + q) = 344/350. With initial state probabilities p0 p1 = 1 0 ,
7 5

n 5 −5

p(n) = 12 12 + λ2 12 12 (2)

The limiting state probabilities are π0 π1 = 7/12 5/12 . Note that pj (n) is within 1% of πj if
|πj − pj (n)| ≤ 0.01πj (3)
These requirements become
7/12
λn2 ≤ 0.01 λn2 ≤ 0.01 (4)
5/12
The minimum value n that meets both requirements is
% &
ln 0.01
n= = 267 (5)
ln λ2
Hence, after 267 time steps, the state probabilities are all within one percent of the limiting state
probability vector. Note that in the packet voice system, the time step corresponded to a 10 ms
time slot. Hence, 2.67 seconds are required.
419
At time n − 1, let pi (n − 1) denote the state probabilities. By Theorem 12.4, the probability of
state k at time n is
∞

pk (n) = pi (n − 1)Pik (1)
i=0
Since Pik = q for every state i,
∞

pk (n) = q pi (n − 1) = q (2)
i=0
Thus for any time n > 0, the probability of state k is q.
In this problem, the arrivals are the occurrences of packets in error. It would seem that N (t) cannot
be a renewal process because the interarrival times seem to depend on the previous interarrival
times. However, following a packet error, the sequence of packets that are correct (c) or in error
(e) up to and including the next error is given by the tree
e •X=1 e •X=2 e •X=3

0.9 0.01 0.01

c c c ...
0.1 0.99 0.99
Assuming that sending a packet takes one unit of time, the time X until the next packet error has
the PMF ⎧
⎨ 0.9 x=1
PX (x) = 0.001(0.99)x−2 x = 2, 3, . . . (1)
⎩
0 otherwise
Thus, following an error, the time until the next error always has the same PMF. Moreover, this
time is independent of previous interarrival times since it depends only on the Bernoulli trials
following a packet error. It would appear that N (t) is a renewal process; however, there is one
additional complication. At time 0, we need to know the probability p of an error for the first
packet. If p = 0.9, then X1 , the time until the first error, has the same PMF as X above and
the process is a renewal process. If p = 0.9, then the time until the first error is different from
subsequent renewal times. In this case, the process is a delayed renewal process.
The hardest part of this problem is that we are asked to find all ways of replacing a branch. The
primary problem with the Markov chain in Problem 12.1.1 is that state 2 is a transient state. We
can get rid of the transient behavior by making a nonzero branch probability P12 or P02 . The
possible ways to do this are:
• Replace P00 = 1/2 with P02 = 1/2
• Replace P01 = 1/2 with P02 = 1/2
• Replace P11 = 1/2 with P12 = 1/2
• Replace P10 = 1/2 with P12 = 1/2
Keep in mind that even if we make one of these replacements, there will be at least one self transition
probability, either P00 or P11 , that will be nonzero. This will guarantee that the resulting Markov
chain will be aperiodic.
420
The chain given in Example 12.11 has two communicating classes as well as the transient state 2. To
create a single communicating class, we need to add a transition that enters state 2. Yet, no matter
how we add such a transition, we will still have two communicating classes. A second transition
will be needed to create a single communicating class. Thus, we need to add two branches. There
are many possible pairs of branches. Some pairs of positive branch probabilities that create an
irreducible chain are
{P12 , P23 } {P50 , P02 } {P51 , P02 } (1)

The idea behind this claim is that if states j and i communicate, then sometimes when we go from
state j back to state j, we will pass through state i. If E[Tij ] = ∞, then on those occasions we pass
through i, the expected time to go to back to j will be infinite. This would suggest E[Tjj ] = ∞
and thus state j would not be positive recurrent. Using a math to prove this requires a little bit of
care.
Suppose E[Tij ] = ∞. Since i and j communicate, we can find n, the smallest nonnegative
integer such that Pji (n) > 0. Given we start in state j, let Gi denote the event that we go through
state i on our way back to j. By conditioning on Gj ,
E [Tjj ] = E [Tjj |Gi ] P [Gi ] + E [Tjj |Gci ] P [Gci ] (1)
Since E[Tjj |Gci ]P [Gci ] ≥ 0,

E [Tjj ] ≥ E [Tjj |Gi ] P [Gi ] (2)
Given the event Gi , Tjj = Tji + Tij . This implies
E [Tjj |Gi ] = E [Tji |Gi ] + E [Tij |Gi ] ≥ E [Tij |Gi ] (3)
Since the random variable Tij assumes that we start in state i, E[Tij |Gi ] = E[Tij ]. Thus E[Tjj |Gi ] ≥
E[Tij ]. In addition, P [Gi ] ≥ Pji (n) since there may be paths with more than n hops that take the
system from state j to i. These facts imply
E [Tjj ] ≥ E [Tjj |Gi ] P [Gi ] ≥ E [Tij ] Pji (n) = ∞ (4)
Thus, state j is not positive recurrent, which is a contradiction. Hence, it must be that E[Tij ] < ∞.

In the solution to Problem 12.1.5, we found that the state transition matrix was
⎡ ⎤
0.998 0.001 0.001
P = ⎣0.004 0.98 0.016⎦ (1)
0.008 0.012 0.98

We can find the stationary probability vector π = π0 π1 π2 by solving π = π P along with
π0 + π1 + π2 = 1. It’s possible to find the solution by hand but its easier to use MATLAB or a
similar tool. The solution is π = [0.7536 0.1159 0.1304].
421
From the Markov chain given in Problem 12.1.1, the state transition matrix is
⎡ ⎤
0.5 0.5 0
P = ⎣ 0.5 0.5 0⎦ (1)
0.25 0.25 0.5

We find the stationary probabilities π = π0 π1 π2 by solving

π = πP 2πj = 1 (2)
j=0
Of course, one equation of π = π P will be redundant. The three independent equations are
π0 = 0.5π0 + 0.5π1 + 0.25π2 (3)

π2 = 0.5π2 (4)
1 = π0 + π 1 + π 2 (5)
From the second equation, we see that π2 = 0. This leaves the two equations:
π0 = 0.5π0 + 0.5π1 (6)

1 = π0 + π 1 (7)
Solving these two equations yields π0 = π1 = 0.5. The stationary probability vector is

π = π0 π1 π2 = 0.5 0.5 0 (8)
If you happened to solve Problem 12.2.2, you would have found that the n-step transition matrix
is ⎡ ⎤
0.5 0.5 0
Pn = ⎣ 0.5 0.5 0 ⎦ (9)
0.5 − (0.5)n+1 0.5 − (0.5)n+1 (0.5)n
From Theorem 12.21, we know that each rows of the n-step transition matrix converges to π . In
this case,
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
0.5 0.5 0 0.5 0.5 0 π
n
lim P = lim ⎣ 0.5 0.5 0 ⎦ ⎣
= lim 0.5 0.5 0 = π⎦
⎦ ⎣ (10)
n→∞ n→∞ n→∞
0.5 − (0.5)n+1 0.5 − (0.5)n+1 (0.5)n 0.5 0.5 0 π

From the problem statement, the Markov chain is
1-p p
p p p p
0 1 2 3 4
1-p 1-p 1-p 1-p
422
The self-transitions in state 0 and state 4 guarantee that the Markov chain is aperiodic. Since the
chain is also irreducible, we can find the stationary probabilities by solving π = π P; however, in
this problem it is simpler to apply Theorem 12.13. In particular, by partitioning the chain between
states i and i + 1, we obtain
πi p = πi+1 (1 − p). (1)
This implies πi+1 = απi where α = p/(1 − p). It follows that πi = αi π0 . REquiring the stationary
probabilities to sum to 1 yields

4
πi = π0 (1 + α + α2 + α3 + α4 ) = 1. (2)
i=0
This implies
1 − α5
π0 = (3)
1−α
Thus, for i = 0, 1, . . . , 4,
5
1− p i
1− α5 1−p p
πi = αi = . (4)
1−α 1 − 1−p
p 1−p

From the problem statement, the Markov chain is
1-p p
p p p p
0 1 i i+1 K
1-p 1-p 1-p 1-p
The self-transitions in state 0 and state K guarantee that the Markov chain is aperiodic. Since the
chain is also irreducible, we can find the stationary probabilities by solving π = π P; however, in
this problem it is simpler to apply Theorem 12.13. In particular, by partitioning the chain between
states i and i + 1, we obtain
πi p = πi+1 (1 − p). (1)
This implies πi+1 = απi where α = p/(1 − p). It follows that πi = αi π0 . REquiring the stationary
probabilities to sum to 1 yields

K
πi = π0 (1 + α + α2 + · · · + αK ) = 1. (2)
i=0
This implies
1 − αK+1
π0 = (3)
1−α
Thus, for i = 0, 1, . . . , K,
K+1
1− p i
1− αK+1 1−p p
πi = αi = . (4)
1−α 1− p 1−p
1−p
423
For this system, it’s hard to draw the entire Markov chain since from each state n there are six
branches, each with probability 1/6 to states n + 1, n + 2, . . . , n + 6. (Of course, if n + k > K − 1,
then the transition is to state n + k mod K.) Nevertheless, finding the stationary probabilities is
not very hard. In particular, the nth equation of π = π P yields
1
πn = (πn−6 + πn−5 + πn−4 + πn−3 + πn−2 + πn−1 ) . (1)
6
Rather than try to solve these equations algebraically, it’s easier to guess that the solution is

π = 1/K 1/K · · · 1/K . (2)
It’s easy to check that 1/K = (1/6) · 6 · (1/K)

This system has three states:
0 front teller busy, rear teller idle
1 front teller busy, rear teller busy
2 front teller idle, rear teller busy
We will assume the units of time are seconds. Thus, if a teller is busy one second, the teller will
become idle in th next second with probability p = 1/120. The Markov chain for this system is
1-p p2+(1-p)2 1-p
1-p p(1-p)
0 1 2
p(1-p) p
We can solve this chain very easily for the stationary probability vector π. In particular,
π0 = (1 − p)π0 + p(1 − p)π1 (1)
This implies that π0 = (1 − p)π1 . Similarly,
π2 = (1 − p)π2 + p(1 − p)π1 (2)
yields π2 = (1 − p)π1 . Hence, by applying π0 + π1 + π2 = 1, we obtain

1−p
π0 = π2 = = 119/358 (3)
3 − 2p
1
π1 = = 120/358 (4)
3 − 2p
The stationary probability that both tellers are busy is π1 = 120/358.
424
In this case, we will examine the system each minute. For each customer in service, we need to
keep track of how soon the customer will depart. For the state of the system, we will use (i, j), the
remaining service requirements of the two customers, To reduce the number of states, we will order
the requirements so that i ≤ j. For example, when two new customers start service each requiring
two minutes of service, the system state will be (2, 2). Since the system assumes there is always a
backlog of cars waiting to enter service, the set of states is
0 (0, 1) One teller is idle, the other teller has a customer requiring one more minute of service
1 (1, 1) Each teller has a customer requiring one more minute of service.
2 (1, 2) One teller has a customer requring one minute of service. The other teller has a customer
requiring two minutes of service.
3 (2, 2) Each teller has a customer requiring two minutes of service.

¼
The resulting Markov chain is shown on the right. Note that when we departing 3 0
from either state (0, 1) or (1, 1) corresponds to both custoemrs finishing service (2,2) (0,1)
¼
and two new customers entering service. The state transiton probabilities reflect
1 ¼ 1 ½
the fact that both customer will have two minute service requirements with
probability 1/4, or both customers will hae one minute service requirements 1 2
(1,1) (1,2)
with probability 1/4, or one customer will need one minute of service and the
½
other will need two minutes of service with probability 1/2.
¼
Writing the stationary probability equations for states 0, 2, and 3 and adding the constraint
j πj = 1 yields the following equations:
π0 = π2 (1)
π2 = (1/2)π0 + (1/2)π1 (2)
π3 = (1/4)π0 + (1/4)π1 (3)
1 = π0 + π 1 + π 2 + π 3 (4)
Substituting π2 = π0 in the second equation yields π1 = π0 . Substituting that result in the third
equation yields π3 = π0 /2. Making sure the probabilities add up to 1 yields

π = π0 π1 π2 π3 = 2/7 2/7 2/7 1/7 . (5)
Both tellers are busy unless the system is in state 0. The stationary probability both tellers are
busy is 1 − π0 = 5/7.

Under construction.

Under construction.
425
Equivalently, we can prove that if Pii = 0 for some i, then the chain cannot be
periodic. So, suppose for state i, Pii > 0. Since Pii = Pii (1), we see that the
0.5
largest d that divides n for all n such that Pii (n) > 0 is d = 1. Hence, state i is
0 1
aperiodic and thus the chain is aperiodic.
0.5
The converse that Pii = 0 for all i implies the chain is periodic is false. As a 0.5
0.5
0.5 0.5
counterexample, consider the simple chain on the right with Pii = 0 for each i.
Note that P00 (2) > 0 and P00 (3) > 0. The largest d that divides both 2 and 3 2
is d = 1. Hence, state 0 is aperiodic. Since the chain has one communicating
class, the chain is also aperiodic.

The Markov chain for this system is
P[N >1|N >0] P[N >2|N >1] P[N >3|N >2] P[N >K-1|N >K-2] P[N >K|N >K-1]
0 1 2 …K-1 K
P[N=2|N >1]
P[N=3|N >2]
P[N=K|N >K-1]
P[N=K+1|N >K]
Note that P [N > 0] = 1 and that
P [N > n, N > n − 1] P [N > n]

P [N > n|N > n − 1] = = . (1)
P [N > n − 1] P [N > n − 1]
Solving π = π P yields
π1 = P [N > 1|N > 0] π0 = P [N > 1] π0 (2)

P [N > 2]
π2 = P [N > 2|N > 1] π1 = π1 = P [N > 2] π0 (3)
P [N > 1]
..
.
P [N > n]
πn = P [N > n|N > n − 1] πn−1 = πn−1 = P [N > n] π0 (4)
P [N > n − 1]
Next we apply the requirement that the stationary probabilities sum to 1. Since P [N ≤ K + 1] = 1,
we see for n ≥ K + 1 that P [N > n] = 0. Thus

K
K ∞

1= πn = π0 P [N > n] = π0 P [N > n] . (5)
n=0 n=0 n=0

∞
From Problem 2.5.11, we recall that n=0 P [N > n] = E[N ]. This implies π0 = 1/E[N ] and that
P [N > n]
πn = . (6)
E [N ]
This is exactly the same stationary distribution found in Quiz 12.5! In this problem, we can view
the system state as describing the age of an object that is repeatedly replaced. In state 0, we start
with a new (zero age) object, and each unit of time, the object ages one unit of time. The random
426
variable N is the lifetime of the object. A transition to state 0 corresponds to the current object
expiring and being replaced by a new object.
In Quiz 12.5, the system state described a countdown timer for the residual life of an object. At
state 0, the system would transition to a state N = n corresponding to the lifetime of n for a new
object. This object expires and is replaced each time that state 0 is reached. This solution and the
solution to Quiz 12.5 show that the age and the residual life have the same stationary distribution.
That is, if we inspect an object at an arbitrary time in the distant future, the PMF of the age of
the object is the same as the PMF of the residual life.

Consider an arbitrary state i ∈ Cl . To show that state i has period L, we must show that Pii (n) > 0
implies n is divisible by L. Consider any sequence of states i, i1 , i2 , . . . , in = i that returns to state i
in n steps. The structure of the Markov chain implies that i1 ∈ Cl+1 , i2 ∈ Cl+2 and so on. In general,
ij ∈ C(l+j) mod L . Moreover, since state i ∈ Cl , in = i only if in ∈ Cl . Since in ∈ C(l+n) mod L , we
must have that (l + n) mod L = l. This occurs if and only if n is divisible by L.

This problem is easy as long as you don’t try to prove that the chain must have a single communi-
cating class. A counterexample is easy to find. Here is a 4 state Markov chain with two recurrent
communicating classes and each state having period 2. The two classes C0 and C1 are shown.
C0 C1
0 1
2 3
From each state i ∈ C0 , all transitions are to states j ∈ C1 . Similarly, from each state i ∈ C1 ,
only transitions to states j ∈ C0 are permitted. The sets {0, 1} and {2, 3} are each communicating
classes. However, each state has period 2.

Since the Markov chain has period d, there exists a state i0 such that Pi0 i0 (d) > 0. For this state i0 ,
let Cn (i0 ) denote the set of states that reachable from i0 in n hops. We make the following claim:
• If j ∈ Cn (i0 ) and j ∈ Cn (i0 ) with n > n, then n = n + kd for some integer k
To prove this claim, we observe that irreducibility of the Markov chain implies there is a sequence
of m hops from state j to state i0 . Since there is an n hop path from i0 to j and an m hop path
from j back to i0 , n + m = kd for some integer k. Similarly, since there is an n hop path from i0
to j, n + m = k d for some integer k . Thus
n − n = (n + m) − (n + m) = (k − k)d. (1)
Now we define
∞
=
Cn = Cn+kd (i0 ), n = 0, 1, . . . , d − 1. (2)
k=0
Because the chain is irreducible, any state i belongs to some set Cn (i0 ), and thus any state i
belongs to at least one set Cn . By our earlier claim, each node i belongs to exactly one set Cn .
427
Hence, {C0 , . . . , Cd−1 } is a partition of the states of the states of the Markov chain. By construction
of the set Cn (i0 ), there exists states i ∈ Cn (i0 ) and j ∈ Cn+1 (i0 ) such that Pij > 0. Hence there
exists i ∈ Cn and j ∈ Cn+1 such that Pij > 0.
Now suppose there exists states i ∈ Cn and j ∈ Cn+m such that Pij > 0 and m > 1. In this
case, the sequence of n + kd hops from i0 to i followed by one hop to state j is an n + 1 + kd hop
path from i0 to j, implying j ∈ Cn+1 . This contradicts the fact that j cannot belong to both Cn+1
and Cn+m . Hence no such transition from i ∈ Cn to j ∈ Cn+m is possible.

A careful reader of the text will observe that the Markov chain in this problem is identical to the
Markov chain introduced in Example 12.21:
1-p
p p p
0 1 2
1-p 1-p 1-p
The stationary probabilities were found in Example 12.26 to be
πi = (1 − α)αi , i = 0, 1, 2, . . . (1)
where α = p/(1 − p). Note that the stationary probabilities do not exist if α ≥ 1, or equivalently,
p ≥ 1/2.

If there are k customers in the system at time n, then at time n + 1, the number of customers in
the system is either n − 1 (if the customer in service departs and no new customer arrives), n (if
either there is no new arrival and no departure or if there is both a new arrival and a departure)
or n + 1, if there is a new arrival but no new departure. The transition probabilities are given by
the following chain:
1-p
p
0 1 2

where α = p(1 − q) and δ = q(1 − p). To find the stationary probabilities, we apply Theorem 12.13
by partitioning the state space between states S = {0, 1, . . . , i} and S = {i + 1, i + 2, . . .} as shown
in Figure 12.4. By Theorem 12.13, for state i > 0,
πi α = πi+1 δ. (1)
This implies πi+1 = (α/δ)πi . A cut between states 0 and 1 yields π1 = (p/δ)π0 . Combining these
results, we have for any state i > 0,
p α i−1
πi = π0 (2)
δ δ
Under the condition α < δ, it follows that

p α i−1
∞
∞
p/δ
πi = π0 + π0 = π0 1+ (3)
δ δ 1 − α/δ
i=0 i=1
428

since p < q implies α/δ < 1. Thus, applying i πi = 1 and noting δ − α = q − p, we have

q p p/(1 − p) i−1
π0 = , πi = , i = 1, 2, . . . (4)
q−p (1 − p)(1 − q) q/(1 − q)
Note that α < δ if and only if p < q, which is both sufficient and necessary for the Markov chain
to be positive recurrent.

From the problem statement, we learn that in each state i, the tiger spends an exponential time
with parameter λi . When we measure time in hours,
λ0 = q01 = 1/3 λ1 = q12 = 1/2 λ2 = q20 = 2 (1)
The corresponding continous time Markov chain is shown below:

1/3
0 1
2 ½
The state probabilities satisfy

1 1 1
p0 = 2p2 p1 = p0 p0 + p1 + p2 = 1 (2)
3 2 3
The solution is
p0 p1 p2 = 6/11 4/11 1/11 (3)

In the continuous time chain, we have states 0 (silent) and 1 (active). The transition rates and the
chain are
0.71
0 1 1
q01 = = 0.7143 q10 = 1.0 (1)
1.4
1
The stationary probabilities satisfy (1/1.4)p0 = p1 . Since p0 + p1 = 1, the stationary probabilities
are
1.4 7 1 5
p0 = = p1 = = (2)
2.4 12 2.4 12
In this case, the continuous time chain and the discrete time chain have the exact same state
probabilities. In this problem, this is not surprising since we could use a renewal-reward process
to calculate the fraction of time spent in state 0. From the renewal-reward process, we know that
the fraction of time spent in state 0 depends only on the expected time in each state. Since in
both the discrete time and continuous time chains, the expected time in each state is the same, the
stationary probabilties must be the same. It is always possible to approximate a continuous time
chain with a discrete time chain in which the unit of time is chosen to be very small. In general
however, the stationary probabilities of the two chains will be close though not identical.
429
From each state i, there are transitions of rate qij = 1 to each of the other k − 1 states. Thus each
state i has departure rate νi = k − 1. Thus, the stationary probabilities satisfy

pj (k − 1) = pj j = 1, 2, . . . , k (1)
i
=j
It is easy to verify that the solution to these equations is

1
pj = j = 1, 2, . . . , k (2)
k

In this problem, we build a two-state Markov chain such that the system in state i ∈ {0, 1} if
the most recent arrival of either Poisson process is type i. Note that if the system is in state 0,
transitions to state 1 occur with rate λ1 . If the system is in state 1, transitions to state 0 occur at
rate λ0 . The continuous time Markov chain is just

0 1

The stationary probabilities satisfy p0 λ1 = p1 λ0 . Thus p1 = (λ1 /λ0 )p0 . Since p0 + p1 = 1, we have
that
p0 + (λ1 /λ0 )p0 = 1. (1)
This implies
λ0 λ1
p0 = , p1 = . (2)
λ0 + λ1 λ0 + λ1
It is also possible to solve this problem using a discrete time Markov chain. One way to do this is
to assume a very small time step ∆. In state 0, a transition to state 1 occurs with probability λ1 ∆;
otherwise the system stays in state 0 with probability 1 − λ1 ∆. Similarly, in state 1, a transition to
state 0 occurs with probability λ0 ∆; otherwise the system stays in state 1 with probability 1 − λ0 ∆.
Here is the Markov chain for this discrete time system:

0 1

Not surprisingly, the stationary probabilities for this discrete time system are
λ0 λ1
π0 = , π1 = . (3)
λ0 + λ1 λ0 + λ1

In Equation (12.93), we found that the blocking probability of the M/M/c/c queue was given by
the Erlang-B formula
ρc /c!
P [B] = PN (c) =
c k
(1)
k=0 ρ /k!
430
The parameter ρ = λ/µ is the normalized load. When c = 2, the blocking probability is
ρ2 /2
P [B] = (2)
1 + ρ + ρ2 /2
Setting P [B] = 0.1 yields the quadratic equation
2 2
ρ2 − ρ − = 0 (3)
9 9
The solutions to this quadratic are √
1±
19
ρ= (4)
9
√
The meaningful nonnegative solution is ρ = (1 + 19)/9 = 0.5954.

When we double the call arrival rate, the offered load ρ = λ/µ doubles to ρ = 160. However, since
we double the number of circuits to c = 200, the offered load per circuit remains the same. In this
case, if we inspect the system after a long time, the number of calls in progress is described by the
stationary probabilities and has PMF

ρn /n!

200 k n = 0, 1, . . . , 200
PN (n) = k=0 ρ /k! (1)
0 otherwise
The probability a call is blocked is

ρ200 /200!
P [B] = PN (200) =
200 = 2.76 × 10−4 (2)
k
k=0 ρ /k!
Note that although the load per server remains the same, doubling the number of circuits to 200
caused the blocking probability to go down by more than a factor of 10 (from 0.004 to 2.76 × 10−4 ).
This is a general property of the Erlang-B formula and is called trunking efficiency by telephone
system engineers. The basic principle ss that it’s more efficient to share resources among larger
groups.
The hard part of calculating P [B] is that most calculators, including MATLAB have trouble
calculating 200!. (In MATLAB, factorial is calculated using the gamma function. That is, 200! =
gamma(201).) To do these calculations, you need to observe that if qn = ρn /n!, then
ρ
qn+1 = qn−1 (3)
n
A simple MATLAB program that uses this fact to calculate the Erlang-B formula for large values
of c is
function y=erlangbsimple(r,c)
%load is r=lambda/mu
%number of servers is c
p=1.0;
psum=1.0;
for k=1:c
p=p*r/k;
psum=psum+p;
end
y=p/psum;
431
Essentially the problems with the calculations of erlangbsimple.m are the same as those of cal-
culating the Poisson PMF. A better program for calculating the Erlang-B formula uses the im-
provements employed in poissonpmf to calculate the Poisson PMF for large values. Here is the
code:
function pb=erlangb(rho,c);
%Usage: pb=erlangb(rho,c)
%returns the Erlang-B blocking
%probability for sn M/M/c/c
%queue with load rho
pn=exp(-rho)*poissonpmf(rho,0:c);
pb=pn(c+1)/sum(pn);

In the M/M/1/c queue, there is one server and the system has capacity c. That is, in addition to
the server, there is a waiting room that can hold c − 1 customers. With arival rate λ and service
rate µ, the Markov chain for this queue is
λ λ λ λ
0 1 c-1 c
µ µ µ µ
By Theorem 12.24, the stationary probabilities satisfy pi−1 λ = pi µ. By defining ρ = λ/mu, we

have pi = ρpi−1 , which implies
pn = ρn p0 n = 0, 1, . . . , c (1)
applying the requirement that the stationary probabilities sum to 1 yields

c

pi = p0 1 + ρ + ρ2 + · · · + ρc = 1 (2)
i=0
This implies
1−ρ
p0 = (3)
1 − ρc+1
The stationary probabilities are
(1 − ρ)ρn
pn = n = 0, 1, . . . , c (4)
1 − ρc+1

Since the arrivals are a Poisson process and since the service requirements are exponentially dis-
tributed, the set of toll booths are an M/M/c/∞ queue. With an arrival rate of λ and a service
rate of µ = 1, the Markov chain is:
λ λ λ λ λ
0 1 c c+1
µ 2µ cµ cµ cµ
432
In the solution to Quiz 12.10, we found that the stationary probabilities for the queue satisfied

p0 ρn /n! n = 1, 2, . . . , c
pn = n−c c (1)
p0 (ρ/c) ρ /c! n = c + 1, c + 2, . . .
where ρ = λ/µ = λ. We must be sure that ρ is small enough that there exists p0 > 0 such that
! "
ρn ρc ρ n−c
∞
c ∞
pn = p0 1 + + =1 (2)
n! c! c
n=0 n=1 n=c+1
This requirement is met if and only if the infinite sum converges, which occurs if and only if
∞
ρ n−c ρ j
∞
= <∞ (3)
c c
n=c+1 j=1
That is, p0 > 0 if and only if ρ/c < 1, or λ < c. In short, if the arrival rate in cars per second is
less than the service rate (in cars per second) when all booths are busy, then the Markov chain has
a stationary distribution. Note that if ρ > c, then the Markov chain is no longer positive recurrent
and the backlog of cars will grow to infinity.
(a) In this case, we have two M/M/1 queues, each with an arrival rate of λ/2. By defining
ρ = λ/µ, each queue has a stationary distribution
pn = (1 − ρ/2) (ρ/2)n n = 0, 1, . . . (1)
Note that in this case, the expected number in queue i is
∞
ρ/2
E [Ni ] = npn = (2)
1 − ρ/2
n=0
The expected number in the system is

ρ
E [N1 ] + E [N2 ] = (3)
1 − ρ/2
(b) The combined queue is an M/M/2/∞ queue. As in the solution to Quiz 12.10, the stationary
probabilities satisfy
p0 ρn /n! n = 1, 2
pn = n−2 2 (4)
p0 ρ ρ /2 n = 3, 4, . . .

The requirement that ∞ n=0 pn = 1 yields

−1
ρ2 ρ2 ρ/2 1 − ρ/2
p0 = 1 + ρ + + = (5)
2 2 1 − ρ/2 1 + ρ/2

The expected number in the system is E[N ] = ∞ n=1 npn . Some algebra will show that
ρ
E [N ] = (6)
1 − (ρ/2)2
We see that the average number in the combined queue is lower than in the system with individual
queues. The reason for this is that in the system with individual queues, there is a possibility that
one of the queues becomes empty while there is more than one person in the other queue.
433
This implies
pn =
⎧
⎪
⎨
⎪
p0
0
pn =
P [H] = pc =
0
n=0
n
(λ + h)n
n!
⎩ (λ + h) λ
0 c−r
lh
p0
c−r
n!
(λ + h)n
n!
c!
l
The LCFS queue operates in a way that is quite different from the usual first come, first served
queue. However, under the assumptions of exponential service times and Poisson arrivals, customers
arrive at rate λ and depart at rate µ, no matter which service discipline is used. The Markov chain
for the LCFS queue is the same as the Markov chain for the M/M/1 first come, first served queue:
ll
n−(c−r)
The requirement that cn=1 pn = 1 yields

1
lh
c-r
l ll
c-r
l
c-r+1
2
l
l ll
c-1
n = c − r + 1, c − r + 2, . . . , c
n = 1, 2, . . . , c − r
p0 n = c − r + 1, c − r + 2, . . . , c
+ (λ + h)c−r
The probability that a handoff call is dropped is

(λ + h)c−r λr
p0 =
c−r
c
n=0
434
λ n−(c−r)
n!
n=c−r+1
(λ+h)n
n!
1
=1
3
c-1
l
c
l ll
It would seem that the LCFS queue should be less efficient than the ordinary M/M/1 queue because
a new arrival causes us to discard the work done on the customer in service. This is not the case,
however, because the memoryless property of the exponential PDF implies that no matter how
much service had already been performed, the remaining service time remains identical to that of
a new customer.

Since both types of calls have exponential holding times, the number of calls in the system can be
used as the system state. The corresponding Markov chain is
When the number of calls, n, is less than c − r, we admit either type of call and qn,n+1 = λ + h.
When n ≥ c − r, we block the new calls and we admit only handoff calls so that qn,n+1 = h. Since
the service times are exponential with an average time of 1 minute, the call departure rate in state
n is n calls per minute. Theorem 12.24 says that the stationary probabilities pn satisfy
⎧
⎪
⎨ λ + h pn−1 n = 1, 2, . . . , c − r
⎪ λ
n
⎩ pn−1
(1)
c
Finally, a handoff call is dropped if and only if a new call finds the system with c calls in progress.
(λ + h)c−r λr /c!

+ λ+h
λ
c−r
c
n=c−r+1
λn
n!
(2)
(3)
(4)
Here is the Markov chain describing the free throws.
0.9
0.6 0.7 0.8
1 2 3 4
0.5
0.1 0.2 0.3 0.4 0 0.4 0.3 0.2 0.1

0.9
0.5
0.8 0.7 0.6
-4 -3 -2 -1
Note that state 4 corresponds to “4 or more consecutive successes” while state −4 corresponds to
“4 or more consecutive misses.” We denote the stationary probabilities by the vector

π = π−4 π−3 π−2 π−1 π0 π1 π2 π3 π4 . (1)
For this vector π, the state transition matrix is

⎡ ⎤
0.9 0 0 0 0 0.1 0 0 0
⎢0.8 0 0 0 0 0.2 0 0 0⎥
⎢ ⎥
⎢ 0 0.7 0 0 0 0.3 0 0 0⎥
⎢ ⎥
⎢0 0 0.6 0 0 0.4 0 0 0⎥
⎢ ⎥
⎢
P=⎢ 0 0 0 0.5 0 0.5 0 0 0⎥ ⎥. (2)
⎢0 0⎥
⎢ 0 0 0.4 0 0 0.6 0 ⎥
⎢0 0 0.7 0 ⎥
⎢ 0 0 0.3 0 0 ⎥
⎣0 0 0 0.2 0 0 0 0 0.8⎦
0 0 0 0.1 0 0 0 0 0.9
To solve the problem at hand, we divide the work into two functions; freethrowmat(n) returns
the n step transition matrix and freethrowp(n) that calculates the probability of a success on the
free throw n.
function Pn=freethrowmat(n); function ps=freethrowp(n);
P=[0.9 0 0 0 0 0.1 0 0 0;... PP=freethrowmat(n-1);
0.8 0 0 0 0 0.2 0 0 0;... p0=[zeros(1,4) 1 ...
0 0.7 0 0 0 0.3 0 0 0;... zeros(1,4)];
0 0 0.6 0 0 0.4 0 0 0;... ps=p0*PP*0.1*(1:9)’;
0 0 0 0.5 0 0.5 0 0 0;...
0 0 0 0.4 0 0 0.6 0 0;...
0 0 0 0.3 0 0 0 0.7 0;...
0 0 0 0.2 0 0 0 0 0.8;...
0 0 0 0.1 0 0 0 0 0.9];
Pn=P^n;
In freethrowp.m, p0 is the initial state probability row vector π (0). Thus p0*PP is the state
probability row vector after n − 1 free throws. Finally, p0*PP*0.1*(1:9)’ multiplies the state
probability vector by the conditional probability of successful free throw given the current state.
The answer to our problem is simply
435
>> freethrowp(11)
ans =
0.5000
>>
In retrospect, the calculations are unnecessary! Because the system starts in state 0, symmetry
of the Markov chain dictates that states −k and k will have the same probability at every time
step. Because state −k has success probability 0.5 − 0.1k while state k has success probability
0.5 + 0.1k, the conditional success probability given the system is in state −k or k is 0.5. Averaged
over k = 1, 2, 3, 4, the average success probability is still 0.5.
Comment: Perhaps finding the stationary distribution is more interesting. This is done fairly
easily:
>> p=dmcstatprob(freethrowmat(1));
>> p’
ans =
0.3123 0.0390 0.0558 0.0929 0 0.0929 0.0558 0.0390 0.3123
About sixty precent of the time the shooter has either made four or more consecutive free throws
or missed four or more free throws. On the other hand, one can argue that in a basketball game,
a shooter rarely gets to take more than a half dozen (or so) free throws, so perhaps the stationary
distribution isn’t all that interesting.

In this problem, we model the system as a continuous time Markov chain. The clerk and the
manager each represent a “server.” The state describes the number of customers in the queue and
the number of active servers. The Markov chain issomewhat complicated because when the number
of customers in the store is 2, 3, or 4, the number of servers may be 1 or may be 2, depending on
whether the manager became an active server.
When just the clerk is serving, the service rate is 1 customer per minute. When the manager
and clerk are both serving, the rate is 2 customers per minute. Here is the Markov chain:

0 1 2c 3c 4c 5 6
1 1 1 1 2 2

2
2 2m 3m 4m
2 2
In states 2c, 3c and 4c, only the clerk is working. In states 2m, 3m and 4m, the manager is also
working. The state space {0, 1, 2c, 3c, 4c, 2m, 3m, 4m, 5, 6, . . .} is countably infinite. Finding the
state probabilities is a little bit complicated because the are enough states that we would like to
use Matlab; however, Matlab can only handle a finite state space. Fortunately, we can use
Matlab because the state space for states n ≥ 5 has a simple structure.
We observe for n ≥ 5 that the average rate of transitions from state n to state n + 1 must equal
the average rate of transitions from state n + 1 to state n, implying
λpn = 2pn+1 , n = 5, 6, . . . (1)
It follows that pn+1 = (λ/2)pn and that
pn = αn−5 p5 , n = 5, 6, . . . , (2)
436
where α = λ < 2 < 1. The requirement that the stationary probabilities sum to 1 implies

4 ∞

1 = p0 + p 1 + (pjc + pjm ) + pn (3)
j=2 n=5

4 ∞

= p0 + p1 + (pjc + pjm ) + p5 αn−5 (4)
j=2 n=5

4
p5
= p0 + p1 + (pjc + pjm ) + (5)
1−α
j=2
This is convenient because for each state
j <
5, we can solve for the staitonary probabilities. In
particular, we use Theorem 12.23 to write i rij pi = 0. This leads to a set of matrix equations
for the state probability vector

p = p0 p1 p2c p3c p3c p4c p2m p3m p4m p5 (6)
The rate transition matrix associated with p is

⎡ ⎤
p0 p1 p2c p3c p4c p2m p3m p4m p5
⎢0 λ 0 0 0 0 0 0 0⎥
⎢ ⎥
⎢1 0 λ 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢0 1 0 λ 0 0 0 0 0⎥
⎢ ⎥
⎢0 0 1 0 λ 0 0 0 0⎥
⎢
Q=⎢ ⎥, (7)
⎢0 0 0 1 0 0 0 0 λ⎥ ⎥
⎢0 2 ⎥
⎢ 0 0 0 0 λ 0 0 ⎥
⎢0 0 0⎥
⎢ 0 0 0 2 0 λ ⎥
⎣0 0 0 0 0 0 2 0 λ⎦
0 0 0 0 0 0 0 2 0
where the first row just shows the correspondence of the state probabilities and the matrix columns.
For each state i, excepting state 5, the departure rate νi from that state equals the sum of entries
of the corresponding row of Q. To find the stationary probabilities, our normal procedure is to
use Theorem 12.23 and solve p R = 0 and p 1 = 1, where R is the same as Q except the zero
diagonal entries are replaced by −νi . The equation p 1 = 1 replaces one column of the set of matrix
equations. This is the approach of cmcstatprob.m.
In this problem, we follow almost the same procedure. We form the matrix R by replacing
the diagonal entries of Q. However, instead of replacing an arbitrary column with the equation
p 1 = 1, we replace the column corresponding to p5 with the equation
p5
p0 + p1 + p2c + p3c + p4c + p2m + p3m + p4m + = 1. (8)
1−α
That is, we solve
p R = 0 0 0 0 0 0 0 0 0 1 . (9)
437
where
⎡ ⎤
−λ λ 0 0 0 0 0 0 1
⎢ 1 −1 − λ λ 0 0 0 0 0 1 ⎥
⎢ ⎥
⎢ 0 1 −1 − λ λ 0 0 0 0 1 ⎥
⎢ ⎥
⎢ 0 0 1 −1 − λ λ 0 0 0 1 ⎥
⎢ ⎥
R=⎢
⎢ 0 0 0 1 −1 − λ 0 0 0 1 ⎥
⎥ (10)
⎢ 0 2 0 0 0 −2 − λ λ 0 1 ⎥
⎢ ⎥
⎢ 0 −2 − ⎥
⎢ 0 0 0 0 2 λ λ 1 ⎥
⎣ 0 0 0 0 0 0 2 −2 − λ 1 ⎦
1
0 0 0 0 0 0 0 2 1−α
Given the stationary distribution, we can now find E[N ] and P [W ].

Recall that N is the number of customers in the system at a time in the distant future. Defining
pn = pnc + pnm , n = 2, 3, 4, (11)
we can write
∞

4 ∞

E [N ] = npn = npn + np5 αn−5 (12)
n=0 n=0 n=5
The substitution k = n − 5 yields

4 ∞

E [N ] = npn + p5 (k + 5)αk (13)
n=0 k=0
4 ∞
5
= npn + p5 + p5 kαk (14)
1−α
n=0 k=0
From Math Fact B.7, we conclude that

4
5 α
E [N ] = npn + p5 + (15)
1 − α (1 − α)2
n=0
4
5 − 4α
= npn + p5 (16)
(1 − α)2
n=0
Furthermore, the manager is working unless the system is in state 0, 1, 2c, 3c, or 4c. Thus
P [W ] = 1 − (p0 + p1 + p2c + p3c + p4c ). (17)
We implement these equations in the following program, alongside the corresponding output.
438
function [EN,PW]=clerks(lam); >> [en05,pw05]=clerks(0.5)
Q=diag(lam*[1 1 1 1 0 1 1 1],1); en05 =
Q=Q+diag([1 1 1 1 0 2 2 2],-1); 0.8217
Q(6,2)=2; Q(5,9)=lam; pw05 =
R=Q-diag(sum(Q,2)); 0.0233
n=size(Q,1); >> [en10,pw10]=clerks(1.0)
a=lam/2; en10 =
R(:,n)=[ones(1,n-1) 1/(1-a)]’; 2.1111
pv=([zeros(1,n-1) 1]*R^(-1)); pw10 =
EN=pv*[0;1;2;3;4;2;3;4; ... 0.2222
(5-4*a)/(1-a)^2]; >> [en15,pw15]=clerks(1.5)
PW=1-sum(pv(1:5)); en15 =
4.5036
pw15 =
0.5772
>>
We see that in going from an arrival rate of 0.5 customers per minute to 1.5 customers per
minute, the average number of customers goes from 0.82 to 4.5 customers. Similarly, the probability
the manager is working rises from 0.02 to 0.57.

Although the inventory system in this problem is relatively simple, the performance analysis is
suprisingly complicated. We can model the system as a Markov chain with state Xn equal to
the number of brake pads in stock at the start of day n. Note that the range of Xn is SX =
{50, 51, . . . , 109}. To evaluate the system, we need to find the state transition matrix for the
Markov chain. We express the transition probabilities in terms of PK (·), the PMF of the number
of brake pads ordered on an arbitary day. In state i, there are two possibilities:
• If 50 ≤ i ≤ 59, then there will be min(i, K) brake pads sold. At the end of the day, the
number of pads remaining is less than 60, and so 50 more pads are delivered overnight. Thus
the next state is j = 50 if K ≥ i pads are ordered, i pads are sold and 50 pads are delivered
overnight. On the other hand, if there are K < i orders, then the next state is j = i − K + 50.
In this case,
P [K ≥ i] j = 50,
Pij = (1)
PK (50 + i − j) j = 51, 52, . . . , 50 + i.
• If 60 ≤ i ≤ 109, then there are several subcases:
– j = 50: If there are K ≥ i orders, then all i pads are sold, 50 pads are delivered overnight,
and the next state is j = 50. Thus
Pij = P [K ≥ i] , j = 50. (2)
– 51 ≤ j ≤ 59: If 50 + i − j pads are sold, then j − 50 pads ar left at the end of the day.
In this case, 50 pads are delivered overnight, and the next state is j with probability
Pij = PK (50 + i − j) , j = 51, 52, . . . , 59. (3)
– 60 ≤ j ≤ i: If there are K = i − j pads ordered, then there will be j ≥ 60 pads at the

end of the day and the next state is j. On the other hand, if K = 50 + i − j pads are
439
ordered, then there will be i − (50 + i − j) = j − 50 pads at the end of the day. Since
60 ≤ j ≤ 109, 10 ≤ j − 50 ≤ 59, there will be 50 pads delivered overnight and the next
state will be j. Thus
Pij = PK (i − j) + PK (50 + i − j) , j = 60, 61, . . . , i. (4)
– For i < j ≤ 109, state j can be reached from state i if there 50 + i − j orders, leaving
i − (50 + i − j) = j − 50 in stock at the end of the day. This implies 50 pads are delivered
overnight and the next stage is j. the probability of this event is
Pij = PK (50 + i − j) , j = i + 1, i + 2, . . . , 109. (5)
We can summarize these observations in this set of state transition probabilities:

⎧
⎪
⎪ P [K ≥ i] 50 ≤ i ≤ 109, j = 50,
⎪
⎪ PK (50 + i − j) 50 ≤ i ≤ 59, 51 ≤ j ≤ 50 + i,
⎪
⎪
⎨
PK (50 + i − j) 60 ≤ i ≤ 109, 51 ≤ j ≤ 59,
Pij = (6)
⎪
⎪ PK (i − j) + PK (50 + i − j) 60 ≤ i ≤ 109, 60 ≤ j ≤ i,
⎪
⎪
⎪
⎪ P (50 + i − j) 60 ≤ i ≤ 108, i + 1 ≤ j ≤ 109
⎩ K
0 otherwise
Note that the “0 otherwise” rule comes into effect when 50 ≤ i ≤ 59 and j > 50 + i. To simplify
these rules, we observe that PK (k) = 0 for k < 0. This implies PK (50 + i − j) = 0 for j > 50 + i.
In addition, for j > i, PK (i − j) = 0. These facts imply that we can write the state transition
probabilities in the simpler form:
⎧
⎪
⎪ P [K ≥ i] 50 ≤ i ≤ 109, j = 50,
⎨
PK (50 + i − j) 50 ≤ i ≤ 59, 51 ≤ j
Pij = (7)
⎪
⎪ P (50 + i − j) 60 ≤ i ≤ 109, 51 ≤ j ≤ 59,
⎩ K
PK (i − j) + PK (50 + i − j) 60 ≤ i ≤ 109, 60 ≤ j
Finally, we make the definitions
βi = P [K ≥ i] , γk = PK (50 + k) , δk = PK (k) + PK (50 + k) . (8)
With these definitions, the state transition probabilities are

⎧
⎪
⎪ βi 50 ≤ i ≤ 109, j = 50,
⎨
γi−j 50 ≤ i ≤ 59, 51 ≤ j
Pij = (9)
⎪
⎪ γ 60 ≤ i ≤ 109, 51 ≤ j ≤ 59,
⎩ i−j
δi−j 60 ≤ i ≤ 109, 60 ≤ j
440
Expressed as a table, the state transition matrix P is
i\j 50 51 · · · 59 60 ··· ··· ··· ··· 109

50 β50 γ−1 · · · γ−9 ··· ··· ··· ··· ··· γ−59
.. .. .. ..
51 β51 γ0 . . . .
.. .. .. .. .. .. ..
. . . . γ−1 . . .
59 β59 γ8 · · · γ0 γ−1 · · · γ−9 ··· ··· γ−50
60 β60 γ9 · · · γ1 δ0 · · · δ−9 ··· δ−49 (10)
.. .. .. .. .. .. .. .. ..
. . . . . . . . .
.. .. .. ..
. . . γ9 . δ−9
.. .. .. .. .. ..
. . . . . δ9 .
.. .. .. .. .. .. .. .. ..
. . . . . . . . .
109 β109 γ58 · · · γ50 δ49 · · · δ9 ··· ··· δ0
In terms of Matlab, all we need to do is to encode the matrix P, calculate the stationary
probability vector π, and then calculate E[Y ], the expected number of pads sold on a typical day.
To calculate E[Y ], we use iterated expectation. The number of pads ordered is the Poisson random
variable K. We assume that on a day n that Xn = i and we calculate the conditional expectation

i−1
E [Y |Xn = i] = E [min(K, i)] = jPK (j) + iP [K ≥ i] . (11)
j=0
Since only states i ≥ 50 are possible, we can write

48
i−1
E [Y |Xn = i] = jPK (j) + jPK (j) + iP [K ≥ i] . (12)
j=0 j=49
Finally, we assume that on a typical day n, the state of the system Xn is described by the stationary
probabilities P [Xn = i] = πi and we calculate

109
E [Y ] = E [Y |Xn = i] πi . (13)
i=50
These calculations are given in this Matlab program:
441
function [pstat,ey]=brakepads(alpha); The first half of brakepads.m constructs P to cal-
s=(50:109)’; culate the stationary probabilities. The first col-
beta=1-poissoncdf(alpha,s-1); umn of P is just the vector
grow=poissonpmf(alpha,50+(-1:-1:-59));

gcol=poissonpmf(alpha,50+(-1:58)); beta = β50 · · · β109 . (14)
drow=poissonpmf(alpha,0:-1:-49);
dcol=poissonpmf(alpha,0:49); The rest of P is easy to construct using toeplitz
P=[beta,toeplitz(gcol,grow)];
function. We first build an asymmetric Toeplitz
P(11:60,11:60)=P(11:60,11:60)...
+toeplitz(dcol,drow); matrix with first row and first column
pstat=dmcstatprob(P);
[I,J]=ndgrid(49:108,49:108); grow = γ−1 γ−2 · · · γ−59 (15)

G=J.*(I>=J); gcol = γ−1 γ0 · · · γ58 (16)
EYX=(G*gcol)+(s.*beta);
pk=poissonpmf(alpha,0:48);
EYX=EYX+(0:48)*pk;
ey=(EYX’)*pstat;
Note that δk = PK (k) + γk . Thus, to construct the Toeplitz matrix in the lower right corner of
P, we simply add the Toeplitz matrix corresponding to the missing PK (k) term. The second half
of brakepads.m calculates E[Y ] using the iterated expectation. Note that

EYX = E [Y |Xn = 50] · · · E [Y |Xn = 109] . (17)
The somewhat convoluted code becomes clearer by noting the following correspondences:

48
i−1
E [Y |Xn = i] = jPK (j) + jPK (j) + iP [K ≥ i] . (18)

j=0 j=49
s.*beta
(0:48)*pk G*gcol
To find E[Y ], we execute the commands:
>> [ps,ey]=brakepads(50);
>> ey
ey =
49.4154
>>
We see that although the store receives 50 orders for brake pads on average, the average number
sold is only 49.42 because once in awhile the pads are out of stock. Some experimentation will show
that if the expected number of orders α is significantly less than 50, then the expected number of
brake pads sold each days is very close to α. On the other hand, if α 50, then the each day the
store will run out of pads and will get a delivery of 50 pads ech night. The expected number of
unfulfilled orders will be very close to α − 50.
Note that a new inventory policy in which the overnight delivery is more than 50 pads or the
threshold for getting a shipment is more than 60 will reduce the expected numer of unfulfilled
orders. Whether such a change in policy is a good idea depends on factors such as the carrying
cost of inventory that are absent from our simple model.

This problem is actually very easy. The state of the system is given by X, the number of cars in
the system.When X = 0, both tellers are idle. When X = 1, one teller is busy, however, we do not
need to keep track of which teller is busy. When X = n ≥ 2, both tellers are busy and there are
n − 2 cars waiting. Here is the Markov chain:
442
l l l l l
0 1 2 7 8
1 2 2 2 2
Since this is a birth death process, we could easily solve this problem using analysis. However,
as this problem is in the Matlab section of this chapter, we might as well construct a Matlab
solution:
function [p,en]=veryfast2(lambda); The code solves the stationary distribution and the ex-
c=2*[0,eye(1,8)]’; pected number of cars in the system for an arbitrary
r=lambda*[0,eye(1,8)]; arrival rate λ.
Q=toeplitz(c,r);
Q(2,1)=1;
p=cmcstatprob(Q);
en=(0:8)*p;
Here is the output:
>> [p,en]=veryfast2(0.75);
>> p’
ans =
0.4546 0.3410 0.1279 0.0480 0.0180 0.0067 0.0025 0.0009 0.0004
>> en
en =
0.8709
>>

Under construction.

Under construction.

Under construction.

Under construction.

Under construction.
443

Sol Manual

Uploaded by

Copyright:

Available Formats

Sol Manual

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

Sol Manual

Uploaded by

Copyright:

Available Formats

Probability and Stochastic Processes

A Friendly Introduction for Electrical and Computer Engineers

September 28, 2005

12.1.7, 12.1.8, 12.5.8, 12.5.9, 12.11.5 – 12.11.9.

– A manual probmatlab.pdf describing the matcode .m functions is also available.

Problem 1.1.1 Solution

the answers are fairly straightforward:

Problem 1.1.2 Solution

the complete Gerlandas pizza menu is

Problem 1.2.1 Solution

S = {ht, hf, mt, mf, lt, lf } . (1)

Problem 1.2.2 Solution

(a) The sample space of the experiment is

S = {aaa, aaf, af a, f aa, f f a, f af, af f, f f f} . (1)

(b) The event that the circuit from Z fails is

ZF = {aaf, af f, f af, f f f } . (2)

The event that the circuit from X is acceptable is

XA = {aaa, aaf, af a, af f } . (3)

(c) Since ZF ∩ XA = {aaf, af f } = φ, ZF and XA are not mutually exclusive.

(e) The event that more than one circuit is acceptable is

C = {aaa, aaf, af a, f aa} . (4)

The event that at least two circuits fail is

(f) Inspection shows that C ∩ D = φ so C and D are mutually exclusive.

(g) Since C ∪ D = S, C and D are collectively exhaustive.

Problem 1.2.3 Solution

S = {A♣, . . . , K♣, A♦, . . . , K♦, A♥, . . . , K♥, A♠, . . . , K♠} . (1)

The event H is the set

Problem 1.2.5 Solution

Problem 1.2.6 Solution

S = {LF, BF, LW, BW } . (1)

(a) The probability that a program is slow is

P [W ] = P [LW ] + P [BW ] = 0.1 + 0.2 = 0.3. (2)

(b) The probability that a program is big is

P [B] = P [BF ] + P [BW ] = 0.2 + 0.2 = 0.4. (3)

(c) The probability that a program is slow or big is

P [W ∪ B] = P [W ] + P [B] − P [BW ] = 0.3 + 0.4 − 0.2 = 0.5. (4)

Problem 1.3.2 Solution

S = {HF, HW, M F, M W } . (1)

P [HW ] = 1 − P [HF ] − P [M F ] − P [M W ] = 1 − 0.2 − 0.3 − 0.1 = 0.4. (4)

(a) The probability a cell phone is slow is

P [W ] = P [HW ] + P [M W ] = 0.4 + 0.1 = 0.5. (5)

(c) The probability that a cell phone is handheld is

P [H] = P [HF ] + P [HW ] = 0.2 + 0.4 = 0.6. (6)

Problem 1.3.4 Solution

Problem 1.3.5 Solution

Problem 1.4.1 Solution

P [L] = 1 − P [3 or fewer billed minutes] (1)

Problem 1.4.3 Solution

Y = {yy, yg, gy} . (1)

So the probability of a pea plant with yellow seeds is

P [Y ] = P [yy] + P [yg] + P [gy] = 3/4. (2)

Problem 1.4.4 Solution

(a) Since A ⊂ A ∪ B, P [A] ≤ P [A ∪ B].

(b) Since B ⊂ A ∪ B, P [B] ≤ P [A ∪ B].

(c) Since A ∩ B ⊂ A, P [A ∩ B] ≤ P [A].

(d) Since A ∩ B ⊂ B, P [A ∩ B] ≤ P [B].

Problem 1.4.5 Solution

P [A ∪ B] = P [A] + P [B] − P [A ∩ B] . (1)

P [A1 ∪ A2 ] = P [A1 ] + P [A2 ] − P [A1 ∩ A2 ] . (2)

1/4 H2 •H1 H2 1/16