1

Solutions to Chapter 3
1. We start with the definition of the CDF of ,
() , [ ]

= [ + ]

= [ ]

for 0

Now for any value we have [ ] + [ ] = 1 and [ ] = () [ = ]

Hence for 0

() =

= 1

= 1 (
) =

= 1 (
)+ =

2. We are given that is a Gaussian random variable distributed as (0 1) i.e. = 0 and

2 = 1 A second random variable is the result of the transformation = () (Fig. 1)

0
() =
2 0

Figure 1:
We are asked to find the pdf of = () We proceed using the indirect method of first
finding the CDF. From the figure we can see that, for 0 the event

{ } = { }
For 0 the event { } = the null event, with probability zero. So the sought after
CDF becomes
() = [ ]

= [ ]

= () ( )

2
Taking the derivative with respect to we get
()

() ( )

= () ( )

= () + ( )
2

() ,

Now plugging in the standard Normal (Gaussian) distribution for we get

1 2
1
1
1
() = 2 +
2
2 2
2

For 0 the density () = 0 Since is a continuous random variable, there is no

probability mass at = 0 so we can set () = 0 there too. The overall pdf for then
becomes
(
1 2
1
1 2 + 1 2 0
2
2
2
() =
0
0
3. (function of RV ) For a given value of = , with 0 there are two solutions for , i.e.
1 () = 12 and 2 () = as shown in the Fig. 2.

Figure 2:
Thus the pdf of is given as

1 ()
2 ()

(2 ())

(1 ()) +
() =



1
= (2) + |1| ()
2
1
(2) + ()
=
2
When 0, there are no solutions, so the general form of the pdf of is

1
(2) + () ()
() =
2

3
where () is the unit step function. For this problem, we have (0 25)thus
() =
=
so that
() =
=

1 2
1

2 25
25
2
1

50
25

(2)2
()2
1 1
1

50 +
50
2 25
25

1
1
1 2

200 + 50 ()
5
2 10

()

4. We are given that is a Gaussian random variable distributed as (0 2) i.e. = 0 and

2 = 2 A second random variable is the result of the transformation = ()

0
() =
22 0
We are asked to find the probability density function (pdf) of = () We find the CDF
first and then dierentiate to find the pdf. From the equation for we can see that, for
0 the event
r

}
{ } = {
2

For 0 the event { } = the null event, with probability zero. So the sought after
CDF becomes
() = [ ]
r

= [
]
2
r

)
= () (
2
Taking the derivative with respect to we get
() ,
=

()

()

q
( 2 )

q
r
2

)
= () (
2

r
1
) q
= () + (
2 4
2

Now plugging in the standard Normal (Gaussian) distribution for we get

1 2
1
1
1
() = 4 + q 8
2
8 2

4
For 0 the density () = 0 Since is a continuous random variable, there is no
probability mass at = 0 so we can set () = 0 there too. The overall pdf for then
becomes
(
1 2
1
1

4 + 1 8 0
2

() =
2
0
0

5. In both parts of this problem the random variable has its probability density given as
exponential with parameter ( 0)i.e. () = ()
(a) Here the function is given as = () = 3 , as shown in Fig. 3.

Figure 3:
Using the indirect method, we see that the distribution function of satisfies () =
( 13 ) since the event { } = { 13 } Upon dierentiation we get
() = (13 )
= ( 13 )

13

1
13
= ( 13 ) 23
3
13
=

().
323
Alternatively, and since the function is monotonic, we can easily use the direct method
to get the pdf of = () = 3 We get

13
= ( )

() = ( 13 )|

Now

= 32 = 3 23 so

13

() = (13 )323
13
=

() same as above.
3 23

5
(b) For this part, only the function has changed, this time () = 2 + 3 We choose to
use the indirect method, and find
() = [ ]

3
=
2

3
=

2
Taking the derivative with respect to the free variable we get

3
1
() =

2
2
which, for the exponential distribution with parameter becomes
() =
=

1 3 3
2 (

)
2
2
(3)
2
( 3)
2

6. By definition () = [ ] = [() ], so () = 1 for all +1 for this

transformation When 1 +1 we can write
() = [() ]
= [ ]
= ()

then calculating the distribution function for the Laplacian density, we get by running integration

1 12 0
() =
1 +
0
2
Thus, we have
1 +1
() = ()

1
1 2 1 0
=
1 +
1 0
2
Clearly, for 1 we must have () = 0 Combining these results, we obtain

1
1

1 12 1 0
() =
1 +

0 1

2
0
1
7. We present two methods to solve this transformation problem:
Method (1): For 0
[ ] = [ ] = [ ln ] = (ln )

6
Hence
() =
For 0
Hence

(ln ) (ln )
1
(ln )
=
= (ln )

()

[ ] = [ ] = [] = 0

1 ln 2
1
exp[ (
) ]()
() =
2

Method (2): A plot of = () is is = ln , for 0 as given in Fig. 4.

Figure 4:

() =

X
( )
= (ln ) |=ln
| 0 ( )|
=1

For 0, there is no real solution to () = 0; hence for 0, the pdf of equals zero
there. Combining solutions for both regions of , we get
() =

(ln )
()

8. (a) Note that the peak is approximately at ln = or = exp when . To investigate

the behavior near zero, to aid in a hand plot, we note
1 (ln )2

=
=

1 (ln )(ln )

1 ln ln

1 1 ln

= ln 1

Now near zero, the exponent ln 1 takes on large positive values, thus ln 1
converges to zero as approaches zero from the right. So the hand plots should just
look to first order like a Gaussian density, with a log scale for the horizontal axis.
Precise plotting with = 2 and = 1, results in the Fig. 5. The plots are in terms of
conventional versus axis plots.

Figure 5:
1
2

exp (ln 2)
2

A second plot is with = 0 and = 4 is given in Fig. 6.

1
24

exp (ln32)

Figure 6:
(b) We first integrate the density () =

1
2

)
exp (ln 2
() We have
2

() = 0 for 0.
For 0 we write
() =

1
(ln )2

exp
2 2
2

We then make the substitution , ln to get (with = )

Z ln
1
( )2

() =
exp

2 2
2

Z ln
( )2
1

exp

=
2 2
2

ln
1
=
+ erf

8
The overall solution for the distribution function (CDF) is then

ln
1
+ erf
()
() =
2

The alternative solution is easier and starts with the distribution function
() , [ ]

= [ ]

= [ ln ],

because ln() is a monotonic increasing function,

= (ln )
on 0

ln
1
+ erf
on 0
=
2

1
ln
=
+ erf
() since () = 0 for 0
2

9. Method (1):
[ ] = [ln ] = [ ] = ( )
() =

( )
1 1
= ( ) = 3 ( )

( )
3

But since 0 all real , ( ) = 1 everywhere. Hence

1 1
() = 3 (exp()3)
3
Method (2): We plot = () first, in Fig. 7.

The only solution to () = 0 is = ln

Figure 7:
for Hence

( )
||=
= ( )

() =

10. Here is the plot of = () in Fig. 8: () = 0 is =

solutions exist to () = 0. Hence
() =

()
| 2
0 () =

(1)

, or = 2 . For 0, no real

Figure 8:
Now

1 1
1

0 () = 2 |=2 =
2
2

thus
() =

( q

2
12 4

0
0

But at = 0 there is a probability mass, jump in the distribution function, and impulse in
the density function. We have
[ = 0] = [ 0]
1
= (0) =
2
thus
Z
1
(0) =
()
2
So, combining results, we have the following answer valid for all ,
r
2 1 4
1
() = () +
2 ()
2

11. (a) From the function given in Fig. P3.11 in the text, we see that the event {
} = { +} = 1 for 1 Hence () = 1 there. Next consider
0 1 again from the figure, we see that in this region of the function with slope
1, { } = { } { 2}, a disjoint union. So in this region of values,
it must be that () = () + (1 (2)) Right at the point = 0 we have
{ } = { 0} { 2} Now consider the remaining region 0 there the
event { } = the null event, since there are no values that map to 0 for the
given function shown in Fig. 9. So () = 0 there. Since : (0 1) we can write

0 1

1
+
erf()
+
1

+
erf(2)

1
() =
2
2
1
1

0
0

=
1 + erf() erf(2) 0 1

1
1
1

Actually, we only said that random variables take on finite values with probability one. So there could be some
events of probability zero that actually attain = However, since they are probability zero, the distribution
function is not aected.

10
Taking derivatives with respect to the free variable we get within each interval

Figure 9:

() =

0
1 2
1 2
2

0
0 1
1

This is not the complete story though as there are jumps in the CDF () at both
= 0 and = 1. Thus we must add to this pdf two impulses located at these two jump
points. From the CDF of given above, we see that
(1) (1 ) = 1 (1 + erf(1) erf(2))
= erf(2) erf(1)

= 0137

so this is the area of the impulse needed at = 1. It thus becomes 0137( 1) The
jump in the CDF () at = 0 is found as
(0) (0 ) = (1 + erf(0) erf(2)) 0
= 1 erf(2)

= 0523

so the needed impulse in the density at = 0 becomes 0523() Combining these

results, we get the pdf () over the full domain of as
1 2
1
() = 0523() + 2 (() ( 1)) + 0137( 1)
2

(b) Equation 3.2-23 in the book will only give us the portion of the solution for 0 1
where in fact there is one root 1 () = and = 1 Using this method, we thus get
1 2
() = 12 2 and miss the impulses at = 0 and = 1 The problem here is
that our function has regions of zero slope. Over such regions, there are uncountably
infinitely many values corresponding to the one value. Thus Equation 3.2-23 does
not hold there.
12. The given pdf of is uniform over [0 2] Thus () = 12 (() ( 2)) The function
(Fig. 10) is given as

0
0

2
0 12
() =
2 2 12 1

0
1

11

Figure 10:
For 1 for the given function the event { } = { +} so for all 1,
the CDF () = 1 When 0 1 there are two solutions to the equation = (), and
they can be given as 1 () = 2 and 2 () = 1 2. So, using Equation 3.2-23, we get
1
1
() = (2) + (1 2)
2
2

0 1

At = 0, we see the probability mass [ = 0] = [ 0] + [ 1] so we must add an

impulse of this area to the density () at = 0. The overall answer for the general pdf
then becomes,

1
1
(2) + (1 2) (() ( 1)) + ( [ 0] + [ 1]) ()
() =
2
2
For the given uniform density of random variable i.e. : [0 2] we have [ 0] = 0
and [ 1] = 12 Also
(2) =
and
(1 2) =

1
1
((2) (2 2)) = (() ( 4))
2
2

1
1
((1 2) (1 2 2)) = ((2 ) ( 2))
2
2

so that
() =

1
1
1
(() ( 4)) + ((2 ) ( 2)) (() ( 1)) + ()
4
4
2

If we define a function
rect() ,

1 12 + 12
0
else

21
1
then we can write () = 12 rect 1
(2)
=
rect
and
so

= 12 rect( 2

2
2
2
4 ) and

= 12 rect
(1 2) = 12 rect 121
= 12 rect 4 The overall answer for the density
2
4
can then be written as

2
1
1
1
1
rect(
) + rect
rect( ) + ()
() =
4
4
4
4
2
2
1
1
1
rect( ) + ()
=
2
2
2

12

Figure 11:
13. First we plot = () in Fig. 11.
For 0 12 : () = 1 2. For 12 1:
() = 2 1; elsewhere () = 0 For 1, no real roots to () = 0, then pdf = 0
there. For = 1, [ = 1] = [ = 0] + [ = 1] = 0 since is a continuous random
variable. For 0, there are no real roots to () = 0, then pdf=0. For 0 1,
there are two roots: 1 = 12 2 ; 2 = 12 + 2 So

X
1
( )
1
1
1
() =
= ( ) + ( + )
||
2
2 2
2
2 2
=1

For = 0, since : (0 2)
1
[ = 0] = [ 0] + [ 1] =
2
Therefore at = 0, () = 12 (). To construct ():

1
1

() = rect
2
2
Therefore for 0 1:

1
1
1
1
+ rect

() = rect
4
4
4
4
And rect() = rect() in general. Therefore

+1
1
1
1
() = rect
+ rect

4
4
4
4
for 0 1 with sketch (Fig. 12) valid in (0 1) For all , we thus have,

Figure 12:

13

1
1

0 1
2 rect( 2 )
1
() = 2 ()
= 0

0
0 or 1
14. (a) See the sketch in Fig. 13.

Figure 13:
(b) () can be computed indirectly without computing (), but it is easier to do part
(c) first and then come back to (b)
(c) For 1: no real solutions to () = 0. For = 1: [ = 1] = [ 1] =
[ 1] = 12 erf(1) $ 0159, so () = 0159( + 1) For 1 1: () = ,
= 0, or = . Therefore () = () For = 1: [ = 1] = [ 1] ' 0159,
so () = 0159( 1) For 1: no real solutions to () = 0 exist. Putting
all these pieces together we get:
1

2
() = 0159( + 1) + 2 rect( ) + 0159( 1)
2
2
which is sketched in Fig. 14. (b) Now that we have (), we obtain () from

Figure 14:

() =

()

= 0159( + 1) + 0159( 1)

erf(1) erf() 1 0
+
erf(1) + erf()
0 1
Note that this last part, involving the erf function is to be added to the step functions
on the line above. Another way of writing the solution comes from noticing that for

14
1 +1 the distribution function of must agree with that of . Thus in this
region, we have the total answer () = () = 05+erf() Then we can write the
solution as

1
1

05 + erf() 1 +1
() =

0
1
which is the same as that a few lines above.
provided in Fig. 15.

A sketch of the distribution function is

Figure 15:
15.
() , [ ]

= [ ]

= [ ]

= 1 [ ] since is continuous,

= 1 ( )

= 1 [

So, upon dierentiation

() =
=
=

( )
2

2 2 + ( )2
1

2 + ()2

16.

= sec
1
=
cos

15

For 1 there are two roots in the interval ( +)

()
1
=
cos
=
= cos1 (1)

The two roots are

1 = + cos1 (1) and
Also () = (cos )1

to show that | 0 ( ())| =

0 () =

sin

cos2

p
2 1 Hence
() =

2 = cos1 (1)

Can then use the right triangle

2
X
( ())
| 0 ( ())|
=1

1
1

2
|| 1

For 1 there are again two roots and we actually obtain the same result as above.
For 1 +1 there are no solutions to () = 0 thus () = 0 there.

Overall we have the solution

() =

1
1
|| 2 1

|| 1
else.

17. Since the random variables and are independent, we can find the density of via
convolution of the two uniform densities and thus
() = () ()
Z +
( ) ()
=

16
Here, by the problem statement : (1 +1) and hence
() =

1
2

1 +1
0
else.

Similarly : (2 +2) and so

() =

1
4

2 +2
0
else.

Computing the convolution graphically, we see that the resulting function will be constant
when the short pulse is completely contained inside the longer pulse and that this
will occur for 1 +1 for which the area is easily computed as 12 14 2 = 14 From
graphical considerations, we can also easily see that the output function must be zero
when the two pulses do not overlap, and that this will occur for all || 3 For 3 || 1
we then just connect these result together via straight lines, to obtain

() =

which is graphed as Fig. 16.

0
3
( + 3) 3 1
1
1 +1
4
1
8 ( + 3) +1 +3
0
+3
1
8

Figure 16:

18.

We wish to calculate the pdf of , , given that and are independent and
exponentially distributed as () = () = exp()() Our approach is to first find
the density of , and then make use of the convolution of densities property to find
the density of = + Now [ ] = [ ] = 1 (), where we have made
use of the fact that is a continuous random variable. Therefore
()

(1 ())
=

= ()

() ,

17
Then, since and are independent,
Z +
() ( )
() =

Z +
=
() (( ))

Z +
=
() ( )

Z +
=
()() ( )

Z +
()
=
max(0)
+

()

max(0)
2 +

=
=
=
=

max(0)

1
2
2 +
1 2

+ 2

0

0
0

0
2

exp(||)
+
2

19. (a) The joint pdf ( ) is given as

( ) =

( )
1
1

( + ) (2 )
3
3
( )

with the transformation and inverse given as

= ( ) = 13 ( + )
= ( ) = 13 (2 )

=+
= 2

We proceed to evaluate the absolute value of the Jacobian as

1


1

1 2 1

3
3

|| = det = det 2
1 = =

9
9
3
3
3

Thus we finally obtain

1
( ) =
3

1
1
( + ) (2 )
3
3

(b) We start by noting that is just the of part (a). Hence () is just the marginal
density for ( ) where

1
1
1
( ) =
( + ) (2 )
3
3
3

18
Combining we have
() =

Upon the substitution =

1
=
3

1
3

1
1
( + ) (2 )
3
3

( + ) inside the integral, we get

and

1
1
(2 ) = (2 (3 )) =
3
3

so that the integral expression then becomes

() =

( )

To get the desired result, we need that and be independent RVs, because then
the joint density ( ) will factor into () ( ), which is the desired
integrand.
20. In this problem
() =

|| 1
0
else,

so the range of is 0 1 For 0 1 i.e. when the first inequality is strict, there is
only one root root to () = = 0: at 1 = Hence
() =

( )
(1 )
= 1

1

for 0 1.

Since () = () it follows that

() = (1 )
= ()
=

there, i.e. for 0 1

However, according to the mapping () the event { = 0} = { 0} { 1} so

[ = 0] = [ 0] + [ 1]
Z

= 0+
1

Thus the complete answer for the pdf of is:

() = 1 () + [() ( 1)]

' 0368() + [() ( 1)]

Labeled sketches of the CDF and pdf follow:

19

21. The only real roots occur when 1 +1 and, as can be seen from the diagram,
these occur at = + 2 for = 2 1 0 1 2 Since () = () and
| 0 ())|= = 1 we obtain
+
X

() =

(+2) rect(2)( + 2)

where rect() , (+ 12 )( 12 )) We note that the product rect(2)( +2) =rect(2)

for 1 For 0, this product is zero. At = 0, only half of rect(2) is overlapped by
() Hence, for = 0, the product is rect( 12 ). Thus,
() =

=1

1
(+2) rect(2) + rect( )
2

To show that the pdf is a legitimate pdf, we note first that it is non-negative, and next
that
Z +
Z +1
Z +1

X
() =
2
+

=1

2 ( 1 ) + (1 1 )

=1
1
1 ( 1 ) + (1 1 )
=
1 2

=
1 ( 1 ) + (1 1 )
1
= ( 1 ) + (1 1 )

= 1 as required for a legitimate pdf.

22. Let and +1 be two numbers generated in sequence by the random number generator
at the th construct. Form = + +1 . If the { } are generated independently, the
pdf of is () = () +1 () and looks like the sketch in Fig. 17. So the form is
right but the mean is wrong! Therefore generate = 1. The pdf of looks like the
sketch in Fig. 18. Equivalently one could subtract 05 from and +1 before adding. In
either case one can get () =tri(2). Here the triangle function tri is defined as

1 2|| 12 + 12
tri() ,
0
else.

20

Figure 17:

Figure 18:

23. We present a MATLAB function histonorm as follows:

function [alpha] = histonorm( samples, sumrange, bins );

sum = zeros( 1, samples );
for i = 1 : bins
x(i) = i;
end
for i = 1 : samples
for j = 1 : sumrange
sum(i) = sum(i) + rand;
end
sum(i) = sum(i) - 6;
end
[z, x] = hist( sum, bins );
plot( x, z);
title( [histogram of , num2str(samples), samples] );
xlabel( bin number )
ylabel( frequency )

The multiple figures in Fig. ?? show the output of histonorm for 10,000 with "smart binning"
on the left and "dumb binning" on the right.

21

Note that the dumb binning result on the right requires a slight modification of the function
as indicated in the comments inside it. The smart binning result which is the default only
plots the domain where the bins are not empty.
While only 10,000 runs were requested in the problem statement, we next show in Fig. ?? the
result of a larger number of trials 100,000 (Fig. ?? and ??), and also a smaller number 1000
trials (Fig. ?? and ??). We can see that the histogram begins to look strongly Gaussian
at the high value of 100,000 trials. A better idea of its Gaussian-ness could be obtained by
viewing a log plot, which is not done here however.

24. (a) We have to calculate the running integral of the density () =

Z
() =
()
Z

exp(||)
=
2

exp(||) Now

22
Because of the absolute value sign, it is easier to consider the two cases 0 and 0
separately. First we evaluate for 0 where () = 2 exp(+) We find
Z

() =
exp()
2

=
exp()|
2
1
exp ()
for 0
=
2
Now we consider the case 0 where () = 2 exp() We note that by symmetry
R0
2 exp() = 12 so we can write
Z

() =
exp(||)
2
Z

1
+
exp()
=
2
0 2

1 1

=
+
exp()|0
2 2
1 1
+ (1 exp ())
=
2 2
1
= 1 exp () for 0
2
Now, as a check, we note that both results agree at their common point = 0 as they
should. Overall, we can write the Laplacian distribution function as

1
0
2 exp ()
() =
1
1 2 exp () 0
(b) Probably the first thing to do here is to note that since : [0 1], we have that
() = {() ( 1)} i.e. just a straight line segment with slope 1 on [0 1] Since
the distribution function is monotone increasing, we have
() = [ 1 ()]
Next, we note that since = 1 so 1 = and hence
() = [ 1 ()]
= ( ())

= () (( ()) ( () 1))

= () (1 0)

= ()

Note: Although it was not asked for in this problem, to actually use this method on a
computer, we would need to calculate (). It is given on (0 1) as
1
0 12
ln 2
() =
1
1
1
ln 2(1) 2 1

23
(c) This method strictly speaking will not work with either jumps or flat regions in the
desired distribution function In a flat region of the corresponding would be a
vertical line, not be a valid function! At a jump of = 1 will not be defined for
some of the input values. This wont work either. On the other hand, there are simple
modifications of this method that can get around these problems and make the basic
method useful in both cases. One simply has to remove the flat regions from before
finding the inverse function. At the jumps, where the inverse function would have a
gap, just fill it in with a horizontal line. With these changes the basic method extends
to both mixed and discrete distribution functions.
25. First we solve () is as follows. Let = , and () = () = (). Let

, , then [ ] = [ ] = 1 (), and so () =

= ().
Therefore
() = ()
Now we can write = + , so that we have the density of given as the convolution
Z
() ( )
() =

Evaluating for 0 (Fig. 19), we get

Figure 19:
() =
=

=
For 0, we get
() =

() ( )
2 2

2
Z

2 2

Combining these two results, we get the formula valid for all

() = ||
2
as found in problem 3.18.

Now let
, || =

0;
0

24

Figure 20:
Consider = () = || (Fig. 20): At root 1 : 0 ()| = 1; At root 2 : 0 ()| = 1.
Then

()1 + ()1 for 0;

() =
0
for 0
Hence if () = () as it is for here, then finally

( ) = 2 ()()
= ()

26. We are told here that and are continuous random variables and are independent. We
are asked for the distribution function and probability density function of , min( ).
For the distribution function, we write
() , [ ]

= 1 [ ]

= 1 [ ] [ ]

= 1 (1 [ ]) (1 [ ])
= 1 (1 () (1 ())

= () + () () ()
For the pdf, we can write
()

( () + () () ())
=

= () + () () () () ()

() ,

For and uniformly distributed as [0 1], we then get,

() = + 2 for [0 1] or generally,

2 2 0 1
=
0
else.
The pdf becomes,
() = 1 + 1 2 for [0 1] or generally,

2 2 0 1
=
0
else.

25
Here are the plots:
() = 2 2 (Fig. 21):

Figure 21:
() = 2 2 (Fig. 22):

Figure 22:
Next we repeat the plots for the case where and are independent and exponentially
distributed with parameter ( 0) i.e. () = () = exp()() Plugging into the
equations, we get () = (1 exp(2))() and () = 2 exp(2)() We notice
that, unlike the case of the two uniform random variables, the minimum of two independent
exponential random variables remains exponential, but with twice the parameter value. The
plots then become, for = 2
() = 4 exp(4) (Fig. 23):
() = 1 exp(4) (Fig. 24):
27.
() = 1 2 ( )
= 1 ()2 ()
= [1 exp()] [1 exp()] 2 ()

= [1 exp()]2 ()
(1 )2 ()

26

Figure 23:

Figure 24:

() =
=
=
=
=

()

[1 exp()]2 ()

[1 exp()]2 ()

2
exp() [1 exp()] ()

(1 )()

This problem has an interpretation for the time to failure of two independent machines. If
we regard the two random variables 1 and 2 as waiting times (to failure), then would
be the time till both machines fail. This has clear application in high-reliability computing.

28. The event { } is the same as

{max(1 2 ) }

= {1 2 }

27
Since { } are independent, then
() = 1 ()2 () ()
Since { } are i.i.d.,

1 () = 2 () = = ()

so that
() = (1 ())

=
()
1

29. The event { } = {min(1 2 ) } is identical with

{ } = {1 2 }
Thus
[ ] = [{1 2 }]
= [1 ] [2 ] [ ]

= (1 1 ())(1 2 ()) (1 ())

Because the { } are i.i.d., this becomes
[ ] = (1 1 ())
and so for the complementary event { }, we have
() = 1 (1 1 ())
30. , max(1 2 ) and the s are independent RVs. Then
() = [ ]

= [1 ] [2 ] [ ]
= ( ())

= (1 ) ()
Hence
()

= ((1 )1 ()

() =

The peak of this curve will occur at

0 = 0 () =
= ( 1)(1 )2 2 (1 )1

= ( 1) (1 )

28
which happens at = 1 or = ln() For = 3, ' 11 and 3 ( ) ' 0444 See
sketch below for = 3.

As for the minimum, we have then , max(1 2 ), where again the s are
independent RVs. Then
() = [ ]

= 1 [ ]

= 1 [1 ] [2 ] [ ]

= 1 (1 1 ())(1 2 ()) (1 ())

= 1 [1 (1 )()]

0
0
=

(1 ) 0
= (1 )()

So the pdf is given as

() = 0 ()
= ()
Here is a sketch for = 1 2 3.

31. We are given = + where are independent and identically distributed (i.i.d.)

29
with : (1 +1) : Then
() = () ()

= () ()

= rect( ) rect( )
2
2

1 ||

rect( )
2
4
4

with sketch:

Next, consider

0
() =
0 0

with sketch:

Now, the real roots of () = 0 are just one = 2 0 and no real roots when 0.
Hence
( 2 )

() = 0
| ()|= 2
where 0 () =

and | 0 ( 2 )| =

1
2

() =

Thus, we have

3
2 0 2
0
else.

Finally, for = 0, we have the probability mass

[ = 0] = [ 0]
Z 0
()
=
2

Z 0
1 ||

=
4
2 2
1
=

30
We can write this as a delta function in the pdf as

3
1
[() ( 2)]
() = () +
2
2
32. We have from the problem statement that
= 1 + 2
= 10 + 20 with 0 , for = 1 2
The pdfs of the primed variables become
1
(0 )

and since the and hence the 0 are independent, we will have
0 (0 ) =

() = 10 () 20 ()
1
=
() 2 ()
2 1
Now we are given that the are distributed as [0 ], thus () = 1 [() ( )]
thus

1
[( ) ( )]
10 () =

1
[() ( )] since 0
=

and so
1
1
[() ( )]
[() ( )]
() =

Performing this convolution, we get a triangle with support [0 2] on the axis and height
1We can then write down the pdf of as

()2

1
1
() =
2 2

0
else.
In this problem, we are given = 110 and = 100 so that = 10 and
1
1
[() ( 10)] [() ( 10)]
() =
10
10

10

100 1
1
=
2 10 10 20
10
0
else.

We can now calculate the probability that the plane will fly at least 5 hours (area of shaded
region in Fig. 25), i.e.
[ 5] = 1 (5)
Z 5
1

= 1
0 100
1 1 25
= 1
( | )
100 2 0
1
= 1 = 0875
8

since is a continuous random variable,

31

Figure 25:
33. Let = + , then
() = [ = ]

X
() ( )
=
=0

X
1
1
2 3
! ( )!
=0

5
X

2 3
!

= 2 3
=

=0

5
!

5 0

Hence, is Poisson distributed with parameter = + So

[ 5] =

5
X

()

=0

54
55
52 53
+
+
+
=
1+5+
2
6
24 120
' 0616
5

34. This is an invertible transformation, with inverse

=+

and

The joint density is then

e
( ) = ( + ) ||

where the Jacobian of the inverse transformation e is determined as

e
= det

1 1
= det
1 1
= 2

32
Thus we have
e
( ) = ( + ) ||
= 2 ( + )

= 2 ( + ) ( )

2 0 + 1 and 0 1
=
0
else,

since and are independent and uniformly distributed (0 1) We can also write the
answer in terms of the unit pulse function rect() defined as

1 05 +05
rect() ,
0
else
The result can then be written as
( ) = 2 rect( + 05) rect( 05)
35. (a) = + where and are independent, and : (1 +1), i.e. () = 12 rect( 2 )
and : (2 +2), i.e. () = 14 rect( 4 ) Then
() = () ()
Z
( ) ()
=

Z
1

1
=
rect(
) rect( )
2 4
4
2
Z +2
1

=
)
rect(
8 2
2
To perform this convolution, we start in region 1, were 3 and there is no overlap
between the two function supports. Here is the sketch.

Clearly there is no overlap here in region 1, and so the output () = 0 here.

region 2: partial overlap 3 1 Here is the sketch:

Using the cross-hatched area as a guide, we can evaluate the integral as

Z
1 +1
() =

8 2
1
=
( + 3) 3 1
8

33
Next comes region 3, complete overlap, sketched as

region 3: 1 +1 From the sketch, we can see that the integral must start at
1 and go to + 1 in this region. Thus
Z
1 +1

() =
8 1
1
=

4
Next comes region 4, partial overlap on the right, where 1 2 and + 1 2 which
together imply 1 3.
region 4: 1 3 Here
Z
1 2

() =
8 1
1
(3 ) 1 3
=
8
Finally comes region 5, no overlap on right.
region 5: 3
() = 0
Our complete answer can be written as

(
+
3)
8
1
() =
4

(3 )

3
3 1
1 +1
1 3
3

with sketch:

As a check, we note that the total area is 2( 12 (2) 14 ) + 2( 14 ) =

1
2

1
2

= 1

(b) Here = 2 where again and are independent. We let = 2 and

= to make use of convolution of densities. Now () = [ ] = [

2 ] = ( 2 ) = () = 2 ( 2 ) and () = [ ] = [ ] = 1 ()
= () = () Hence

1
() = rect( ) and
4
4

1
() = rect( ),
4
4

34
with sketches

so
() = () ()

1
rect( ) rect( )
=
8
4
4

1
triag( )
=
4
4
where function triag is given as
(1 ||) || 1
0
|| 1

triag() ,
A sketch of this is given as

36. We look at the transformation problem for two independent Normal random variables and
: (0 2 ) transformed to , 2 + 2 and , We thus have
= ( ) = 2 + 2

and = ( ) =

This is a non-invertible transformation with two real roots, for ||

1 :
2 :

p
= = + 2 , and
p
= = 2

Now at both roots the magnitude of the Jacobian is the same,

p
|1 | = |2 | = 2 2 where
"
#

12 = det = 2 = 2 2

35
Hence

p
p
1

( 2 ) + ( 2 )
2 2
(
1 1
2 ) || 0
exp(2
2
2
2

=
0
else.

( ) =

We can find the marginal density either by integrating out the unwanted variable in this
joint density, or by using the result of Example 3.3-10 that will be Exponential distributed
(equivalently Chi-square with 2 degrees of freedom). Either way the answer is
() =

( )

=
=

Z +
1
1

exp(2 2 )

2
2
2
1
exp(2 2 ) ()
2 2

37. We want to find the pdf of the two variables

= +
= +
The joint pdf of and are given as
( ) =
where ( ) =

1
[2
22 (12 )

2 2

1
p
()
2
1

2 + 2 ]. Consider the inverse transformation, i.e.,

=
+

= +

Note that from the above transformation, the solution of the two equations are given as =
The term in the exponent of the pdf will be 2 1 2 [2 2 + 2 ],

+ and = + .
2 (1 )
and this is given as
h
i
1
1
2
2
)2 2(
)(
+ (
2
[
(

+
2+
]
=

+
)

+
)
2 2 (1 2 )
2 2 (1 2 )
In this exponent, if the cross terms (terms that contain ) vanish, then we would be able to
split the pdf in to the product of the two marginal pdfs. In other words, if the coecients
of the terms is zero, then we would be able to write = . Therefore, we need
(2
+ 2
2
2
) = 0

36
the coecient of will be zero. Then
. If we chose
= = ,
=
+

will give us
( ) =

1
2

1
[2 + 2 2] = 22 (1
2 2 + 2(1 + )2 2 ]. Therefore,
2 ) [2(1 )

i
i
h

2
2

1
1+

. The magnitude of the Jacobian is given as

+

2 (12 )

2h

where ( ) =

+
2
,

( ) =

|| = mag

.
2

=
2|
|

With 1 , 1 2
, 2 , 1 + 2. Hence,
2

1 2
1 2
1
1
( ) =
2 1
2 2
2 1
22
= () ()

38. Let ( ) , +
and ( ) ,
2
2 . The
0 || || and are four in number.

1 = + +
2 = +
3 = +
4 = + +

real roots of ( ) = , ( ) = occur for

1 = +
2 = +
3 =
4 =

We note that can be negative, but never greater in magnitude than . The magnitude of
is

p

abs = 2|| = 2 2 2

and this is the same for all the four roots. Now observe: 2 + 2 = 2 for all the roots, and
p
1 1 = 2 2 = 3 3
p
2 2 = 2 2 = 4 4

Hence

( ) =
=

2
2
22 2

p
2

4 2 2 1 2

2(1 )

1

p
12 cosh
(1 2 )(2 2 )

2
2
2+2 2

2(1 )

!
2 2
for 0 || ||
1 2

where the hyperbolic cosine function cosh() , 12 ( + ). For 0 or || ||, there

are no real roots of the transformation equations, so that we have ( ) = 0 there.

37
39. From Eq. 3.4-4 in Example 3.4-1, we have the linear transformation
= + =
with the one solution, i.e. = 1:
= ( ) =

+
= ( ) =

2
2

Proceeding with the direct approach, the matrix in the Jacobian e becomes:

1
1

2
2
= 1

2 2

with determinant or Jacobian given as

e
=

1
2
1
2

1
2
12

= 1

So, the magnitude of the Jacobian is 12 Then by Eq. 3.4-11, we can write

+
e

( ) = ||
2
2

1
+
=

2
2
2
which agrees with Eq. 3.4-4 that was derived in the Example 3.4-1 using the indirect method.
40. Here the transformation is give as
= ( ) , cos + sin

(1)

= ( ) , sin cos

(2)

The random variables and are independent Normal distributed as (0 1) To find the
inverse, we can multiply (1) by cos and then multiply (2) by sin and add the results to get
= cos + sin
since cos2 + sin2 = 1 Similarly, we multiply (1) by sin and multiply (2) by cos and
subtract to get
= sin cos
Thus there is a unique inverse and one root at ( ) = ( cos + sin sin cos )
Calculating the Jacobian, we get

cos sin
= det
= 1 so that || = 1
sin cos
The transformed pdf then becomes,
( ) = ( cos + sin sin cos ) 1

1
1
2
2
exp ( cos + sin ) + ( sin cos )
=
2
2

1
1
+
exp 2 + 2 ,
=
2
2

38
This independent Normal joint density is thus unchanged by this transformation! Note that
this transformation is closely related to a rotation of Cartesian coordinates in the plane.
In fact, if were replaced in the transformation by , the transformation would be a
coordinate rotation by angle + in the plane.
41. We look at the transformation problem for two independent Normal random variables and
: (0 2 ) transformed to , 2 + 2 and , 2 We thus have
= ( ) = 2 + 2

and = ( ) = 2

This is a non-invertible transformation with two real roots, for ||

r
2

= , and
1 : = +
2
2
r
2

2 : =
=
2
2

Now at both roots the magnitude of the Jacobian is the same,

r
2
|1 | = |2 | = 4
where
2
"
#
r

12 = det = 4 = 4
2

Hence
( ) =

q
2
4 2
( 1
1
4 2

!
r
2
2
) + (
)
(
2
2
2
2

2
(
2)

exp(2 2 ) | 2 |

else.

42. First we determine the value of

1

=
=
=

+ Z +

( )

Z 1 Z +1

(2 + 2 )
0
1

Z 1 Z +1
2
2

( + )

3
=
4

Hence ( ) = 34 (2 + 2 ) with support on 0 1 1 +1

(a) i) For 0 1

( ) = 0

39
i) For 1 1
( ) = 1
iii)

0 1 1
( ) =
=
=
=
=

iv) For 0 1 1 1
( ) =
=
v) For 1 1 1
( ) =
=

3
4

Z +1

( + )
"
+1 #
3
3 3
+
4 3
3 1
3
3

+
4 3
3
3
3

3 2 3 2
+
4 3
3
1 3
( + )
2
0

Z Z
3
2
2
( + )
4 0
1

1 3
( + 1) + (3 + 1)
4

Z Z
3 1
2
2
( + )
4 0
1

1
( + 1) + ( 3 + 1)
4

43. (a) The range of is a subset of the real line 1 if is a real-valued random variable.
(b) A reasonable probability space for is ( F ) where is the sample space of the
underlying experiment, F is the field of events defined on , and is the set function that
assigns to every set (event) F the number [] A reasonable probability space for
is ( B ) where , the range of , is the induced sample space under the mapping
, B is the Borel field of events over 1 and is a set function that assigns to every set
B, the number [].
(c) The event {| () } under the mapping is the set ( ] 1 .
(d) In the I/O viewpoint, the inverse image is computed as follows:
{ } = {2 + 3 }

3
=

3
=

40
44. Define the event = {max(1 2 ) }. Then
( ) = [ |] [] + [ | ] [ ]
Now
[] = [1 ] [2 ] by independence,
= (1 )()(1 )()
= (1 )2 ()

and so [ ] = 1 (1 )2 () Observe that = 0 at given event , while = at

given event . So
[ |] = () and [ | ] = ()
Hence,
( ) = [ |] [] + [ | ] [ ]

= ()(1 )2 () + ()(1 (1 )2 ())

= (1 )2 ()() + ()(1 (1 )2 ())

For any fixed 0 and = we see ( ) = (1 )2 + 1(1 (1 )2 ) = 1 and

for = and any fixed , we have ( ) = () Does this make sense? Think about
it!