You are on page 1of 10

ORFE 526 - Probability

Problem Set 4
Theo Gutman-Solo

October 14, 2014

Definitions
1. A normed vector space V is a vector space with an operation || || :
V R called the norm which satisfies
Postitive definiteness, x V ||V || 0, with equality only holding for x = 0.
C, X V ||X|| = |a|||X||1
Triangle inequality x, y V , ||x + y|| ||x|| + ||y||
2. An inner product space is a vector space together with an operation
< , >: V V C satisfying the following properties
Congujate symmetry, < x, y >= < y,x >
Linearity in first argument < x + y, z >=< x, z > + < y, z >
Positive definiteness < x, x > 0, with equality if and only if
x=0
3. A metric space is a set X with an operation d : X X R called the
metric which satisfies
Symmetry d(x, y) = d(y, x)
Positive definiteness d(x, y) 0, with equality if and only if x = y
Triangle inequality, d(x, y) d(x, z) + d(z, y)
1

Note this assumes the vector space is defined over a subfield of the complex numbers

The space is complete with respect to the metric d if every cauchy


sequence converges to a point inside X.
4. A banach space is a complete normed vector space X, in other words
for every cauchy sequence xn X, there exists a point x X such
that limn ||xn x|| = 0. (|| || is the norm of the space)
5. A hilbert space is an inner product space that is also a complete metric
space with respect to the norm induced by the inner product. d(x, y) =

< x y, x y >

Properties of CDFs

2.1
1.



(a1 , b1 ] (a2 , b2 ] = (a1 , b1 ] (a2 , b2 ] + (, a1 ] (, b2 ]


+ (, b1 ] (, a2 ] (, b1 ] (, a2 ]

(, b1 ] (, a2 ]


= (a1 , b1 ] (a2 , b2 ] + (, a1 ] (, b2 ]


+ (a1 , b1 ] (, a2 ] + (, a1 ] (, a2 ]


(, b1 ] (, a2 ] (, b1 ] (, a2 ]
= F (a1 , a2 ) F (a1 , b2 ) F (b1 , a2 )


+ (a1 , b1 ] (a2 , b2 ] + (, a1 ] (, b2 ]

+ (a1 , b1 ] (, a2 ]
= F (a1 , a2 ) F (a1 , b2 ) F (b1 , a2 )


+ (a1 , b1 ] (, b2 ] + (, a1 ] (, b2 ]
= F (a1 , a2 ) F (a1 , b2 ) F (b1 , a2 )

+ (, b1 ] (, b2 ]
= F (a1 , a2 ) F (a1 , b2 ) F (b1 , a2 ) + F (b1 , b2 )
2. This follows directly from continuity from above. Note that since the
series xk is monotonically decreasing to x this implies that i
i+1
i
i
(, xi+1
1 ] (, x2 ] (, x1 ] (, x2 ]

This means that we have a monotonically decreasing sequence and


assuming there exists a N such that F (xa ) < we can apply
2

continuity from above as follows.




lim F (xk ) = lim F (xk+a ) = lim (, x1k+a ] (, xk+a
]
2
k
k
k
!

\
= lim
(, x1k+a ] (, xk+a
]
2
k

i=1


= (, x1 ] (, x2 ]
If there doesnt exist an a N such that F (xa ) < , then F (xk ) =
for all k N, This means that F (x) = so the identity still holds.
3. Follows directly from continuity from below. If (xk ) is a monotonically increasing sequence it follows that k, (, xk1 ] (, xk2 ]
k+1
(, xk+1
1 ] (, x2 ]. This is an ascending family of sets.


lim F (xk ) = lim (, xk1 ] (, xk2 ]
k
k
!

[
= lim
(, xk1 ] (, xk2 ]
k

k=1

= ((, ) (, ))
=
4. This follows directly from continuity from at . Assume x2n
as n . Then i, (, xi1 ] (, xi2 ] (, x11 ] (, xi2 ],
this immediately implies that i, F (xi1 , xi2 ) F (x11 , xi2 ). Let Si =
(, x11 ] (, x2i ], it is clear that i, Si+1 Si . Its also clear that

Si =

i=1

Since for any point (a, b) because xn2 there exists an xn2 < b
which implies that (a, b)
/ Sn and as a result isnt a member of the
intersection. Finally
!
n
\

n
1 n
1
i
lim F (x ) lim F (x1 , x2 ) = lim
(, x1 ] (, x2 ]
n

lim
n

(, x11 ]

i=1

(, xi2 ]

0
However since the CDF is nonnegative this immediately implies that
lim F (xn ) = 0
n

2.2
These properties follow simply from the subadditivity of measures.
x1 x2 (, x1 ] (, y] (, x2 ] (, y)
((, x1 ] (, y]) ((, x2 ] (, y])
F (x1 , y) F (x2 , y)
y1 y2 (, x] (, y1 ] (, x] (, y2 )
((, x] (, y1 ]) ((, x] (, y2 ])
F (x, y1 ) F (x, y2 )

2.3

1 : x0
1 : x < 0, y 0
G(x, y) =

0 : x < 0, y < 0
This function is clearly increasing in x and y, but consider points a =
(1, 1) and b = (1, 1).
G(1, 1) G(1, 1) G(1, 1) + G(0, 0) = 1 1 1 + 0 = 1

Linearity of the Lebesgue Integral on simple functions

For simple integrable functions f and g the lebesgue integral is a linear


functional. Note we can represent simple functions in a convenient manner.
f=

n
X

xi 1Ei

i=1

Where (x1 , x2 , ..., xn ) = range(f ) which is finite and Ei = f 1 (xi ). It is


important to note that the collection Ei is pairwise disjoint and its union
spans the domain. Let fi be the i-th element in range(f ), and Fi = f 1 (fi ).
gj and Gj are defined analogously.
Next consider the set collection {Fi Gj }ij . This set is clearly pairwise
disjoint and f + g is obviously constant over these sets. Finally

n [
m
n
m
n
[
[
[
[
Fi
(Fi Gi ) =
Gi =
Fi =
i=1 j=1

i=1

j=1

i=1

We conclude that since (Fi Gj )ij are pairwise disjoint, span the space
omega and (f + g) is constant on each element we can represent (f + g) as
a sum of simple functions indexed by (Fi Gj )ij .
Z
(f + g) =

n X
m
X
i=1 j=1
n X
m
X

(fi + gj )(Fi Gi )
fi (Fi Gi ) + gj (Fi Gi )

i=1 j=1
n X
m
X

fi (Fi
i=1 j=1
n
m
X
X

(Fi Gi ) +

fi

i=1

j=1

n
X

n
X

m
[

Fi Gi +

fi (Fi ) +

i=1

m
X

Gi )

(Fi Gi )

gj

j=1

Z
=

gj (Fi
i=1 j=1
m
n
X
X
j=1

fi

i=1

Gi ) +

n X
m
X

i=1

gj

n
[

!
Ei Fi

j=1
n
X

gi (Gi )

j=1

Z
f+

Properties of the lebesgue integral

In the following problems I have assumed that all functions are non-negative
for simplicity. This does not cause any loss of generality as all functions can
be written as the difference of non-negative functions. Everything is assumed
to be on a measureable space (, F, )

4.1
Z

d : is simple, 0 cf
( n
)
X
= sup
wi (Wi ) : n N, i, x Wi F 0 wi cf (x)

cf d = sup

i=1
n
X

= sup c

)
wi (Wi ) : n N, i, x Wi F 0 wi f (x)

i=1

= c sup

( n
X

)!
wi (Wi ) : n N, i, x Wi F 0 wi f (x)

i=1


Z

= c sup
d : is simple, 0 f
Z
= c f d

4.2
Z

Z
f d

Z
gd

(f g)d 0
Z

(max[f g, 0] max[g f, 0])d 0


Z
max[f g, 0]d max[g f, 0]d 0

However using the assumption that [f < g] = 0. The second integral on


the last line must be zero. This is because it is suprenum of simple functions
of the form
X
ai 1Ai : i, ai 6= 0 Ai {x : (g(x) f (x)) > 0}
But this implies that all these simple functions integrate identically to zero.
Every term in the sum of the simple function vanishes because either the
coefficient ai is zero or (Ai ) is zero since it is the subset of {x : (f (x) <
g(x))} which is null by assupmtion.

4.3
This follows from the contrapositive. Assume that [f 6= 0] = [|f | =
6 0] 6= 0,
this means that for some n N the set Cn = {x : |f (x)| 1/n} has positive
measure. This follows because the limit of the measure of these sets as
6

n = {x : f (x) =
6 0} which is nonzero by assumption. If C is the set
mentioned above we have
Z
Z
Z
1
1
1
1C |f |
1C |f | 0 < (1C ) |f |
n
n
n

5
Let (, F, ) be a measure, (E, E) a measureable space and f : (, F)
(E, E) a measureable function.

5.1

Pushforward of Measure

f [B] = [f 1 (B)] defines a measure on (E, E). First note f [B] = [f 1 (B)]
0 under the assumption that is a measure, which implies non-negativity.
Next the empty set is a null set, this follows trivially. f [] = [f 1 ()] =
[] = 0. Finally f satisfies - additivity. This follows trivially from the
property of the inverse function.
!

[
[
1
f
Bi =
f 1 (Bi )
i=1

i=1

f 1 (A) f 1 (B) = A B =
This implies that
f

!
Bi

= f 1

i=1

!!
Bi

i=1

(f 1 (bi )) =

i=1

!
f 1 (Bi )

i=1

f (Bi )

i=1

Taking these facts together we conclude that f is a measure.

5.2
Choose an increasing sequence of simple functions n : E R that converge
pointwise to g so that
Z
Z
lim
n df =
gdf
n E

Such a sequence is guaranteed to exist since g is measureable by assumption.


n =

mn
X

zin 1Zin

i=1

Using the elements of the original sequence create a new sequence n :


R
mn
X
n =
zin 1f 1 (Zin )
i=1

This series is clearly increasing. It also follows immediately that the series
n converges pointwise to g f .



lim n (x) = lim n f (x) = g f (x)
n

It is easy to verify that for all n these simple functions have the same integral.
Z
mn
X
n df =
zin f [Zin ]
E

i=1

mn
X

zin [f 1 (Zin )]

Zi=1
= n d
This immediately implies that
Z
Z
lim
n d = lim
n df
n

n E

Finally the monotone convergence theorem implies that because n and n


converge to g f and g respectively
Z
Z
Z
Z
g f d = lim
n d = lim
n df =
gdf

n E

6
6.1

Right Continuity of the CDF and quintile function

Right continuity of the CDF follows directly from continuity from above
property of a measure.
!

\
1
lim F (x) = lim P (X x) = P
, x +
= lim P ((, x])
xa
xa
i
n
i=1

Finally to justify the use of continuity from above note that because P is a
probability measure P ((, x + 1]) < .
Right continuity of the Quantile Function follows from right continuity of
the measure.
lim q(x) = sup{x R : F (x) x} = sup{x R : F (x) a} = q(a)

xa+

6.2
This follows from the fact that as we take a finer and finer partition the term
[F (xi+1 F (xi )] in the stieltjes integral starts to approximate F 0 (Xi )(xi+1
xi ) which gives a riemann partition of the space.

6.3
This follows by integration by parts.
Z
Z 0
(P [X > x] 1)dx +
(P [X > x])dx


(1 CDF (x)) 1 dx +

Z 0

(1 CDF (X))dx
0


CDF (x) dx +

(1 CDF (X))dx
0

|(CDF (x)x)]0
Z

Z
xp(x)dx +

xp(x)dx + |(1 CDF (x))x]


0

xp(x)dx
0

xp(x)dx
0

xp(x)dx

E[X]

6.4
[u (0, 1) : u F (X) 1] = [(0, F (X))] = F (X) 0 = F (X)
Next note that q is clearly a strictly increasing function, because the the set
{x R : F (x) u} can only gain elements as u increases.

6.5
F (x) is uniformly distributed on (0, 1) if X is continuous (and thus invertible). Let U be distributed according to F (X). For x (0, 1)
P (U x) = P (F (X) x) = P (X F 1 (x)) = F (F 1 (x)) = x

10

You might also like