Tutorial 02
1 .
Cunivariate) Distribution function .
Any F :
/R -
> [0 1) , is a distribution function (d .
f) if :
.
i) F(-0) = im F(x) = 0 ; F() = F(x) = 1 .
X -
> 8
ii) FX
iii) i right continuous.
.
2 Characterization 11.
of probability measure on
Pinduces F IP F(x) 1P((
prob then x3)
: = 0
a . measure
-
.
,
is a d .
f . and satisfies X =
IP.
Finduces I :
F df then PEXF is
prob
a
.
..
a measure
and satisfies (P)c -
2, x)) =
F(x) ·
3 .
Generalized inverse & quantile function
If F : IR- 11 " ,
the
generalized inverse F
-
of F is
defined by
"
F (y) =
inf(x-1R :
F(x) = y3 , Y GR
If F is a d .
f ..
then F" is the quantile function .
4
. Multivariate d .
J
.
Any F :
IRd - > [0 1), is a
joint/multivariate dif -
if
is him F(x) = 0
Hj = 1 , ..., d;
Min F (x) =
1
Xj + C
ii) F is
d-increasing
iii) F right-continuous
Characterizes 1RM.
prob. measure on
I . Random variable ·
Vector ,
sequence
(22 #) measurable
, a
space , a measurable
function
X: --
> & is called a :
i) random variablee- if d =
1
vi) random rector , if 122
iii) random sequence , in d = 0.
6. Distribution of X
If (2 , F .
1) a
prob space
and N: - >
IRA measurable
,
then IPX(B) =
1P(X"(B) is a
prob. measure on 11R9 B(1RM).
,
i. e
. the distribution of .
X
7
. Connection between X
. 1Px and E
X induces 1PX , which is characterized
by .
F
=Laxm iR
=
IP(B)
IPX
8. Quantile transform & Probability transform
i) Let dif
+
F be a .
and UwUnifi . 1) .
Then F (r) - .
F
ii) If X-F , F continuous ,
then F(x) ~Ul 1). .
Exercises
>
-
1. Let X = (X X2) be a
. random vector with d .
f
. It
and F and F2
margins , .
Express IP( > &) in terms
of H .
F and E .
show that
In
particular ,
in
general
| P( * (e) + 1 -
1P( ***)
Sol :
#
.
2 Suppose X and Y are v U's
. . on (2 F IP) - ,
,
and let AE F . Show that if we let
\ XIes,we
2 (r) =
then E is an U U
..
Sol :
It show E measurable
suffices to is , i. e
.
↓ BE FIRD) .
Z"(B) E F .
2 (B) =
[Z" (B) 1H] U [E"(B) 1 As
= IX" (B) 1 H] U [Y"(B) Al
X Y are nu's ,
so X"(B) .
Y"(B) EF
By property of 6-algebra ,
"(B) EF
.
3 Suppose you want to
flip a
fair coin , but
don't have coin-you only have
you a a
X-Uniflo How would do ?
uniform v U
.
.
,
1).
you it
Sal :
distribution,
We want to
sample from a Bernoulli
Bern (2) ,
given X-Unif(0 ,
1) .
The Bernoulli d .
.
f
Yo
Fix =
/2 1-
: o
11 "
4 Ye
I
I Y
0
(y)
=
=
F
+
yeto 1)
.
=
,
,
I , 0 . .
W
Thus F
+
(X) ~ Bern(z)
4 Let
. X be an U V
. . on C .
F 1P), ,
and
F its d .
f. Show that
F continuous at xo > 1P(X = .
X ) = 0
Sal :
We
already have F right continuous ,
thus
F continuous at XoE) F left continuous No
at
E) im F(x) =
F(xi)
X > Xo
-
>
# lim
X
[F(x0) -
F(x] = lim
* >
-
Xo
Px ((X XoS) ,
X -
(continuity from above) = 1PX(X0) =
1P(X =
x0)
= 0
5 .
Let X be Ur.
an with d .
f .
F.
Let Ver The distribution
Unif 10 1)
.
,
generalizal function
of X is defined by ,
F(x x),
(P(X(x) + Xp(X
= =
x)
and the
generalized distribution transform Vof X is.
V F(X v) = ,
Show U w Unif (0 , 1) .
Sol :
First let &2 (X) =
sup[x 1P(X = x) 23
: <
Note we have U if and
only if
(X , v)e((x , x) :
(P(X(x) + x(p(X =
x) =
2) = S
Leei) :
B =
1P(X = (X) So
.
Denote 9 =
IP(X < (X) ,
then
S =
<X < 9(x)U(X 92(x) =
,
9 +
VB22)
Thus IP(U =2) 1P(S) = =
& + BIP(V) : C
Ketii) :
B = 0
| P(V = <) =
|p(X =
q(x)) 1P(X = 92(x) = =
2