stat 712 fa 2021 lec 10 slides transformations of multiple
TRANSCRIPT
STAT 712 fa 2021 Lec 10 slides
Transformations of multiple random variables
Karl B. Gregory
University of South Carolina
These slides are an instructional aid; their sole purpose is to display, during the lecture,definitions, plots, results, etc. which take too much time to write by hand on the blackboard.
They are not intended to explain or expound on any material.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 1 / 15
X X Y X Ä Y Y
Bivariate transformations
1 Bivariate transformations
2 Sums of independent random variables
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 2 / 15
Bivariate transformations
We often wish to find the distribution of a function of two (or more) rvs:
Exercise: Let X1 and X2 be indep. Exponential(�) rvs. Let Y = X1/(X1 + X2).1 Find the cdf of Y .2 Find the pdf of Y .
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 3 / 15
X X f x x te EE I x o xc o
X X e o o x o o
Y E O I
For y E o
Fy y P Y E P EY EDSG.xade.dz
3
P X E y
P X Ll x E Xy
Bivariate transformations
Theorem (Bivariate transformation method)
Let (X1,X2) be a pair of cont. rvs with joint pdf fX1,X2 on X and
Y1 = g1(X1,X2) and Y2 = g2(X1,X2),
where g1 and g2 define a 1:1 transformation of X onto Y (define these).
Let g�11 and g�1
2 be the functions satisfying
y1 = g1(x1, x2)y2 = g2(x1, x2)
() x1 = g�11 (y1, y2)
x2 = g�12 (y1, y2)
for all (x1, x2) 2 X and (y1, y2) 2 Y.
Then the joint pdf of (Y1,Y2) is given by
fY1,Y2(y1, y2) = fX1,X2(g�11 (y1, y2), g
�12 (y1, y2))|J(y1, y2)|, for (y1, y2) 2 Y,
where J(y1, y2) is the Jacobian (next slide), if J(y1, y2) is not always 0 on Y.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 4 / 15
Jacobian
Bivariate transformations
JacobianIn the setup of the previous slide, the Jacobian of the transformation is defined as
J(y1, y2) =
�����
@@y1
g�11 (y1, y2)
@@y2
g�11 (y1, y2)
@@y1
g�12 (y1, y2)
@@y2
g�12 (y1, y2)
����� .
For real numbers a, b, c , d ,����a bc d
���� = ad � bc .
This is called the determinant.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 5 / 15
Bivariate transformations
Exercise: Let X1,X2 be independent Normal(0, 1) rvs.1 Find the joint pdf of Y1 = X1/X2 and Y2 = X2.2 Find the marginal pdf of Y1.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 6 / 15
f laxe t e e C E R R
X 442 Gi Ya YaY G x.x
x2 Ya Gi Y Ya42 X2 Jz _x2
Gilly ya Gilly Yet
sic y Gysi y Ö
fairly je je YI Fu Cy _a ER
eEyi 1yd dieam
f itilitik dye
e
1 4,2 2
G G der
u Yi Ye Fu dy End
feinlitik d
I e
Hilldu
E Ei Ittf
Bivariate transformations
Exercise: Let X1,X2 have joint pdf
fX1,X2(x1, x2) =1�2 exp
�x1 + x2
�
�1(x1 > 0, x2 > 0).
1 Find the joint pdf of Y1 = X1/(X1 + X2) and Y2 = X1 + X2.2 Find the marginal pdf of Y1.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 7 / 15
Bivariate transformations
Exercise: Let X1 ⇠ Beta(1, 1) and X2 ⇠ Beta(2, 1) be independent rvs.1 Find the joint pdf of Y1 = X1X2 and Y2 = X2.2 Find the marginal pdf of Y1.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 8 / 15
K.x.la fürl GE gECi as zcocx.ci
J O X2 2
2 V2 I O LA LI O L 22 2
e
Y E Y_Y O Y Yz 2
x 4 42 5 Y y
Y Xix g a e
y X2 82 x x X2 Y Gj y y
ILY E Ei E0
Yo Ey Yo
Y 2 Ya I E I o Y y a
2 1 o y 42 2
For y Elo
2
July J 2 den
2ydHi
pdf of Beta 2 1 ß 2
Bivariate transformations
Exercise: Let Z1,Z2 have the bivariate Normal distribution, with joint pdf
fZ1,Z2(z1, z2) =12⇡
1p1 � ⇢2
exp
�1
2z21 � 2⇢z1z2 + z2
21 � ⇢2
�.
1 Find the joint pdf of U1 = Z1 + Z2 and U2 = Z1 � Z2.2 Find the marginal pdfs of U1 and U2.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 9 / 15
rho
PI gives correlation between Z Zz
Bivariate transformations
Exercise: Let Z1,Z2 have the bivariate Normal distribution, with joint pdf
fZ1,Z2(z1, z2) =12⇡
1p1 � ⇢2
exp
�1
2z21 � 2⇢z1z2 + z2
21 � ⇢2
�.
1 Find the joint pdf of U1 = min{Z1,Z2} and U2 = max{Z1,Z2}? (Not 1:1).2 Find the marginal pdf of U2 (work on this at home).
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 10 / 15
U I ZEE EE
DE
Fu n P Uz En P m s zieh n
t.tn IGE G
Sums of independent random variables
1 Bivariate transformations
2 Sums of independent random variables
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 11 / 15
Ya f Canal Sy al H _ER
Sums of independent random variables
We now extend our definition of indepence to a collection of more than two rvs.
Mutual independence for a collection of rvsLet the rvs X1, . . . ,Xn have joint pdf/pmf f (x1, . . . , xn) and marginal pdfs/pmfssuch that Xi ⇠ fXi , i = 1, . . . , n. If
f (x1, . . . , xn) =nY
i=1
fXi (xi ) for all (x1, . . . , xn) 2 Rn,
we say that X1, . . . ,Xn are mutually independent rvs.
Expectation of a product of functionsIf X1, . . . ,Xn are mutually independent, then
E(g1(X1) · · · · · gn(Xn)) =Qn
i=1 Egi (Xi )
for any functions g1, . . . , gn : R ! R.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 12 / 15
e
Sums of independent random variables
Theorem (mgf method for sums of independent rvs)Let X1, . . . ,Xn be ind. rvs with mgfs MX1 , . . . ,MXn , resp. Let Y = X1 + · · ·+ Xn.
The mgf of Y is given by
MY (t) =nY
i=1
MXi (t).
Moreover, if X1, . . . ,Xn are ind. and all have mgf MX (are iid), then
MY (t) = [MX (t)]n.
Exercise: Prove the above.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 13 / 15
X Y X Y Multi Mylt Myltl
Sums of independent random variables
Exercise: Let X1, . . . ,Xn be ind. chi-squared rvs with dfs ⌫1, . . . , ⌫n, resp. Findthe distribution of Y = X1 + · · ·+ Xn.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 14 / 15
My t 1 275 s nn
e e
Mylt Mx.lt I 1 2 5
ab.ae abt1 2
E
not of chisquaredwith
df Evi
Sums of independent random variables
Exercise: Let X1, . . . ,Xn be ind. Normal rvs with means µ1, . . . , µn and variances�2
1 , . . . ,�2n , resp.
1 Find the distribution of Y = X1 + · · ·+ Xn.2 Find the distribution of V = a1X1 + · · ·+ anXn, for a1, . . . , an 2 R.3 Find the distribution of X̄n = (X1 + · · ·+ Xn)/n.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 15 / 15
My t eMit Eth
My t My f eMit Eth
in
eE
t E a Ek
not of Normal Eni E E
Arlt Maxtertanx
elakttank t
El el't kt Cent x
a Xr
e
CastleO
EF e
X
editXi
Mx ait
eMilsit t a sit
E si t t E air t z
89 ofNormal Kairo a
Multivariate transformations, including non-1:1
1 Bivariate transformations
2 Multivariate transformations, including non-1:1
3 Sums of independent random variables
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 11 / 22
Multivariate transformations, including non-1:1
We first define the joint pdf/pmf of a collection of d random variables.
Multivariate pdf/pmfThe joint pdf of a d-tuple of continuous rvs (X1, . . . ,Xd) is the functionf : Rd ! [0,1) such that
P((X1, . . . ,Xd) 2 A) =
Z· · ·
Z
Af (x1, . . . , xd)dx1 · · · dxd
for all A 2 B(R).
If X1, . . . ,Xd are discrete, then their joint pmf is the function p given by
p(x1, . . . , xd) = P(X1 = x1, . . . ,Xd = xd) for all (x1, . . . , xd) 2 Rd .
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 12 / 22
d
e
I
2
Multivariate transformations, including non-1:1
We extend our definition of independence to a collection of more than two rvs.
Mutual independence for a collection of rvsLet the rvs X1, . . . ,Xd have joint pdf/pmf f (x1, . . . , xd) and marginal pdfs/pmfssuch that Xi ⇠ fXi , i = 1, . . . , d . If
f (x1, . . . , xd) =Qd
j=1 fXj (xj) for all (x1, . . . , xd) 2 Rd ,
we say that X1, . . . ,Xd are mutually independent rvs.
Theorem (Quick check for mutual independence)The random variables X1, . . . ,Xd with joint pdf/pmf f are mutually independentif and only if there exist functions g1, . . . , gd such that
f (x1, . . . , xd) =Qd
j=1 gj(xj) for all (x1, . . . , xd) 2 Rd .
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 13 / 22
y4 g x hlu
__
Multivariate transformations, including non-1:1
Multivariate transformationLet (X1, . . . ,Xd) have joint pdf f with support A. Consider the rvs
U1 = g1(X1, . . . ,Xd) . . . Ud = gd(X1, . . . ,Xd).
for some functions g1, . . . , gd : Rd ! Rd such that, given a partitionA0,A1, . . . ,Am of A, where P((X1, . . . ,Xd) 2 A0) = 0, the transformation is1 : 1 on each of the sets A1, . . . ,Am.
Let the inverse transformations be given by g�11k , . . . , g�1
dk , k = 1, . . . ,m such that
x1 = g�11k (u1, . . . , ud) . . . xd = g�1
1k (u1, . . . , ud)
is the unique value in Ak mapped to (u1, . . . , ud). Then
fU1,...,Ud (u1, . . . , ud) =mX
k=1
f (g�11k (u1, . . . , ud), . . . , g
�1dk (u1, . . . , ud))|Jk(u1, . . . , ud)|,
where Jk(u1, . . . , ud) is the Jacobian of the transformation on Ak , k = 1, . . . ,m.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 14 / 22
E0
Ard
g m mm
der
Multivariate transformations, including non-1:1
Exercise: Let X ⇠ Normal(0, 1). Find the pdf of U = X 2.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 15 / 22
A P
d L
U E o
u xing xd
ftp EE gitarre gilt F
TG detff.si Elms d.tl n.til
1 ErEr
Multivariate transformations, including non-1:1
Exercise: Let Z1,Z2ind⇠ Normal(0, 1) and consider the rvs
Y1 = Z 21 + Z 2
2
Y2 = Z1/q
Z 21 + Z 2
2 .
1 Find the joint pdf of (Y1,Y2).2 Check whether Y1 and Y2 are independent.3 Find the marginal pdf of Y2.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 16 / 22
fz.zdz.in E e geEk
Tnf e o a 1
µ2,355
yay 1
y Zitze g Zi Zz
JE g z z
Z Fk
zu ITy.si
Multivariate transformations, including non-1:1
Exercise: Let X1,X2,X3ind⇠ Exponential(1) and consider
Y1 =X1
X1 + X2
Y2 =X1 + X2
X1 + X2 + X3
Y3 = X1 + X2 + X3.
1 Find the joint pdf of (Y1,Y2,Y3).2 Check whether Y1, Y2, and Y3 are mutually independent.3 Find the marginal distributions Y1, Y2, and Y3.
Karl B. Gregory (U. of South Carolina) STAT 712 fa 2021 Lec 10 slides 17 / 22
Sy nix e e er I x o Xc so 7170
g 711 2,4
82 x X2 XEin D
can
ng x xD
ECO
AEIHNHIEÜH 4 4 43 4434 43
Athit 4243 4243 414243 141 Yay