Hc

f a continuous Rv
revisiting basic definitions expectation variance

joint distributions conditionals independence

continuous Bayes rule

Many of the concepts we covered for discrete rv s follow exactly

for continuous v.v s First recall the definitions of the distribution
functions except in the continuous case we will be calling them
densities instead of masses

pdf of a continuous v.v is

It.at abjpggspyfunctin
IP FEB f f x dx BCIR

Its co.d is defined

tivednsityq.gg EgR
Note that we do not define the pdf as f x P X x because the
event
corresponding to one
point XER has probability 0
Recallthisgraph
for ftp p
nfmass
fromprevious lecture

Def A random variable X is continuous if its cdf can be written as

 over the pdf
an
integral
f
1
F x f u du

Note this expression is analogous to the discrete case x

Ex Px Z

Note that f x does not necessarily have to lie within 0,1

However its area under the curve must lie within 0,1 because
of definition 1

T
f .

x x

Examples Continuous uniform distribution on 0

note it is 1

1
co.es
24 at

Tx
The pdf cdf must satisfy the following properties
1 x 0 XER
i e it is continuous everywhere
EE discontinuities and doesn't blow
up
to as anywhere
3
f f x dx 2
4
F b Fx a
f x dx P a x̅ b
 ootie

event a x̅ b

One can relate the terminologies of mass discrete case and density
continuous case through an infinitesimal interpretation
P x e f z dz f x E

for infinitesimally small E 0

mass usually equal to the Density

is

of an object times its Area Volume

Thus
1f
the continuous
e.g think about the
tank of water
it is more useful to
mass of a cubical

in case
quantify probability
as a density than mass

The definitions of the expectation and variance are also similar in the

continuous case as in the discrete case

Def The expectation or

expectedvalue or mean of v.v x̅ with pdf

Sx is defined as

aant.IE I Var X IE X M where E x̅

u
All the properties follow similarly e g linearity expected value of g x

Lemmy If x̅ has density with f x 0 so and cdf Fx

 IE x̅ 1 Fx x ax

proof is similar to discrete case

Joint distributions and independence all follow the same

properties in

the continuous case as the discrete case just with different

notation
Px x
y x y
However the biggest distinction is that P X x has zero probability
in the continuous case since XE IR is
point with zero mass just a

Thus we often deal with the cdf instead of the pdf

Def The jointdensityfunctions of X Y is Fye 1122 0,1 given by

x
yeR Fx x
y P x
Tey fxy.lu v audu

Def The marginaldensityfunctions corresponding to joint Fye are

F IP F y p Tey
fig fig
x x
x x y y y y

f x fx x y dy f y fxy x y dx

Examples 2D Uniform distribution tt

C if 0 x 2 and o y s
ftp.y lx.y
use

y
Here constant CER must satisfy normalization

Spa Fx x
y dx dy 1 So'f c dx dy 1
421
Lemma Random variables X Y are independent if
Exy x
y Fx x
Fy y fx x
y f x
fy.ly
 EE it a be R
O if a

if acx b
E x

2 if b

EX f taxax.at
Fi fx
Vars
b1
Exponential continuous limit of a waiting time distribution
e
g time between consecutive bus arrivals earthquakes other
uncertain events
e if so
Exponential X x̅ Exp x f x e
else
Here SO is called the
rateparameters

E x
je
E X

varix t

We've actually seen a type of waiting time distribution before in the

discrete case the Geometric v.v In fact the Geometric and
Exponential r.ve's can be related to each other as follows

E Ep
1
Faean n E P k t
p 1 1 p Knew
 Fexp x 1 e xx 4 70

s
Define S In l p so that e 1 p Then Exp NS Foam n

This can be interpreted using our favorite coin tossing scenario Suppose
wetoss the coin very quickly every 84 1 seconds a coin with prob of
H p l e t Then the time until the first H appears is very close
to being Exponentially distributed with parameter X
This relationship will play an
important role later when we discuss
Poisson processes

One of the most continuous v.v and

usst important
IT h NCM 02 where m is the mean x̅ M
O is the variance Var X 02

m
x e e exer

Clearly IE x̅ M Var X 02 One can

verify these using the

integral definitions of the mean variance

atx

x x

The standard normal arises when M O 52 1

2
x E e

We will talk much more about normal r.vn's later

 Gammas a generalization of the Exponential distribution
There are two common parametrizations used in practice
x̅ Gamma α 0 or
Gammg.la
shapt scale ratex t
Here α O X 0

In the first parametrization

f x e F x 8 x 8
IE x̅ 20 Var X 202
In the second parametrization

f x x é F x 8 x xx

IE x̅ Var X 4 2

where P is the gamma function

Mα fit etat whenever 230

x 1 if αE IN

and 8 is the incomplete gamma function which we won't write

as

Gamma 1 X is the same as anExp X distribution

If x̅ In are i i d with each x̅ Exp x then

E Gamma n nx
 ions and conditional expectations the fact that
P X x 0 requires us to consider a more careful definition since

Prix y x fyyx.ly x IP Y y X x

becomes undefined
Again let's use the cdf let xeR be s t f x 0 Consider the

infinitesimal length EER and the probability

P Y y x x̅extE
PCYIY.EE E
S t
II
S t

x is the function

EII.gg ytionoftgivenx
f
Fyix.ly x IP Fy1X x
Ef g du

for EIR sit x 0

any
Correspondingly it becomes convenient to define the conditionaldensity
of Fy as
fy Ix given by the integrand of 2

fall
From this definition we have the continuousBaysrule
 ii
I IiIII
E iit

of a specific event A given Y y P AIY y we

get
PAIGE y P A y Y y S

from the usual Bayes rule

PE jfd f II
The denominator can be evaluated via the total probability theorem

f y IPCA fya y PAY fyya.ly

so that
IP A fyla y
iiii 1tt t

Weargivenalightbulbwithexpantilly distributed lifetime Y

but the
parameter 1 is alse a v.v X Unif 1

safe
face

fy Flax Exp X
x
If we empirically the lifetime of the bulb to be ye Rt
measure we

can compute an
expression for its conditional pdf
 masts.FI ti T
tnEII
transmitted we
it so
1 w.p 1 p
The received
signal is Y S w̅ where w̅ NCO 1 is independent
of S This means Y S s N s 1

What is IP 5 11 Y y the probability that I was transmitted

given we received yer
PS.CA
Using Bayes PCS 11 Fy
ffYt
e
E iif

en

Det Thecaonalexpectation of enemy

IE YT X x
y fix ylx dy