1
00:00:00,750 --> 00:00:03,290
In this segment, we start
a discussion of multiple

2
00:00:03,290 --> 00:00:05,710
continuous random variables.

3
00:00:05,710 --> 00:00:09,760
Here are some objects that we're
already familiar with.

4
00:00:09,760 --> 00:00:14,090
But exactly as in the discrete
case, if we are dealing with

5
00:00:14,090 --> 00:00:17,930
two random variables, it is
not enough to know their

6
00:00:17,930 --> 00:00:20,250
individual PDFs.

7
00:00:20,250 --> 00:00:23,230
We also need to model the
relation between the two

8
00:00:23,230 --> 00:00:27,780
random variables, and this is
done through a joint PDF,

9
00:00:27,780 --> 00:00:32,800
which is the continuous analog
of the joint PMF.

10
00:00:32,800 --> 00:00:37,180
We will use this notation to
indicate joint PDFs where we

11
00:00:37,180 --> 00:00:41,650
use f to indicate that we're
dealing with a density.

12
00:00:41,650 --> 00:00:45,520
So what remains to be done is to
actually define this object

13
00:00:45,520 --> 00:00:47,800
and see how we use it.

14
00:00:47,800 --> 00:00:52,440
Let us start by recalling that
joint PMFs were defined in

15
00:00:52,440 --> 00:00:55,910
terms of the probability that
the pair of random variables X

16
00:00:55,910 --> 00:01:01,350
and Y take certain specific
values little x and little y.

17
00:01:01,350 --> 00:01:05,970
Regarding joint PDFs, we start
by saying that it has to be

18
00:01:05,970 --> 00:01:08,050
non-negative.

19
00:01:08,050 --> 00:01:10,880
However, a more precise
interpretation in terms of

20
00:01:10,880 --> 00:01:14,490
probabilities has to
wait a little bit.

21
00:01:14,490 --> 00:01:19,270
Joint PDFs will be used to
calculate probabilities.

22
00:01:19,270 --> 00:01:21,300
And this will be done
in analogy with

23
00:01:21,300 --> 00:01:22,740
the discrete setting.

24
00:01:22,740 --> 00:01:25,820
In the discrete setting, the
probability that the pair of

25
00:01:25,820 --> 00:01:30,660
random variables falls inside a
certain set is just the sum

26
00:01:30,660 --> 00:01:35,390
of the probabilities of all of
the possible pairs inside that

27
00:01:35,390 --> 00:01:37,310
particular set.

28
00:01:37,310 --> 00:01:39,880
For the continuous case,
we introduce

29
00:01:39,880 --> 00:01:41,990
an analogous formula.

30
00:01:41,990 --> 00:01:46,120
We use the joint density instead
of the joint PMF.

31
00:01:46,120 --> 00:01:50,700
And instead of having a
summation, we now integrate.

32
00:01:50,700 --> 00:01:55,450
As in the discrete setting,
we have one total unit of

33
00:01:55,450 --> 00:01:56,700
probability.

34
00:01:58,630 --> 00:02:04,090
The joint PDF tells us how this
unit of probability is

35
00:02:04,090 --> 00:02:06,460
spread over the entire
continuous

36
00:02:06,460 --> 00:02:08,360
two-dimensional plane.

37
00:02:08,360 --> 00:02:13,080
And we use it, we use the joint
PDF, to calculate the

38
00:02:13,080 --> 00:02:19,180
probability of a certain set
by finding the volume under

39
00:02:19,180 --> 00:02:24,270
the joint PDF that lies
on top of that set.

40
00:02:24,270 --> 00:02:27,490
This is what this integral
really represents.

41
00:02:27,490 --> 00:02:32,170
We integrate over a particular
two-dimensional set, and we

42
00:02:32,170 --> 00:02:36,510
take this value that
we integrate.

43
00:02:36,510 --> 00:02:40,360
And we can think of this as the
height of an object that's

44
00:02:40,360 --> 00:02:43,400
sitting on top of that set.

45
00:02:43,400 --> 00:02:46,990
Now, this relation here, this
calculation of probabilities,

46
00:02:46,990 --> 00:02:51,140
is not something that we
are supposed to prove.

47
00:02:51,140 --> 00:02:54,160
This is, rather, the
definition of

48
00:02:54,160 --> 00:02:57,450
what a joint PDF does.

49
00:02:57,450 --> 00:03:01,920
A legitimate joint PDF is any
function of two variables,

50
00:03:01,920 --> 00:03:06,820
which is non-negative and
which integrates to 1.

51
00:03:06,820 --> 00:03:11,670
And we will say that two random
variables are jointly

52
00:03:11,670 --> 00:03:18,420
continuous if there is a
legitimate joint PDF that can

53
00:03:18,420 --> 00:03:21,890
be used to calculate the
associated probabilities

54
00:03:21,890 --> 00:03:24,710
through this particular
formula.

55
00:03:24,710 --> 00:03:27,770
So we have really an indirect
definition.

56
00:03:27,770 --> 00:03:32,290
Instead of defining the joint
PDF as a probability, we

57
00:03:32,290 --> 00:03:37,610
actually define it indirectly by
saying what it does, how it

58
00:03:37,610 --> 00:03:42,220
will be used to calculate
probabilities.

59
00:03:42,220 --> 00:03:44,780
A picture will be
helpful here.

60
00:03:44,780 --> 00:03:48,200
Here's a plot of a possible
joint PDF.

61
00:03:48,200 --> 00:03:53,000
These are the x and y-axes.

62
00:03:53,000 --> 00:03:58,990
And the function being plotted
is the joint PDF of these two

63
00:03:58,990 --> 00:04:01,580
random variables.

64
00:04:01,580 --> 00:04:05,730
This joint PDF is higher at
some places and lower at

65
00:04:05,730 --> 00:04:08,800
others, indicating that certain
regions of the x,y

66
00:04:08,800 --> 00:04:11,480
plane are more likely
than others.

67
00:04:11,480 --> 00:04:15,630
The joint PDF determines the
probability of a set B by

68
00:04:15,630 --> 00:04:21,470
integrating over that set B.
Let's say it's this set.

69
00:04:21,470 --> 00:04:24,940
Integrating the PDF
over that set.

70
00:04:24,940 --> 00:04:30,190
Pictorially, what this means is
that we look at the volume

71
00:04:30,190 --> 00:04:36,790
that sits on top of that set,
but below the PDF, below the

72
00:04:36,790 --> 00:04:41,780
joint PDF, and so we obtain some
three-dimensional object

73
00:04:41,780 --> 00:04:43,100
of this kind.

74
00:04:43,100 --> 00:04:47,940
And this integral corresponds
to actually finding this

75
00:04:47,940 --> 00:04:52,740
volume here, the volume that
sits on top of the set B but

76
00:04:52,740 --> 00:04:55,060
which is below the joint PDF.

77
00:04:58,080 --> 00:05:01,410
Let us now develop some
additional understanding of

78
00:05:01,410 --> 00:05:03,680
joint PDFs.

79
00:05:03,680 --> 00:05:12,430
As we just discussed, for any
given set B, we can integrate

80
00:05:12,430 --> 00:05:15,890
the joint PDF over that set.

81
00:05:15,890 --> 00:05:18,770
And this will give us
the probability of

82
00:05:18,770 --> 00:05:21,120
that particular set.

83
00:05:21,120 --> 00:05:24,460
Of particular interest is the
case where we're dealing with

84
00:05:24,460 --> 00:05:29,420
a set which is a rectangle, in
which case the situation is a

85
00:05:29,420 --> 00:05:30,750
little simpler.

86
00:05:30,750 --> 00:05:33,380
So suppose that we have
a rectangle where the

87
00:05:33,380 --> 00:05:38,460
x-coordinate ranges from A to B
and the y-coordinate ranges

88
00:05:38,460 --> 00:05:44,420
from some C to some D. Then, the
double integral over this

89
00:05:44,420 --> 00:05:48,520
particular rectangle can be
written in a form where we

90
00:05:48,520 --> 00:05:52,620
first integrate with respect to
one of the variables that

91
00:05:52,620 --> 00:05:57,950
ranges from A to B. And then, we
integrate over all possible

92
00:05:57,950 --> 00:06:01,480
values of y as they
range from C to D.

93
00:06:01,480 --> 00:06:05,270
Of particular interest is the
special case where we're

94
00:06:05,270 --> 00:06:10,290
dealing with a small rectangle
such as this one.

95
00:06:10,290 --> 00:06:15,990
A rectangle with sizes equal to
some delta where delta is a

96
00:06:15,990 --> 00:06:19,000
small number.

97
00:06:19,000 --> 00:06:22,920
In that case, the double
integral, which is the volume

98
00:06:22,920 --> 00:06:28,810
on top of that rectangle,
is simpler to evaluate.

99
00:06:28,810 --> 00:06:33,010
It is equal to the value of
the function that we're

100
00:06:33,010 --> 00:06:36,640
integrating at some point in the
rectangle --- let's take

101
00:06:36,640 --> 00:06:37,940
that corner ---

102
00:06:37,940 --> 00:06:42,640
times the area of that little
rectangle, which is equal to

103
00:06:42,640 --> 00:06:44,390
delta square.

104
00:06:44,390 --> 00:06:47,820
So we have an interpretation of
the joint PDF in terms of

105
00:06:47,820 --> 00:06:50,820
probabilities of small
rectangles.

106
00:06:50,820 --> 00:06:54,140
Joint PDFs are not
probabilities.

107
00:06:54,140 --> 00:06:57,560
But rather, they are probability
densities.

108
00:06:57,560 --> 00:07:03,020
They tell us the probability
per unit area.

109
00:07:03,020 --> 00:07:05,190
And one more important
comment.

110
00:07:05,190 --> 00:07:08,450
For the case of a single
continuous random variable, we

111
00:07:08,450 --> 00:07:12,750
know that any single point
has 0 probability.

112
00:07:12,750 --> 00:07:16,620
This is again, true for the case
of two jointly continuous

113
00:07:16,620 --> 00:07:17,440
random variables.

114
00:07:17,440 --> 00:07:19,460
But more is true.

115
00:07:19,460 --> 00:07:24,350
If you take a set B
that has 0 area.

116
00:07:24,350 --> 00:07:26,890
For example, a certain curve.

117
00:07:26,890 --> 00:07:34,890
Suppose that this curve is the
entire set B. Then, the volume

118
00:07:34,890 --> 00:07:40,320
under the joint PDF that's
sitting on top of that curve

119
00:07:40,320 --> 00:07:43,570
is going to be equal to 0.

120
00:07:43,570 --> 00:07:48,460
So 0 area sets have
0 probability.

121
00:07:48,460 --> 00:07:50,930
And this is one of the
characteristic features of

122
00:07:50,930 --> 00:07:54,590
jointly continuous
random variables.

123
00:07:54,590 --> 00:07:57,920
Now, let's think of a particular
situation.

124
00:07:57,920 --> 00:08:04,160
Suppose that X is a continuous
random variable, and let Y be

125
00:08:04,160 --> 00:08:08,160
another random variable, which
is identically equal to X.

126
00:08:08,160 --> 00:08:11,730
Since X is a continuous random
variable, Y is also a

127
00:08:11,730 --> 00:08:13,690
continuous random variable.

128
00:08:13,690 --> 00:08:17,480
However, in this situation, we
are certain that the outcome

129
00:08:17,480 --> 00:08:20,890
of the experiment is going
to fall on the line

130
00:08:20,890 --> 00:08:23,080
where x equals y.

131
00:08:23,080 --> 00:08:26,910
All the probability lies
on top of a line, and

132
00:08:26,910 --> 00:08:30,140
a line has 0 area.

133
00:08:30,140 --> 00:08:33,840
So we have positive probability
on the set of 0

134
00:08:33,840 --> 00:08:37,159
area, which contradicts what
we discussed before.

135
00:08:37,159 --> 00:08:41,100
Well, this simply means that
X and Y are not jointly

136
00:08:41,100 --> 00:08:42,230
continuous.

137
00:08:42,230 --> 00:08:44,420
Each one of them is continuous,
but together

138
00:08:44,420 --> 00:08:47,840
they're not jointly
continuous.

139
00:08:47,840 --> 00:08:52,230
Essentially, joint continuity
is something more than

140
00:08:52,230 --> 00:08:56,370
requiring each random variable
to be continuous by itself.

141
00:08:56,370 --> 00:09:01,010
For joint continuity, we want
the probability to be really

142
00:09:01,010 --> 00:09:03,820
spread over two dimensions.

143
00:09:03,820 --> 00:09:07,630
Probability is not allowed
to be concentrated on a

144
00:09:07,630 --> 00:09:09,140
one-dimensional set.

145
00:09:09,140 --> 00:09:11,850
On the other hand, in this
example, the probability is

146
00:09:11,850 --> 00:09:14,450
concentrated on a
one-dimensional set.

147
00:09:14,450 --> 00:09:16,220
And we do not have
joint continuity.