1
00:00:01,670 --> 00:00:03,465
In this brief segment,
we will discuss

2
00:00:03,465 --> 00:00:07,970
an important application
of the convolution formula.

3
00:00:07,970 --> 00:00:11,360
Suppose that X is a normal
random variable with a given

4
00:00:11,360 --> 00:00:12,300
mean and variance.

5
00:00:12,300 --> 00:00:14,950
So that the PDF of
X takes this form.

6
00:00:14,950 --> 00:00:18,140
And similarly, Y is normal
with a given mean and variance.

7
00:00:18,140 --> 00:00:20,020
So its PDF takes this form.

8
00:00:20,020 --> 00:00:22,800
We assume that X and
Y are independent.

9
00:00:22,800 --> 00:00:26,740
And we're interested in the
sum of the two random variables

10
00:00:26,740 --> 00:00:30,410
X and Y. And we wish
to derive the PDF of Z.

11
00:00:30,410 --> 00:00:35,040
Of course, the PDF of Z is given
by the convolution formula.

12
00:00:35,040 --> 00:00:40,040
And now we plug in here, the
form for the density of X.

13
00:00:40,040 --> 00:00:43,250
And here, we plug the
form of the density of Y.

14
00:00:43,250 --> 00:00:45,950
Except that instead
of the argument Y,

15
00:00:45,950 --> 00:00:49,240
we need to put in the
argument z minus x.

16
00:00:49,240 --> 00:00:51,320
So we obtain this
form, where here we

17
00:00:51,320 --> 00:00:54,690
have a z minus x instead of y.

18
00:00:54,690 --> 00:00:58,460
Now this is an integral that
looks pretty complicated.

19
00:00:58,460 --> 00:01:01,260
But it is not too hard to do.

20
00:01:01,260 --> 00:01:05,470
One just needs to be patient,
rearrange terms, collect terms.

21
00:01:05,470 --> 00:01:07,620
And the details of
the calculations

22
00:01:07,620 --> 00:01:09,720
are not as interesting.

23
00:01:09,720 --> 00:01:12,240
So we will skip them for now.

24
00:01:12,240 --> 00:01:14,590
And I will just tell you
that the final answer

25
00:01:14,590 --> 00:01:16,310
takes this form.

26
00:01:16,310 --> 00:01:17,450
What is this form?

27
00:01:17,450 --> 00:01:20,130
Well, it's
exponential of minus z

28
00:01:20,130 --> 00:01:23,870
minus something squared
divided by a constant.

29
00:01:23,870 --> 00:01:25,610
And we recognize
that this is the form

30
00:01:25,610 --> 00:01:28,240
of a normal random variable.

31
00:01:28,240 --> 00:01:31,510
It's a normal random
variable whose mean

32
00:01:31,510 --> 00:01:37,970
is given by this term
here, it's mu x plus mu y.

33
00:01:37,970 --> 00:01:40,750
And the variance of that
normal random variable

34
00:01:40,750 --> 00:01:43,520
is that constant that appears
next to the factor of 2

35
00:01:43,520 --> 00:01:45,130
in the denominator.

36
00:01:45,130 --> 00:01:48,320
So the sum of these two
normal random variables,

37
00:01:48,320 --> 00:01:50,850
these two independent
normal random variables,

38
00:01:50,850 --> 00:01:52,166
is also normal.

39
00:01:52,166 --> 00:01:53,789
The fact that this
is the mean and this

40
00:01:53,789 --> 00:01:57,190
is the variance of the sum,
of course, is not a surprise.

41
00:01:57,190 --> 00:02:00,490
What is important in this
result that we have here

42
00:02:00,490 --> 00:02:03,800
is that the sum is
actually normal.

43
00:02:03,800 --> 00:02:05,600
Now, we carried
out this argument

44
00:02:05,600 --> 00:02:09,590
for the case of the sum of
two normal random variables.

45
00:02:09,590 --> 00:02:13,630
But suppose that we
had the sum of three

46
00:02:13,630 --> 00:02:15,860
independent normal
random variables,

47
00:02:15,860 --> 00:02:17,870
what can we say about it?

48
00:02:17,870 --> 00:02:22,640
By the result that we just
discussed, this sum is normal.

49
00:02:22,640 --> 00:02:24,340
This is assumed to be normal.

50
00:02:24,340 --> 00:02:27,490
We assume that X, Y,
and W are independent.

51
00:02:27,490 --> 00:02:31,560
Therefore, this sum
is independent from W.

52
00:02:31,560 --> 00:02:33,980
So we're dealing
with the sum of two

53
00:02:33,980 --> 00:02:37,140
independent normal
random variables again.

54
00:02:37,140 --> 00:02:40,430
So this sum here is going
to be normal as well.

55
00:02:40,430 --> 00:02:43,070
And we continue this
argument by induction,

56
00:02:43,070 --> 00:02:45,110
and conclude that
more generally,

57
00:02:45,110 --> 00:02:48,329
the sum of any finite number
of independent normal random

58
00:02:48,329 --> 00:02:50,160
variables is normal.

59
00:02:50,160 --> 00:02:52,850
This is a very important,
but also useful fact.

60
00:02:52,850 --> 00:02:54,780
It means that when
we start working

61
00:02:54,780 --> 00:02:57,050
with normal random
variables, very often

62
00:02:57,050 --> 00:03:00,150
we stay within the realm
of normal random variables.

63
00:03:00,150 --> 00:03:02,890
We can form linear
functions of them,

64
00:03:02,890 --> 00:03:04,830
take linear
combinations of them,

65
00:03:04,830 --> 00:03:10,480
and still remain in the world
of normal random variables.