WEBVTT

00:00:01.310 --> 00:00:05.530
Besides PMFs and PDFs, we can
also describe the distribution

00:00:05.530 --> 00:00:09.330
of a random variable, as
we know, using a CDF.

00:00:09.330 --> 00:00:12.260
A CDF is always well-defined.

00:00:12.260 --> 00:00:14.970
And for the case of a continuous
random variable,

00:00:14.970 --> 00:00:18.640
the CDF can be found by
integrating the PDF.

00:00:18.640 --> 00:00:23.230
And conversely, we can recover
the PDF from the CDF by

00:00:23.230 --> 00:00:25.040
differentiating.

00:00:25.040 --> 00:00:28.010
There is something similar that
happens for the case of

00:00:28.010 --> 00:00:30.410
multiple random variables,
as well.

00:00:30.410 --> 00:00:37.390
We can define the joint CDF as
the probability that X and Y,

00:00:37.390 --> 00:00:43.290
the pair X-Y, takes values
that are below certain

00:00:43.290 --> 00:00:46.360
numbers, little x
and little y.

00:00:46.360 --> 00:00:51.930
So we are talking about the
probability of the blue set in

00:00:51.930 --> 00:00:53.180
this diagram.

00:00:55.780 --> 00:01:05.140
This probability can be found
by integrating the joint PDF

00:01:05.140 --> 00:01:07.830
over the blue set.

00:01:07.830 --> 00:01:12.300
And, since we're using x and y
to be some specific numbers,

00:01:12.300 --> 00:01:16.070
let us use some different dummy
variables to carry out

00:01:16.070 --> 00:01:17.320
the integration.

00:01:22.310 --> 00:01:25.000
What is the range of
the integration?

00:01:25.000 --> 00:01:29.160
The first variable, which is s
in this integral, ranges from

00:01:29.160 --> 00:01:31.226
minus infinity up to x.

00:01:34.021 --> 00:01:37.520
And the other variable, which
is the one that we're

00:01:37.520 --> 00:01:42.150
integrating with respect to,
in the outer integral--

00:01:42.150 --> 00:01:43.380
the t variable--

00:01:43.380 --> 00:01:45.910
ranges from minus
infinity to y.

00:01:50.000 --> 00:01:53.870
Now, let us see what happens if
we start taking derivatives

00:01:53.870 --> 00:01:55.610
of this expression.

00:01:55.610 --> 00:01:59.960
If we take the derivative of
this expression with respect

00:01:59.960 --> 00:02:04.475
to y, what is left is
the inner integral.

00:02:07.010 --> 00:02:11.810
And if we take, now, a
derivative with respect to x

00:02:11.810 --> 00:02:15.550
of this inner integral,
we will be left with

00:02:15.550 --> 00:02:20.090
just the joint PDF.

00:02:20.090 --> 00:02:24.600
And it will be the joint PDF
evaluated at the particular

00:02:24.600 --> 00:02:26.329
limits of the integration.

00:02:26.329 --> 00:02:32.420
So, it's going to be f sub
xy at little x, little y.

00:02:32.420 --> 00:02:35.010
So, we have this particular
formula.

00:02:35.010 --> 00:02:38.320
By taking derivative with
respect to x, and then with

00:02:38.320 --> 00:02:41.075
respect to y, or maybe in
the opposite order.

00:02:41.075 --> 00:02:42.790
It doesn't matter.

00:02:42.790 --> 00:02:48.000
This particular derivative
gives us back the PDF.

00:02:48.000 --> 00:02:50.190
Let us look at an example.

00:02:50.190 --> 00:02:52.730
Suppose that we have a uniform

00:02:52.730 --> 00:02:56.820
distribution on the unit square.

00:02:56.820 --> 00:03:01.540
So the PDF is equal to 1
on this green square.

00:03:01.540 --> 00:03:04.610
And is equal to 0 otherwise.

00:03:04.610 --> 00:03:08.970
So, in this example, if we take
some x and y, so that the

00:03:08.970 --> 00:03:16.060
xy pair falls inside the
rectangle, the probability of

00:03:16.060 --> 00:03:20.400
the blue set is going to be just
the probability of that

00:03:20.400 --> 00:03:22.140
little rectangle here.

00:03:22.140 --> 00:03:25.860
Because everything outside
has zero probability.

00:03:25.860 --> 00:03:30.240
With a uniform joint PDF,
which is equal to 1, the

00:03:30.240 --> 00:03:33.329
probability is just the area
of the set that we are

00:03:33.329 --> 00:03:34.150
considering.

00:03:34.150 --> 00:03:37.390
And since this set that we are
considering is a rectangle

00:03:37.390 --> 00:03:38.120
with [sides]

00:03:38.120 --> 00:03:44.280
x and y, the joint CDF is
equal to x times y.

00:03:44.280 --> 00:03:47.340
Now, if we take the derivative
of this expression with

00:03:47.340 --> 00:03:53.180
respect to x, and then with
respect to y, then we're left

00:03:53.180 --> 00:03:56.095
just with a constant
equal to 1--

00:03:59.620 --> 00:04:04.590
which is as it should be, so
that it integrates to one.

00:04:04.590 --> 00:04:08.870
So, we have seen that CDFs
also apply to the case of

00:04:08.870 --> 00:04:12.610
multiple random variables, and
that we can recover the joint

00:04:12.610 --> 00:04:14.350
PDF from the joint CDF.