WEBVTT

00:00:00.190 --> 00:00:03.380
In this problem, we're given a
random variable X which has a

00:00:03.380 --> 00:00:09.980
uniform distribution in the
interval negative 1 to 1.

00:00:09.980 --> 00:00:15.030
In other words, if we were to
draw out the PDF of X, we see

00:00:15.030 --> 00:00:19.290
that in the interval negative
1 to 1, it has value 1/2.

00:00:22.460 --> 00:00:26.750
Now we're given a sequence
random variables X1, X2, and

00:00:26.750 --> 00:00:32.380
so on, where each Xi has the
same distribution as X and

00:00:32.380 --> 00:00:35.480
different Xi's are
independent.

00:00:35.480 --> 00:00:38.820
For part a, we would like to
know if the sequence Xi

00:00:38.820 --> 00:00:41.070
converges to some number--

00:00:41.070 --> 00:00:42.390
let's call it c--

00:00:42.390 --> 00:00:48.100
in probability as i
goes to infinity--

00:00:48.100 --> 00:00:49.710
whether this is true.

00:00:49.710 --> 00:00:52.030
Let's first recall the
definition of convergence in

00:00:52.030 --> 00:00:53.410
probability.

00:00:53.410 --> 00:00:57.900
If this does happen, then by
definition, we'll have that

00:00:57.900 --> 00:01:05.660
for every epsilon greater than
0, the probability Xi minus c

00:01:05.660 --> 00:01:11.510
greater equal to epsilon, this
quantity will go to 0 in the

00:01:11.510 --> 00:01:15.490
limit of i going to infinity.

00:01:15.490 --> 00:01:20.450
In other words, with very high
probability, we will find Xi

00:01:20.450 --> 00:01:24.220
to be very concentrated around
the number c if this were to

00:01:24.220 --> 00:01:28.580
be the PDF of Xi.

00:01:28.580 --> 00:01:30.430
Now, can this be true?

00:01:30.430 --> 00:01:34.740
Well, we know that each Xi is
simply a uniform distribution

00:01:34.740 --> 00:01:36.380
over negative 1 to 1.

00:01:36.380 --> 00:01:38.790
It doesn't really change
as we increase i.

00:01:38.790 --> 00:01:41.420
So intuitively, the
concentration around any

00:01:41.420 --> 00:01:43.840
number c is not going
to happen.

00:01:43.840 --> 00:01:47.980
So we should not expect a
convergence in probability in

00:01:47.980 --> 00:01:49.050
this sense.

00:01:49.050 --> 00:01:53.250
For part b, we would like to
know whether the sequence Yi,

00:01:53.250 --> 00:01:58.370
defined as Xi divided by i,
converges to anything in

00:01:58.370 --> 00:02:01.050
probability.

00:02:01.050 --> 00:02:04.250
Well, by just looking at the
shape of Yi, we know that

00:02:04.250 --> 00:02:09.340
since the absolute value of Xi
is less than 1, then we expect

00:02:09.340 --> 00:02:14.380
the absolute value of
Yi is less than 1/i.

00:02:14.380 --> 00:02:17.400
So eventually, Yi gets
very close to 0

00:02:17.400 --> 00:02:19.450
as i goes to infinity.

00:02:19.450 --> 00:02:23.650
So it's safe to bet that maybe
Yi will converge to 0 in

00:02:23.650 --> 00:02:26.400
probability.

00:02:26.400 --> 00:02:28.870
Let's see if this is
indeed the case.

00:02:28.870 --> 00:02:35.460
The probability of Yi minus 0
greater equal to epsilon is

00:02:35.460 --> 00:02:39.940
equal to the probability of Yi
absolute value greater equal

00:02:39.940 --> 00:02:41.440
to epsilon.

00:02:41.440 --> 00:02:46.020
Now, previously we know that the
absolute value of Yi is at

00:02:46.020 --> 00:02:49.270
most 1/i by the definition
of Yi.

00:02:49.270 --> 00:02:53.000
And hence the probability right
here is upper bounded by

00:02:53.000 --> 00:02:57.370
the probability of 1i greater
equal to epsilon.

00:02:57.370 --> 00:02:59.740
Notice in this expression,
there is nothing random.

00:02:59.740 --> 00:03:01.410
i is simply a number.

00:03:01.410 --> 00:03:11.245
Hence this is either 1 if i is
less equal to 1/epsilon, or 0

00:03:11.245 --> 00:03:15.700
if i is greater than
1/epsilon.

00:03:15.700 --> 00:03:20.520
Now, this tells us, as long as
i is great enough-- it's big

00:03:20.520 --> 00:03:22.720
enough compared to epsilon--

00:03:22.720 --> 00:03:24.720
we know that this quantity
here is [INAUDIBLE]

00:03:24.720 --> 00:03:25.500
0.

00:03:25.500 --> 00:03:30.570
And that tells us in the limit
of i goes to infinity

00:03:30.570 --> 00:03:35.910
probability of Yi deviating
from 0 by more than

00:03:35.910 --> 00:03:40.070
epsilon goes to 0.

00:03:40.070 --> 00:03:47.080
And that shows that indeed, Yi
converges to 0 in probability

00:03:47.080 --> 00:03:51.100
because the expression right
here, this limit, holds for

00:03:51.100 --> 00:03:53.180
all epsilon.

00:03:53.180 --> 00:03:55.400
Now, in the last part of the
problem, we are looking at a

00:03:55.400 --> 00:04:01.100
sequence Zi defined by Xi raised
to the i-th power.

00:04:01.100 --> 00:04:03.940
Again, since we know Xi is some
number between negative 1

00:04:03.940 --> 00:04:07.380
and 1, this number raised to the
i-th power is likely to be

00:04:07.380 --> 00:04:08.630
very small.

00:04:08.630 --> 00:04:10.760
And likely to be small in the
sense that it will have

00:04:10.760 --> 00:04:12.830
absolute value close to 0.

00:04:12.830 --> 00:04:16.660
So a safe guess will be the
sequence Zi converges to 0 as

00:04:16.660 --> 00:04:20.500
well as i goes to infinity.

00:04:20.500 --> 00:04:22.490
How do we prove this formally?

00:04:22.490 --> 00:04:26.810
We'll start again with a
probability that Zi stays away

00:04:26.810 --> 00:04:31.970
from 0 by more than epsilon
and see how that evolves.

00:04:31.970 --> 00:04:36.770
And this is equal to the
probability that Xi raised to

00:04:36.770 --> 00:04:40.920
the i-th power greater
equal to epsilon.

00:04:40.920 --> 00:04:46.330
Or again, we can write this by
taking out the absolute value

00:04:46.330 --> 00:04:51.410
that Xi is less equal to
negative epsilon raised to the

00:04:51.410 --> 00:04:57.970
1 over i-th power or Xi greater
equal to epsilon 1

00:04:57.970 --> 00:05:01.470
over i-th power.

00:05:01.470 --> 00:05:05.610
So here, we'll divide into two
cases, depending on the value

00:05:05.610 --> 00:05:07.000
of epsilon.

00:05:07.000 --> 00:05:10.670
In the first case, epsilon
is greater than 1.

00:05:10.670 --> 00:05:15.130
Well, if that's the case, then
we know epsilon raised to some

00:05:15.130 --> 00:05:18.590
positive power is still
greater than 1.

00:05:18.590 --> 00:05:22.760
But again, Xi cannot have any
positive density be on the

00:05:22.760 --> 00:05:25.470
interval negative 1 or 1.

00:05:25.470 --> 00:05:28.650
And hence we know the
probability above, which is Xi

00:05:28.650 --> 00:05:31.780
less than some number smaller
than negative 1 or greater

00:05:31.780 --> 00:05:34.720
than some number bigger
than 1 is 0.

00:05:34.720 --> 00:05:36.910
So that case is handled.

00:05:36.910 --> 00:05:40.210
Now let's look at a case where
epsilon is less than 1,

00:05:40.210 --> 00:05:41.870
greater than 0.

00:05:41.870 --> 00:05:47.010
So in this case, epsilon to the
1/i will be less than 1.

00:05:47.010 --> 00:05:50.380
And it's not that difficult
to check that since Xi has

00:05:50.380 --> 00:05:55.280
uniform density between negative
1 and 1 of magnitude

00:05:55.280 --> 00:06:02.290
1/2, then the probability here
was simply 2 times 1/2 times

00:06:02.290 --> 00:06:07.590
the distance between epsilon
to the 1 over

00:06:07.590 --> 00:06:11.390
i-th power and 1.

00:06:11.390 --> 00:06:15.080
So in order to prove this
quantity converge to 0, we

00:06:15.080 --> 00:06:19.200
simply have to justify why
does epsilon to the 1/i

00:06:19.200 --> 00:06:23.570
converge to 1 as i
goes to infinity.

00:06:23.570 --> 00:06:25.120
For that, we'll recall
the properties

00:06:25.120 --> 00:06:26.560
of exponential functions.

00:06:26.560 --> 00:06:30.910
In particular, if a is a
positive number and x is its

00:06:30.910 --> 00:06:35.920
exponent, if we were to take the
limit as x goes to 0 and

00:06:35.920 --> 00:06:39.370
look at the value of a to the
power of x, we see that

00:06:39.370 --> 00:06:41.190
this goes to 1.

00:06:41.190 --> 00:06:46.920
So in this case, we'll let a be
equal to epsilon and x be

00:06:46.920 --> 00:06:48.640
equal to 1/i.

00:06:48.640 --> 00:06:52.450
As we can see that as i goes to
infinity, the value of x,

00:06:52.450 --> 00:06:55.070
which is 1/i, does go to 0.

00:06:55.070 --> 00:06:59.540
And therefore, in the limit i
going to infinity, the value

00:06:59.540 --> 00:07:04.500
of epsilon to the 1 over
i-th power goes to 1.

00:07:04.500 --> 00:07:09.380
And that shows if we plug this
limit into the expression

00:07:09.380 --> 00:07:13.860
right here that indeed, the term
right here goes to 0 as i

00:07:13.860 --> 00:07:15.150
goes to infinity.

00:07:15.150 --> 00:07:20.740
And all in all, this implies the
probability of Zi minus 0

00:07:20.740 --> 00:07:25.040
absolute value greater equal to
epsilon in the limit of i

00:07:25.040 --> 00:07:30.370
going to infinity converges to
0 for all positive epsilon.

00:07:30.370 --> 00:07:34.230
And that completes our proof
that indeed, Zi converges to 0

00:07:34.230 --> 00:07:35.480
in probability.