WEBVTT
00:00:08.466 --> 00:00:10.830
BEN HARRIS: Hi,
and welcome back.
00:00:10.830 --> 00:00:12.910
Today we're going to do
a problem about the four
00:00:12.910 --> 00:00:15.160
fundamental subspaces.
00:00:15.160 --> 00:00:18.650
So here we have a
matrix B. B is written
00:00:18.650 --> 00:00:21.400
as the product of a
lower triangular matrix
00:00:21.400 --> 00:00:24.470
and an upper triangular matrix.
00:00:24.470 --> 00:00:28.100
And we're going to find
a basis for, and compute
00:00:28.100 --> 00:00:30.470
the dimension of,
each of the four
00:00:30.470 --> 00:00:33.280
fundamental subspaces of B.
00:00:33.280 --> 00:00:37.400
I'll give you a minute to try
that on your own, to hit pause,
00:00:37.400 --> 00:00:39.762
and then I'll be right
back in just a minute
00:00:39.762 --> 00:00:40.845
and we can do it together.
00:00:52.000 --> 00:00:52.840
OK.
00:00:52.840 --> 00:00:54.580
We're back.
00:00:54.580 --> 00:00:58.890
Now, the first
thing to notice is
00:00:58.890 --> 00:01:04.910
that this is an LU
decomposition of B.
00:01:04.910 --> 00:01:08.120
So we have L here and U here.
00:01:08.120 --> 00:01:11.350
Now let's go one
space at a time.
00:01:11.350 --> 00:01:12.870
Let's start with
the column space.
00:01:16.190 --> 00:01:19.720
And first, let's just say what
the dimension of the column
00:01:19.720 --> 00:01:21.490
space is.
00:01:21.490 --> 00:01:24.220
OK, so let's look
at our U matrix.
00:01:24.220 --> 00:01:26.290
How many pivots do we have?
00:01:26.290 --> 00:01:31.520
We have two pivots, so the
column space has dimension 2.
00:01:31.520 --> 00:01:35.250
This is the number of pivots.
00:01:35.250 --> 00:01:36.100
Good.
00:01:36.100 --> 00:01:40.655
Now, how do we find a
basis for the column space?
00:01:50.610 --> 00:01:52.400
How do we find that basis?
00:01:52.400 --> 00:01:55.090
Well, in lecture,
Professor Strang
00:01:55.090 --> 00:01:59.980
had started with a matrix
B. He did elimination on it,
00:01:59.980 --> 00:02:03.460
and then he took the pivot
columns in the original matrix.
00:02:03.460 --> 00:02:04.260
And that's great.
00:02:04.260 --> 00:02:05.510
That works.
00:02:05.510 --> 00:02:09.210
You can also take the pivot
columns in the L matrix.
00:02:09.210 --> 00:02:12.230
You can see by multiplying
this out that it will amount
00:02:12.230 --> 00:02:15.550
to essentially the same thing.
00:02:15.550 --> 00:02:18.520
So a basis for
this column space,
00:02:18.520 --> 00:02:23.890
I can just take these two
pivot columns of my L matrix,
00:02:23.890 --> 00:02:29.860
[1, 2, -1] and [0, 1, 0].
00:02:29.860 --> 00:02:31.580
Good.
00:02:31.580 --> 00:02:32.640
OK.
00:02:32.640 --> 00:02:36.060
So there's the dimension
of and the basis
00:02:36.060 --> 00:02:41.070
for the column space of B.
00:02:41.070 --> 00:02:43.700
Next, let's do the
null space together.
00:02:45.880 --> 00:02:46.380
OK.
00:02:46.380 --> 00:02:49.570
What's the dimension
of the null space?
00:02:53.030 --> 00:02:58.610
Well, the dimension
of the null space
00:02:58.610 --> 00:03:06.380
is always the number of columns
minus the number of pivots.
00:03:06.380 --> 00:03:06.880
Right?
00:03:06.880 --> 00:03:11.260
It's the number
of free variables.
00:03:11.260 --> 00:03:15.300
So here, that's just one.
00:03:15.300 --> 00:03:16.640
Good.
00:03:16.640 --> 00:03:21.070
And how do we find this one
vector in the null space?
00:03:34.190 --> 00:03:39.220
Well, what we do
is we can just plug
00:03:39.220 --> 00:03:44.380
in 1 for our free
variable, and we can
00:03:44.380 --> 00:03:47.770
backsolve to get the other two.
00:03:47.770 --> 00:03:53.360
So this equation tells me
that my second number is -1,
00:03:53.360 --> 00:03:58.920
and this equation tells me
that that third variable -3/5,
00:03:58.920 --> 00:04:00.000
if I can fit it in here.
00:04:04.590 --> 00:04:10.890
That's a -3/5, if it's difficult
to see that on the tape.
00:04:10.890 --> 00:04:14.810
Now, let's move on,
next, to the row space.
00:04:17.600 --> 00:04:19.380
Next is the row space.
00:04:19.380 --> 00:04:23.170
So how do we find the
dimension of the row space?
00:04:23.170 --> 00:04:27.730
I'm going to write row space
as column space of B transpose.
00:04:27.730 --> 00:04:29.070
How do we find that?
00:04:29.070 --> 00:04:32.450
Well, remember that one of
our big facts in this class
00:04:32.450 --> 00:04:34.780
is that the dimension
of the row space
00:04:34.780 --> 00:04:37.040
is the same as the dimension
of the column space.
00:04:37.040 --> 00:04:40.060
It's just the number of pivots.
00:04:40.060 --> 00:04:41.030
So that's good.
00:04:41.030 --> 00:04:43.660
It's 2.
00:04:43.660 --> 00:04:49.600
And how do we find a
basis for the row space?
00:04:49.600 --> 00:04:53.460
There are a couple ways
of thinking about this.
00:04:53.460 --> 00:04:57.290
One way to think about it is:
we got this upper triangular
00:04:57.290 --> 00:05:00.110
matrix from B by
doing elimination.
00:05:00.110 --> 00:05:04.420
And elimination doesn't
change the row space.
00:05:04.420 --> 00:05:12.700
So I can just use the two
pivot rows of the matrix U.
00:05:12.700 --> 00:05:25.620
Basis for my row
space here is: I just
00:05:25.620 --> 00:05:34.680
put these two pivot
rows together,
00:05:34.680 --> 00:05:38.650
and I get a basis
for this row space.
00:05:38.650 --> 00:05:42.110
The last one is always the
toughest and the trickiest.
00:05:42.110 --> 00:05:46.410
We have to do the left null
space or the null space of B
00:05:46.410 --> 00:05:46.910
transpose.
00:05:51.040 --> 00:05:54.730
First, let's compute
its dimension.
00:05:54.730 --> 00:05:59.100
What's the dimension of
this left null space?
00:05:59.100 --> 00:06:02.950
Well, there's a similar
formula to the one
00:06:02.950 --> 00:06:05.520
we used when we were
computing the dimension
00:06:05.520 --> 00:06:07.710
of the null space.
00:06:07.710 --> 00:06:12.210
It's just the number of rows
minus the number of pivots.
00:06:12.210 --> 00:06:13.930
So there are three rows.
00:06:13.930 --> 00:06:15.900
Our matrix is 3 by 3.
00:06:15.900 --> 00:06:17.370
And there are two pivots.
00:06:17.370 --> 00:06:20.330
So this is just
one-dimensional, again.
00:06:23.810 --> 00:06:29.180
We need to compute, now,
this left null space.
00:06:33.810 --> 00:06:36.950
Let me go back to
our original matrix.
00:06:36.950 --> 00:06:44.110
The way to do this is to take
B equals L*U, and invert L,
00:06:44.110 --> 00:06:48.660
and get E*B equals U. So we need
to move L over to the left-hand
00:06:48.660 --> 00:06:49.160
side.
00:06:51.670 --> 00:06:56.390
If we do that-- I'm just
going to write that down here.
00:06:59.990 --> 00:07:03.000
So what's the inverse
of the L matrix?
00:07:03.000 --> 00:07:13.720
We just get 1, -2,
1, 0, 1, 1, times B,
00:07:13.720 --> 00:07:17.670
is our U matrix, this
upper triangular matrix.
00:07:25.830 --> 00:07:27.710
Now that I moved L
over to the other side,
00:07:27.710 --> 00:07:35.400
I can read off the vectors
in my left null space.
00:07:35.400 --> 00:07:39.140
Now, I'm looking at
not my pivot variables
00:07:39.140 --> 00:07:42.670
but my free variables, because
it's some sort of null space.
00:07:42.670 --> 00:07:45.940
but I want to look
at this E matrix.
00:07:45.940 --> 00:07:50.040
So the third row of this
E matrix, the third row
00:07:50.040 --> 00:07:53.760
corresponds to
the three row here
00:07:53.760 --> 00:07:57.660
and when I multiply this
by B, I just get zeros,
00:07:57.660 --> 00:07:59.765
so this is in the
left null space.
00:08:03.570 --> 00:08:11.190
A basis for this
left null space is--
00:08:11.190 --> 00:08:17.460
see if I can fit it here--
just this [1, 0, 1].
00:08:17.460 --> 00:08:17.960
Good.
00:08:21.530 --> 00:08:30.850
So we've found the dimension of
and basis for all of the four
00:08:30.850 --> 00:08:32.559
fundamental subspaces.
00:08:32.559 --> 00:08:35.860
Before I move on, I
just want to recall
00:08:35.860 --> 00:08:38.809
which of the L matrix
or the U matrix
00:08:38.809 --> 00:08:43.150
we used for each
of these subspaces.
00:08:43.150 --> 00:08:49.580
So for the column space,
we used the pivot columns
00:08:49.580 --> 00:08:51.400
of the L matrix.
00:08:51.400 --> 00:08:56.510
For the null space, we
looked at the U matrix.
00:08:56.510 --> 00:09:02.080
For the row space, we also
looked at the U matrix.
00:09:02.080 --> 00:09:07.390
And for the left null space, we
needed to invert the L matrix
00:09:07.390 --> 00:09:10.000
and look at the free row.
00:09:14.730 --> 00:09:16.230
We're done with the problem.
00:09:16.230 --> 00:09:18.460
But the last thing
that's useful is
00:09:18.460 --> 00:09:22.790
to draw a picture,
which I have right here.
00:09:22.790 --> 00:09:25.270
I know in lecture
Professor Strang
00:09:25.270 --> 00:09:28.560
has drawn you some sort of
cartoon pictures of what
00:09:28.560 --> 00:09:30.230
these subspaces look like.
00:09:30.230 --> 00:09:33.390
But here I want to try
to actually draw them
00:09:33.390 --> 00:09:35.050
in a special case.
00:09:35.050 --> 00:09:41.440
So if you can read my drawing
here, what do we have?
00:09:41.440 --> 00:09:50.450
We have the row space here,
and the null space here.
00:09:53.064 --> 00:09:54.470
Right?
00:09:54.470 --> 00:10:02.640
And so B maps this
picture into this picture.
00:10:02.640 --> 00:10:06.180
The null space here--
all the scalar multiples
00:10:06.180 --> 00:10:09.710
of this vector-- all go
to 0, because they're
00:10:09.710 --> 00:10:11.520
in the null space.
00:10:11.520 --> 00:10:16.580
That's exactly
what B takes to 0.
00:10:16.580 --> 00:10:19.430
B takes everything else,
including the row space,
00:10:19.430 --> 00:10:20.695
into this column space.
00:10:23.270 --> 00:10:25.490
And what does B transpose do?
00:10:25.490 --> 00:10:30.480
Well, B transpose kills
this left null space,
00:10:30.480 --> 00:10:32.580
kills this vector,
and it take everything
00:10:32.580 --> 00:10:35.170
else into the row
space, into the column
00:10:35.170 --> 00:10:37.870
space of B transpose.
00:10:37.870 --> 00:10:38.910
OK.
00:10:38.910 --> 00:10:40.870
Thanks for doing this
exercise together.
00:10:40.870 --> 00:10:43.580
I hope this picture is helpful.