WEBVTT

00:00:00.500 --> 00:00:01.730
GILBERT STRANG: OK.

00:00:01.730 --> 00:00:07.780
So this is a "prepare the way"
video about symmetric matrices

00:00:07.780 --> 00:00:11.000
and complex matrices.

00:00:11.000 --> 00:00:16.850
We'll see symmetric matrices
in second order systems

00:00:16.850 --> 00:00:18.580
of differential equations.

00:00:18.580 --> 00:00:21.410
Symmetric matrices are the best.

00:00:21.410 --> 00:00:23.440
They have special
properties, and we

00:00:23.440 --> 00:00:28.700
want to see what are the special
properties of the eigenvalues

00:00:28.700 --> 00:00:29.920
and the eigenvectors?

00:00:29.920 --> 00:00:32.540
And I guess the
title of this lecture

00:00:32.540 --> 00:00:34.630
tells you what those
properties are.

00:00:34.630 --> 00:00:36.860
So if a matrix is
symmetric-- and I'll

00:00:36.860 --> 00:00:40.670
use capital S for a
symmetric matrix--

00:00:40.670 --> 00:00:44.730
the first point is the
eigenvalues are real,

00:00:44.730 --> 00:00:46.330
which is not automatic.

00:00:46.330 --> 00:00:49.490
But it's always true if
the matrix is symmetric.

00:00:49.490 --> 00:00:52.210
And the second, even
more special point

00:00:52.210 --> 00:00:56.890
is that the eigenvectors are
perpendicular to each other.

00:00:56.890 --> 00:00:59.600
Different eigenvectors
for different eigenvalues

00:00:59.600 --> 00:01:01.360
come out perpendicular.

00:01:01.360 --> 00:01:03.290
Those are beautiful properties.

00:01:03.290 --> 00:01:04.030
They pay off.

00:01:07.400 --> 00:01:11.070
So that's the symmetric matrix,
and that's what I just said.

00:01:11.070 --> 00:01:13.980
Real lambda, orthogonal x.

00:01:13.980 --> 00:01:17.990
Also, we could look at
antisymmetric matrices.

00:01:17.990 --> 00:01:21.010
The transpose is
minus the matrix.

00:01:21.010 --> 00:01:24.140
In that case, we don't
have real eigenvalues.

00:01:24.140 --> 00:01:28.770
In fact, we are sure to have
pure, imaginary eigenvalues.

00:01:28.770 --> 00:01:33.770
I times something on
the imaginary axis.

00:01:33.770 --> 00:01:38.060
But again, the eigenvectors
will be orthogonal.

00:01:38.060 --> 00:01:42.810
However, they will
also be complex.

00:01:42.810 --> 00:01:47.720
When we have
antisymmetric matrices,

00:01:47.720 --> 00:01:49.480
we get into complex numbers.

00:01:49.480 --> 00:01:52.810
Can't help it, even
if the matrix is real.

00:01:52.810 --> 00:01:59.110
And then finally is the
family of orthogonal matrices.

00:01:59.110 --> 00:02:04.460
And those matrices have
eigenvalues of size 1,

00:02:04.460 --> 00:02:06.280
possibly complex.

00:02:06.280 --> 00:02:09.180
But the magnitude
of the number is 1.

00:02:09.180 --> 00:02:12.050
And again, the eigenvectors
are orthogonal.

00:02:12.050 --> 00:02:19.280
This is the great family of
real, imaginary, and unit

00:02:19.280 --> 00:02:22.300
circle for the eigenvalues.

00:02:22.300 --> 00:02:23.230
OK.

00:02:23.230 --> 00:02:25.440
I want to do examples.

00:02:25.440 --> 00:02:27.620
So I'll just have an
example of every one.

00:02:30.590 --> 00:02:33.350
So there's a symmetric matrix.

00:02:33.350 --> 00:02:36.610
There's a antisymmetric matrix.

00:02:36.610 --> 00:02:40.130
If I transpose it,
it changes sign.

00:02:40.130 --> 00:02:45.420
Here is a combination, not
symmetric, not antisymmetric,

00:02:45.420 --> 00:02:47.150
but still a good matrix.

00:02:47.150 --> 00:02:52.680
And there is an orthogonal
matrix, orthogonal columns.

00:02:52.680 --> 00:02:55.560
And those columns have length 1.

00:02:55.560 --> 00:02:59.160
That's why I've got the
square root of 2 in there.

00:02:59.160 --> 00:03:03.170
So these are the
special matrices here.

00:03:03.170 --> 00:03:06.830
Can I just draw a little
picture of the complex plane?

00:03:06.830 --> 00:03:09.340
There is the real axis.

00:03:09.340 --> 00:03:10.965
Here is the imaginary axis.

00:03:13.660 --> 00:03:22.900
And here's the unit circle,
not greatly circular but close.

00:03:22.900 --> 00:03:28.810
Now-- eigenvalues
are on the real axis

00:03:28.810 --> 00:03:34.180
when S transpose equals S.
They're on the imaginary axis

00:03:34.180 --> 00:03:40.420
when A transpose equals minus A.
And they're on the unit circle

00:03:40.420 --> 00:03:44.980
when Q transpose
Q is the identity.

00:03:44.980 --> 00:03:48.450
Q transpose is Q
inverse in this case.

00:03:48.450 --> 00:03:50.590
Q transpose is Q inverse.

00:03:50.590 --> 00:03:52.630
Here the transpose
is the matrix.

00:03:52.630 --> 00:03:54.920
Here the transpose
is minus the matrix.

00:03:54.920 --> 00:03:56.790
And you see the
beautiful picture

00:03:56.790 --> 00:04:00.100
of eigenvalues, where they are.

00:04:00.100 --> 00:04:04.400
And the eigenvectors for
all of those are orthogonal.

00:04:04.400 --> 00:04:05.960
Let me find them.

00:04:05.960 --> 00:04:13.500
Here that symmetric matrix
has lambda as 2 and 4.

00:04:13.500 --> 00:04:15.700
The trace is 6.

00:04:15.700 --> 00:04:17.329
The determinant is 8.

00:04:17.329 --> 00:04:18.769
That's the right answer.

00:04:18.769 --> 00:04:20.680
Lambda equal 2 and 4.

00:04:20.680 --> 00:04:28.130
And x would be 1
and minus 1 for 2.

00:04:28.130 --> 00:04:32.940
And for 4, it's 1 and 1.

00:04:32.940 --> 00:04:35.300
Orthogonal.

00:04:35.300 --> 00:04:39.520
Orthogonal eigenvectors--
take the dot product of those,

00:04:39.520 --> 00:04:42.520
you get 0 and real eigenvalues.

00:04:42.520 --> 00:04:43.920
What about A?

00:04:46.530 --> 00:04:49.080
Antisymmetric.

00:04:49.080 --> 00:04:55.355
The equation I-- when I do
determinant of lambda minus A,

00:04:55.355 --> 00:05:00.930
I get lambda squared plus
1 equals 0 for this one.

00:05:00.930 --> 00:05:04.730
That leads me to lambda
squared plus 1 equals 0.

00:05:04.730 --> 00:05:11.260
So that gives me lambda is
i and minus i, as promised,

00:05:11.260 --> 00:05:14.140
on the imaginary axis.

00:05:14.140 --> 00:05:18.460
And I guess that that matrix
is also an orthogonal matrix.

00:05:18.460 --> 00:05:21.090
And those eigenvalues,
i and minus i,

00:05:21.090 --> 00:05:23.320
are also on the circle.

00:05:23.320 --> 00:05:26.730
So that A is also a Q. OK.

00:05:26.730 --> 00:05:29.780
What are the
eigenvectors for that?

00:05:29.780 --> 00:05:32.030
I think that the
eigenvectors turn out

00:05:32.030 --> 00:05:36.370
to be 1 i and 1 minus i.

00:05:39.940 --> 00:05:41.050
Oh.

00:05:41.050 --> 00:05:42.700
Those are orthogonal.

00:05:42.700 --> 00:05:45.970
I'll have to tell you
about orthogonality

00:05:45.970 --> 00:05:48.210
for complex vectors.

00:05:48.210 --> 00:05:50.630
Let me complete these examples.

00:05:50.630 --> 00:05:53.470
What about the
eigenvalues of this one?

00:05:53.470 --> 00:05:55.880
Well, that's an easy one.

00:05:55.880 --> 00:06:00.060
Can you connect that to A?

00:06:00.060 --> 00:06:04.570
B is just A plus 3
times the identity--

00:06:04.570 --> 00:06:06.670
to put 3's on the diagonal.

00:06:06.670 --> 00:06:09.500
So I'm expecting
here the lambdas

00:06:09.500 --> 00:06:13.340
are-- if here they
were i and minus i.

00:06:13.340 --> 00:06:17.460
All I've done is add 3 times the
identity, so I'm just adding 3.

00:06:17.460 --> 00:06:19.270
I'm shifting by 3.

00:06:19.270 --> 00:06:23.560
I'll have 3 plus
i and 3 minus i.

00:06:23.560 --> 00:06:24.960
And the same eigenvectors.

00:06:29.380 --> 00:06:32.210
So that's a complex number.

00:06:32.210 --> 00:06:37.020
That matrix was not
perfectly antisymmetric.

00:06:37.020 --> 00:06:39.250
It's not perfectly symmetric.

00:06:39.250 --> 00:06:43.850
So that gave me a 3 plus i
somewhere not on the axis

00:06:43.850 --> 00:06:45.750
or that axis or the circle.

00:06:45.750 --> 00:06:49.660
Out there-- 3 plus
i and 3 minus i.

00:06:49.660 --> 00:06:55.860
And finally, this one,
the orthogonal matrix.

00:06:55.860 --> 00:06:57.425
What are the
eigenvalues of that?

00:07:01.320 --> 00:07:02.650
Let's see.

00:07:02.650 --> 00:07:06.830
I can see-- here I've added 1
times the identity, just added

00:07:06.830 --> 00:07:11.140
the identity to minus 1, 1.

00:07:11.140 --> 00:07:14.990
So again, I have this minus
1, 1 plus the identity.

00:07:14.990 --> 00:07:24.430
So I would have 1 plus i and
1 minus i from the matrix.

00:07:24.430 --> 00:07:28.330
And now I've got a division
by square root of 2,

00:07:28.330 --> 00:07:31.730
square root of 2.

00:07:31.730 --> 00:07:36.560
And those numbers
lambda-- you recognize

00:07:36.560 --> 00:07:42.150
that when you see that number,
that is on the unit circle.

00:07:42.150 --> 00:07:43.770
Where is it on the unit circle?

00:07:43.770 --> 00:07:45.440
1 plus i.

00:07:45.440 --> 00:07:48.440
1 plus i over square root of 2.

00:07:48.440 --> 00:07:50.540
Square root of 2
brings it down there.

00:07:50.540 --> 00:07:51.740
There's 1.

00:07:51.740 --> 00:07:53.230
There's i.

00:07:53.230 --> 00:07:54.770
Divide by square root of 2.

00:07:54.770 --> 00:07:56.560
That puts us on the circle.

00:07:56.560 --> 00:08:00.960
That's 1 plus i over
square root of 2.

00:08:00.960 --> 00:08:08.950
And here is 1 plus i, 1 minus
i over square root of two.

00:08:08.950 --> 00:08:09.775
Complex conjugates.

00:08:13.950 --> 00:08:16.990
When I say "complex
conjugate," that means

00:08:16.990 --> 00:08:21.300
I change every i to a minus i.

00:08:21.300 --> 00:08:23.750
I flip across the real axis.

00:08:23.750 --> 00:08:25.500
I'd want to do that in a minute.

00:08:25.500 --> 00:08:29.000
So are there more lessons
to see for these examples?

00:08:29.000 --> 00:08:33.559
Again, real eigenvalues and
real eigenvectors-- no problem.

00:08:33.559 --> 00:08:36.340
Here, imaginary eigenvalues.

00:08:36.340 --> 00:08:38.510
Here, complex eigenvalues.

00:08:38.510 --> 00:08:42.350
Here, complex eigenvalues
on the circle.

00:08:42.350 --> 00:08:43.730
On the circle.

00:08:43.730 --> 00:08:44.760
OK.

00:08:44.760 --> 00:08:48.970
And each of those
facts that I just

00:08:48.970 --> 00:08:52.470
said about the location
of the eigenvalues--

00:08:52.470 --> 00:08:57.000
it has a short proof, but maybe
I won't give the proof here.

00:08:57.000 --> 00:08:59.890
It's the fact that
you want to remember.

00:08:59.890 --> 00:09:03.910
Can I bring down again, just
for a moment, these main facts?

00:09:03.910 --> 00:09:08.210
Real, from
symmetric-- imaginary,

00:09:08.210 --> 00:09:14.810
from antisymmetric--
magnitude 1, from orthogonal.

00:09:14.810 --> 00:09:15.520
OK.

00:09:15.520 --> 00:09:20.620
Now I feel I've talking
about complex numbers,

00:09:20.620 --> 00:09:27.850
and I really should say-- I
should pay attention to that.

00:09:27.850 --> 00:09:29.530
Complex numbers.

00:09:29.530 --> 00:09:32.950
So I have lambda as a plus ib.

00:09:35.690 --> 00:09:38.630
What do I mean by the
"magnitude" of that number?

00:09:38.630 --> 00:09:41.810
What's the magnitude
of lambda is a plus ib?

00:09:41.810 --> 00:09:47.600
Again, I go along a, up b.

00:09:47.600 --> 00:09:54.770
Here is the lambda,
the complex number.

00:09:54.770 --> 00:09:56.630
And I want to know
the length of that.

00:09:56.630 --> 00:09:59.250
Well, everybody knows
the length of that.

00:09:59.250 --> 00:10:04.310
Thank goodness Pythagoras
lived, or his team lived.

00:10:04.310 --> 00:10:07.890
It's the square root of
a squared plus b squared.

00:10:11.080 --> 00:10:19.060
And notice what that-- how do I
get that number from this one?

00:10:19.060 --> 00:10:20.510
It's important.

00:10:20.510 --> 00:10:25.110
If I multiply a plus
ib times a minus

00:10:25.110 --> 00:10:32.180
ib-- so I have lambda-- that's
a plus ib-- times lambda

00:10:32.180 --> 00:10:37.040
conjugate-- that's a minus ib--
if I multiply those, that gives

00:10:37.040 --> 00:10:38.520
me a squared plus b squared.

00:10:38.520 --> 00:10:41.030
So I take the square
root, and this

00:10:41.030 --> 00:10:45.020
is what I would call the
"magnitude" of lambda.

00:10:45.020 --> 00:10:50.080
So the magnitude of a number
is that positive length.

00:10:50.080 --> 00:10:52.890
And it can be found-- you
take the complex number

00:10:52.890 --> 00:10:54.940
times its conjugate.

00:10:54.940 --> 00:10:58.010
That gives you a squared
plus b squared, and then

00:10:58.010 --> 00:11:00.410
take the square root.

00:11:00.410 --> 00:11:02.700
Basic facts about
complex numbers.

00:11:02.700 --> 00:11:03.200
OK.

00:11:03.200 --> 00:11:05.970
What about complex vectors?

00:11:05.970 --> 00:11:08.220
What is the dot product?

00:11:08.220 --> 00:11:11.710
What is the correct
x transpose x?

00:11:11.710 --> 00:11:13.850
Well, it's not x transpose x.

00:11:13.850 --> 00:11:23.280
Suppose x is the vector 1 i, as
we saw that as an eigenvector.

00:11:23.280 --> 00:11:25.750
What's the length
of that vector?

00:11:25.750 --> 00:11:30.850
The length of that vector is
not 1 squared plus i squared.

00:11:30.850 --> 00:11:36.000
1 squared plus i squared would
be 1 plus minus 1 would be 0.

00:11:36.000 --> 00:11:40.390
The length of that vector
is the size of this squared

00:11:40.390 --> 00:11:45.330
plus the size of this
squared, square root.

00:11:45.330 --> 00:11:46.220
Here we go.

00:11:46.220 --> 00:11:51.920
The length of x squared-- the
length of the vector squared--

00:11:51.920 --> 00:11:54.565
will be the vector.

00:11:57.330 --> 00:12:04.850
As always, I can find
it from a dot product.

00:12:04.850 --> 00:12:11.802
But I have to take
the conjugate of that.

00:12:11.802 --> 00:12:14.440
If I want the
length of x, I have

00:12:14.440 --> 00:12:18.722
to take-- I would usually
take x transpose x, right?

00:12:18.722 --> 00:12:23.780
If I have a real vector x,
then I find its dot product

00:12:23.780 --> 00:12:26.170
with itself, and
Pythagoras tells me

00:12:26.170 --> 00:12:27.740
I have the length squared.

00:12:27.740 --> 00:12:32.840
But if the things are complex--
I want minus i times i.

00:12:32.840 --> 00:12:37.320
I want to get lambda
times lambda bar.

00:12:37.320 --> 00:12:40.870
I want to get a positive number.

00:12:40.870 --> 00:12:44.200
Minus i times i is plus 1.

00:12:44.200 --> 00:12:47.520
Minus i times i is plus 1.

00:12:47.520 --> 00:12:49.970
So I must, must do that.

00:12:52.570 --> 00:12:55.030
So that's really what
"orthogonal" would mean.

00:12:59.290 --> 00:13:03.280
"Orthogonal complex
vectors" mean--

00:13:03.280 --> 00:13:13.140
"orthogonal vectors" mean that
x conjugate transpose y is 0.

00:13:13.140 --> 00:13:18.780
That's what I mean by
"orthogonal eigenvectors"

00:13:18.780 --> 00:13:20.830
when those eigenvectors
are complex.

00:13:20.830 --> 00:13:24.100
I must remember to take
the complex conjugate.

00:13:24.100 --> 00:13:26.830
And I also do it for matrices.

00:13:26.830 --> 00:13:30.815
So if I have a symmetric
matrix-- S transpose S.

00:13:30.815 --> 00:13:31.925
I know what that means.

00:13:35.630 --> 00:13:38.280
But suppose S is complex.

00:13:38.280 --> 00:13:39.780
Suppose S is complex.

00:13:39.780 --> 00:13:48.400
Then for a complex matrix, I
would look at S bar transpose

00:13:48.400 --> 00:13:52.100
equal S.

00:13:52.100 --> 00:13:55.990
Every time I transpose,
if I have complex numbers,

00:13:55.990 --> 00:13:57.820
I should take the
complex conjugate.

00:13:57.820 --> 00:13:59.930
MATLAB does that automatically.

00:13:59.930 --> 00:14:07.920
If you ask for x prime,
it will produce-- not just

00:14:07.920 --> 00:14:11.460
it'll change a column to
a row with that transpose,

00:14:11.460 --> 00:14:12.870
that prime.

00:14:12.870 --> 00:14:15.800
And it will take the
complex conjugate.

00:14:15.800 --> 00:14:18.731
So we must remember
always to do that.

00:14:18.731 --> 00:14:19.230
Yeah.

00:14:19.230 --> 00:14:24.210
And in fact, if S
was a complex matrix

00:14:24.210 --> 00:14:28.440
but it had that property--
let me give an example.

00:14:28.440 --> 00:14:32.740
So here's an S, an
example of that.

00:14:32.740 --> 00:14:38.500
1, 2, i, and minus i.

00:14:38.500 --> 00:14:41.020
So I have a complex matrix.

00:14:41.020 --> 00:14:45.340
And if I transpose it and
take complex conjugates,

00:14:45.340 --> 00:14:49.390
that brings me
back to S. And this

00:14:49.390 --> 00:15:00.860
is called a "Hermitian matrix"
among other possible names.

00:15:00.860 --> 00:15:04.340
Hermite was a important
mathematician.

00:15:04.340 --> 00:15:08.490
He studied this
complex case, and he

00:15:08.490 --> 00:15:14.020
understood to take the conjugate
as well as the transpose.

00:15:14.020 --> 00:15:19.730
And sometimes I would write
it as SH in his honor.

00:15:19.730 --> 00:15:24.460
So if I want one
symbol to do it-- SH.

00:15:24.460 --> 00:15:29.230
In engineering, sometimes
S with a star tells me,

00:15:29.230 --> 00:15:33.980
take the conjugate when
you transpose a matrix.

00:15:33.980 --> 00:15:38.830
So that's main facts about-- let
me bring those main facts down

00:15:38.830 --> 00:15:45.730
again-- orthogonal eigenvectors
and location of eigenvalues.

00:15:45.730 --> 00:15:48.310
Now I'm ready to solve
differential equations.

00:15:48.310 --> 00:15:50.030
Thank you.