WEBVTT

00:00:00.500 --> 00:00:01.420
GILBERT STRANG: OK.

00:00:01.420 --> 00:00:07.010
So this is a video in which we
go for second-order equations,

00:00:07.010 --> 00:00:08.680
constant coefficients.

00:00:08.680 --> 00:00:12.030
We look for the
impulse response,

00:00:12.030 --> 00:00:14.860
the key function in
this whole business,

00:00:14.860 --> 00:00:17.650
and the step response, too.

00:00:17.650 --> 00:00:19.850
So those are the responses.

00:00:19.850 --> 00:00:23.780
So I'm going to call g--
that will be the impulse

00:00:23.780 --> 00:00:26.000
response, where the
right-hand side is

00:00:26.000 --> 00:00:32.259
a delta function, an impulse,
a sudden force at the moment t

00:00:32.259 --> 00:00:33.360
equals 0.

00:00:33.360 --> 00:00:35.200
So that's the equation.

00:00:35.200 --> 00:00:36.680
That's the impulse.

00:00:36.680 --> 00:00:40.700
And g is the response, and
we want a formula for it.

00:00:40.700 --> 00:00:45.790
Then the other possibility,
very interesting possibility,

00:00:45.790 --> 00:00:49.550
is when the right-hand
side is a step function.

00:00:49.550 --> 00:00:52.920
And then we want the
response to that function.

00:00:52.920 --> 00:00:54.010
I click a switch.

00:00:56.520 --> 00:00:59.800
The machine starts
working, and it

00:00:59.800 --> 00:01:03.430
approaches a steady response.

00:01:03.430 --> 00:01:06.040
The solution rises from 0.

00:01:06.040 --> 00:01:13.950
So it starts at r of 0
equals r prime of 0 equals 0.

00:01:13.950 --> 00:01:16.630
The step response
starts from rest.

00:01:16.630 --> 00:01:20.430
The action happens when I
click a switch at t equals 0,

00:01:20.430 --> 00:01:24.130
and then r of t will
rise to a constant.

00:01:24.130 --> 00:01:26.700
Very, very important solutions.

00:01:26.700 --> 00:01:31.710
But we'll focus
especially on this one.

00:01:31.710 --> 00:01:32.530
OK.

00:01:32.530 --> 00:01:36.400
So that's our equation with
the right-hand side delta.

00:01:36.400 --> 00:01:38.760
And of course, that
right-hand side

00:01:38.760 --> 00:01:43.460
is not totally familiar,
not as nice as e to the st.

00:01:43.460 --> 00:01:46.770
But there is something
that-- there is another way

00:01:46.770 --> 00:01:51.150
to approach it-- that's
a key idea here--

00:01:51.150 --> 00:01:55.210
that gives us this all-important
function from solving

00:01:55.210 --> 00:01:56.350
a null equation.

00:01:56.350 --> 00:01:58.700
How's that?

00:01:58.700 --> 00:02:01.570
I start with a null
equation, but now this

00:02:01.570 --> 00:02:03.885
had no initial condition.

00:02:03.885 --> 00:02:08.229
So this one started from g
of 0 and g prime of 0 both 0.

00:02:08.229 --> 00:02:11.860
And everything happened,
boom, from the delta function.

00:02:11.860 --> 00:02:14.320
This is the same function.

00:02:14.320 --> 00:02:17.810
Except when I look to
see what happens at t

00:02:17.810 --> 00:02:24.800
equals 0, what happens is g
prime immediately jumps to 1.

00:02:24.800 --> 00:02:29.650
So another way I can approach
g, the computation of g,

00:02:29.650 --> 00:02:33.080
is to think of it-- I'm just
looking for a null solution.

00:02:33.080 --> 00:02:38.190
I'm looking for the null
solution that starts from 0.

00:02:38.190 --> 00:02:46.020
But it starts with an initial
derivative, slope equal 1.

00:02:46.020 --> 00:02:52.240
So I know that g
is a combination.

00:02:52.240 --> 00:02:57.475
So I know how to solve equations
like that, null equation.

00:02:57.475 --> 00:03:01.140
You remember s1 and s2?

00:03:01.140 --> 00:03:06.190
I look at s squared-- I've
made this coefficient 1-- so s

00:03:06.190 --> 00:03:11.010
squared plus Bs plus C equals 0.

00:03:11.010 --> 00:03:15.710
That gives us s1 and s2.

00:03:19.100 --> 00:03:21.710
And now I'll tell
you what the g is.

00:03:21.710 --> 00:03:25.620
So that gives us the s1 and
the s2 in the null solution,

00:03:25.620 --> 00:03:28.000
and we're looking
for a null solution.

00:03:28.000 --> 00:03:33.000
So our g of t is
some combination of e

00:03:33.000 --> 00:03:36.680
to the s1t and e to the s2t.

00:03:40.040 --> 00:03:41.910
OK?

00:03:41.910 --> 00:03:44.157
It's some combination of those.

00:03:44.157 --> 00:03:47.120
And we want it to be 0.

00:03:47.120 --> 00:03:53.990
So no surprise, if I subtract
those-- I'm starting at t

00:03:53.990 --> 00:03:54.490
equals 0.

00:03:54.490 --> 00:03:57.200
I'm starting--
this is 1 minus 1.

00:03:57.200 --> 00:03:58.340
It's 0.

00:03:58.340 --> 00:04:04.820
And now I just have to
fix the initial slope,

00:04:04.820 --> 00:04:07.610
the first derivative, to be 1.

00:04:07.610 --> 00:04:09.290
Well, what's the
derivative of this?

00:04:09.290 --> 00:04:11.310
This brings down an s1.

00:04:11.310 --> 00:04:14.090
This brings down an s2.

00:04:14.090 --> 00:04:16.959
At t equals 0, I'm
getting an s1 minus s2.

00:04:16.959 --> 00:04:22.600
So I'll just divide
by that, s1 minus s2.

00:04:22.600 --> 00:04:24.340
There you go.

00:04:24.340 --> 00:04:29.760
That's the impulse
response-- a null solution

00:04:29.760 --> 00:04:35.760
that satisfies these
special initial conditions.

00:04:35.760 --> 00:04:38.830
So that's the function
in mathematics

00:04:38.830 --> 00:04:41.640
that's sometimes called
the fundamental solution.

00:04:41.640 --> 00:04:46.990
It's a solution from which
you can create all solutions.

00:04:46.990 --> 00:04:49.440
It's really the
mother of solutions

00:04:49.440 --> 00:04:53.430
to this second-order
differential equation.

00:04:53.430 --> 00:04:55.990
Because if I have
another forcing function,

00:04:55.990 --> 00:04:58.140
this tells me that growth rate.

00:04:58.140 --> 00:05:03.950
It's just like e to the at
for a first-order equation.

00:05:03.950 --> 00:05:05.490
Remember the growth
rate e to the

00:05:05.490 --> 00:05:08.880
at for the simple first-order
equation with interest

00:05:08.880 --> 00:05:11.480
rate coefficient a?

00:05:11.480 --> 00:05:13.310
Now we have two.

00:05:13.310 --> 00:05:16.800
Instead of a, we
have an s1 and an s2,

00:05:16.800 --> 00:05:20.300
and that's the special function.

00:05:20.300 --> 00:05:21.580
OK.

00:05:21.580 --> 00:05:25.710
We need to get more insight
on that for particular cases.

00:05:25.710 --> 00:05:32.140
So let me show you
the same function when

00:05:32.140 --> 00:05:35.230
I have no damping.

00:05:35.230 --> 00:05:38.870
Start with that case,
always the easiest case.

00:05:38.870 --> 00:05:42.270
When B is 0-- B was the
damping coefficient,

00:05:42.270 --> 00:05:46.370
the first derivative in
our differential equation.

00:05:46.370 --> 00:05:48.720
Can I just bring down the
differential equation?

00:05:48.720 --> 00:05:52.720
When B is 0 here,
that's no damping.

00:05:52.720 --> 00:05:55.680
I just have the second
derivative and the function.

00:05:55.680 --> 00:05:58.470
That's when things
oscillate forever.

00:05:58.470 --> 00:06:00.020
So that's what will happen.

00:06:00.020 --> 00:06:05.690
With B equals 0, I
have pure oscillation.

00:06:05.690 --> 00:06:13.070
The s1 and s2 are cosines
and sines that oscillate.

00:06:13.070 --> 00:06:18.150
Or it's neater to stay
with exponentials, the i

00:06:18.150 --> 00:06:20.810
omega and minus i
omega, where that's

00:06:20.810 --> 00:06:23.530
omega n, the natural frequency.

00:06:23.530 --> 00:06:29.470
Now, if I just plug in that
s1 and s2-- the plus is s1

00:06:29.470 --> 00:06:32.020
and the minus is s2--
I plug it in there,

00:06:32.020 --> 00:06:35.580
I get a nice formula for g of t.

00:06:35.580 --> 00:06:40.600
So that's what g of t
looks like for no damping.

00:06:40.600 --> 00:06:43.880
It just oscillates.

00:06:43.880 --> 00:06:44.830
OK.

00:06:44.830 --> 00:06:48.190
The next case is underdamping.

00:06:48.190 --> 00:06:51.260
It's good to see all
these cases each time.

00:06:51.260 --> 00:06:54.200
So that's a small value
of B. Underdamping

00:06:54.200 --> 00:07:02.770
means there is some
damping, but it's

00:07:02.770 --> 00:07:07.380
small enough so there's
now a real part,

00:07:07.380 --> 00:07:10.400
but there's still
an imaginary part.

00:07:10.400 --> 00:07:16.610
So this is, in a way, the trick
is the case when s is complex.

00:07:16.610 --> 00:07:20.470
If I go higher with the
dampening, increase B further,

00:07:20.470 --> 00:07:22.035
then I'll hit a
point where there

00:07:22.035 --> 00:07:24.950
are two real solutions equal.

00:07:24.950 --> 00:07:28.220
And if I push B beyond
that, I've got overdamping,

00:07:28.220 --> 00:07:31.230
and those two real
solutions separate.

00:07:31.230 --> 00:07:34.020
They're different,
but they're real.

00:07:34.020 --> 00:07:36.700
And then my formula,
in that case,

00:07:36.700 --> 00:07:39.480
overdamping, that would
be the best formula.

00:07:39.480 --> 00:07:43.510
But with underdamping, I
can see the oscillation.

00:07:43.510 --> 00:07:47.920
If I just plug in those two
solutions for s1 and s2,

00:07:47.920 --> 00:07:52.800
you'll see that I have the
e to the minus B over 2t

00:07:52.800 --> 00:07:54.390
appears throughout.

00:07:54.390 --> 00:07:57.870
But then I have the sine
of omega over omega,

00:07:57.870 --> 00:08:02.570
same as I had before, except
now the damping frequency

00:08:02.570 --> 00:08:06.370
is a bit slower than
the natural frequency.

00:08:06.370 --> 00:08:08.460
Damping slows that
frequency down.

00:08:08.460 --> 00:08:11.540
And in a different
video, we had a formula

00:08:11.540 --> 00:08:14.740
for omega damping,
omega damping.

00:08:14.740 --> 00:08:24.100
And then increase the damping
some more, then this part--

00:08:24.100 --> 00:08:27.580
this omega damping goes to 0.

00:08:27.580 --> 00:08:30.560
We don't see any imaginary
part in the solution.

00:08:30.560 --> 00:08:33.344
We see two equal real values.

00:08:33.344 --> 00:08:34.299
They're simple.

00:08:34.299 --> 00:08:37.169
They have to be just minus B/2.

00:08:37.169 --> 00:08:42.950
So that's a case of two
s's coming together.

00:08:42.950 --> 00:08:45.140
And when two things
come together,

00:08:45.140 --> 00:08:47.980
we're used to seeing
a factor t appear.

00:08:47.980 --> 00:08:51.360
So I have that they came
together at minus B/2.

00:08:51.360 --> 00:08:53.500
So I have the
exponential of that.

00:08:53.500 --> 00:08:58.180
But I have a factor t
from the merge of the two.

00:08:58.180 --> 00:09:04.350
And then if B increases beyond
that, that's my formula.

00:09:04.350 --> 00:09:06.020
The two s's are real.

00:09:09.310 --> 00:09:11.890
I don't think one
memorizes all this.

00:09:11.890 --> 00:09:14.600
I had to look them up and
write them on the board

00:09:14.600 --> 00:09:16.910
before starting this video.

00:09:16.910 --> 00:09:22.320
But I hope you see that
they're extremely nice.

00:09:22.320 --> 00:09:25.670
The no-damping case
with [? pure ?]

00:09:25.670 --> 00:09:33.350
frequencies and the underdamping
case with a real-- a decay.

00:09:33.350 --> 00:09:37.770
The critical damping when
you increase B further,

00:09:37.770 --> 00:09:41.570
you just have that
and no oscillation.

00:09:41.570 --> 00:09:45.200
And then beyond
that is overdamping.

00:09:45.200 --> 00:09:45.820
OK.

00:09:45.820 --> 00:09:51.060
So we're good for
the impulse response.

00:09:51.060 --> 00:09:54.520
And now I just have to say,
what's the step response?

00:09:54.520 --> 00:10:02.030
So can I end this video by
going back to my equation?

00:10:02.030 --> 00:10:04.700
I have to bring the board
down to show it to you.

00:10:04.700 --> 00:10:08.740
So now I'm going to deal
with the step response.

00:10:08.740 --> 00:10:11.290
So the equation is the same.

00:10:11.290 --> 00:10:14.460
I'm calling the
solution r for response.

00:10:14.460 --> 00:10:16.800
And the point is, the
right-hand side is now

00:10:16.800 --> 00:10:19.480
a step instead of a delta.

00:10:19.480 --> 00:10:22.980
So we'd like to solve that
equation starting from rest.

00:10:22.980 --> 00:10:28.800
So a switch went on, and I
want a formula for r of t.

00:10:28.800 --> 00:10:30.440
That's all that remains.

00:10:30.440 --> 00:10:35.100
And actually,
that's just-- well,

00:10:35.100 --> 00:10:40.140
you can see what the
particular solution is.

00:10:40.140 --> 00:10:42.080
We look at a
particular solution.

00:10:42.080 --> 00:10:45.720
Well, this right-hand side is 1.

00:10:45.720 --> 00:10:49.380
This right-hand side
is 1 beyond t equals 0.

00:10:49.380 --> 00:10:53.380
So I'm looking for a way
to get 1 out of that.

00:10:53.380 --> 00:10:55.560
Well, or it can be
just a constant.

00:10:55.560 --> 00:10:58.740
The particular solution
is the steady state

00:10:58.740 --> 00:10:59.915
that we're approaching.

00:10:59.915 --> 00:11:03.090
And there's one little
cool thing to do.

00:11:03.090 --> 00:11:08.660
Sometimes people who have the
dimensions and units of things

00:11:08.660 --> 00:11:11.665
clearly in their mind
will put a C in there.

00:11:16.630 --> 00:11:20.790
This is really the good thing
to do is to have that C in there

00:11:20.790 --> 00:11:25.195
because now the units
for r are the same as-- r

00:11:25.195 --> 00:11:26.880
is going to go to the 1.

00:11:26.880 --> 00:11:31.085
The steady state will
be 1 now because I

00:11:31.085 --> 00:11:33.810
have Cr equals C times 1.

00:11:33.810 --> 00:11:39.440
And out in infinity, the
simple solution is r equal 1.

00:11:39.440 --> 00:11:41.700
If when r is 1, its
derivative is 0.

00:11:41.700 --> 00:11:43.070
Second derivative's 0.

00:11:43.070 --> 00:11:45.190
r equal 1 is a solution.

00:11:45.190 --> 00:11:47.370
It's a particular solution.

00:11:47.370 --> 00:11:49.981
It's the steady-state solution.

00:11:49.981 --> 00:11:51.304
Good.

00:11:51.304 --> 00:11:57.270
But that r of t equal 1
does not start correctly.

00:11:57.270 --> 00:12:01.230
We want to start at
0, with a slope of 0.

00:12:01.230 --> 00:12:07.550
So I have to subtract off one
of these particular solutions

00:12:07.550 --> 00:12:09.250
with e to the s1.

00:12:09.250 --> 00:12:12.070
And now I have to
get it so that I

00:12:12.070 --> 00:12:15.960
have to subtract it off so
that this thing starts from 0.

00:12:15.960 --> 00:12:17.500
Let me see if I can do it.

00:12:17.500 --> 00:12:24.520
I think maybe if I
do an s2 e to the s1t

00:12:24.520 --> 00:12:30.700
and subtract off
an s1 e to the s2t.

00:12:35.610 --> 00:12:38.610
Do you see what that
one has achieved?

00:12:38.610 --> 00:12:45.520
At t equals 0, I have--
well, at least at t equals 0,

00:12:45.520 --> 00:12:48.700
I've made the derivative
0 because the derivative

00:12:48.700 --> 00:12:50.720
will bring down an s1 there.

00:12:50.720 --> 00:12:53.190
That derivative will
bring down an s2.

00:12:53.190 --> 00:12:57.610
And when I put t equals 0,
I get s1, s2 minus s1, s2.

00:12:57.610 --> 00:12:59.320
Good, good, good.

00:12:59.320 --> 00:13:00.110
OK.

00:13:00.110 --> 00:13:05.930
Now I think that together
they're all correct.

00:13:05.930 --> 00:13:10.940
I need to divide by s1 minus s2.

00:13:10.940 --> 00:13:13.310
Let me just say,
I think that's it.

00:13:13.310 --> 00:13:14.640
I think that's it.

00:13:14.640 --> 00:13:17.740
It wouldn't be a bad
idea if I just checked.

00:13:17.740 --> 00:13:23.480
And having checked, I've
learned that that's a plus sign.

00:13:23.480 --> 00:13:24.840
OK.

00:13:24.840 --> 00:13:27.980
So the graph of r.

00:13:27.980 --> 00:13:30.030
This is a graph of r of t.

00:13:30.030 --> 00:13:35.830
It starts from 0,
and it rises to 1.

00:13:35.830 --> 00:13:38.700
Asymptotically it's 1.

00:13:38.700 --> 00:13:42.520
This is a graph of r of t.

00:13:42.520 --> 00:13:45.480
And in practice, that's
a very important number.

00:13:45.480 --> 00:13:47.350
What is the rise time?

00:13:47.350 --> 00:13:50.440
How far do you have to go in
time before it rises up to,

00:13:50.440 --> 00:13:53.800
let's say, 95% of 1?

00:13:53.800 --> 00:13:56.930
All these questions are
extremely practical questions

00:13:56.930 --> 00:13:59.076
for an engineer.

00:13:59.076 --> 00:14:01.110
What's the rise time?

00:14:01.110 --> 00:14:03.290
And you're playing
with this formula.

00:14:03.290 --> 00:14:07.610
So let me just make
another comment

00:14:07.610 --> 00:14:11.740
on this r of t step response.

00:14:11.740 --> 00:14:15.160
My other comment is
I've emphasized g of t,

00:14:15.160 --> 00:14:19.240
the impulse response is like
responsible for everything.

00:14:19.240 --> 00:14:21.170
It's always with us.

00:14:21.170 --> 00:14:24.690
And how are those connected?

00:14:24.690 --> 00:14:27.450
That's my final
question in this video.

00:14:27.450 --> 00:14:32.270
How is r of t
connected to g of t?

00:14:32.270 --> 00:14:35.420
Well, let me ask about
the right-hand sides.

00:14:35.420 --> 00:14:38.370
How is the step
function connected

00:14:38.370 --> 00:14:40.720
to the delta function?

00:14:40.720 --> 00:14:46.220
Answer, the step function is the
integral of the delta function.

00:14:46.220 --> 00:14:47.780
The integral of
the delta function

00:14:47.780 --> 00:14:50.230
is 0 as long as
you're integrating off

00:14:50.230 --> 00:14:51.850
where the delta function is 0.

00:14:51.850 --> 00:14:55.910
But as soon as you
pass the big spike,

00:14:55.910 --> 00:15:00.900
then the integral jumps to 1,
and you have a step function.

00:15:00.900 --> 00:15:04.850
So the step function is the
integral of the delta function.

00:15:04.850 --> 00:15:10.650
So the step response is the
integral of the delta response.

00:15:10.650 --> 00:15:13.390
It's the integral of
the impulse response.

00:15:13.390 --> 00:15:15.910
r is the integral of g.

00:15:15.910 --> 00:15:21.850
r is the integral of g with
the correct initial conditions

00:15:21.850 --> 00:15:26.980
that gave us this and eventually
[INAUDIBLE] approach 1.

00:15:26.980 --> 00:15:34.490
So that is the two key
solutions, you could say.

00:15:34.490 --> 00:15:39.780
The impulse response important
in theory and in practice.

00:15:39.780 --> 00:15:42.820
The step response extremely
important in practice

00:15:42.820 --> 00:15:47.725
because turning on
a switch is so basic

00:15:47.725 --> 00:15:53.930
an operation in engineering.

00:15:53.930 --> 00:15:55.220
Good.

00:15:55.220 --> 00:15:57.120
Thank you.