video_id
stringclasses
7 values
text
stringlengths
2
29.3k
5DUQ3-Y_gX4
ALAN EDELMAN: I call it the way for big boys and big girls.
5DUQ3-Y_gX4
STEVEN G. JOHNSON: Yes, for big kids.
5DUQ3-Y_gX4
ALAN EDELMAN: For grown-ups.
5DUQ3-Y_gX4
STEVEN G. JOHNSON: The big-kid way.
5DUQ3-Y_gX4
ALAN EDELMAN: Kid way.
5DUQ3-Y_gX4
STEVEN G. JOHNSON: OK.
5DUQ3-Y_gX4
So df, let's just do it slowly.
5DUQ3-Y_gX4
So what we want to do is we want to take f.
5DUQ3-Y_gX4
We're going to take--
5DUQ3-Y_gX4
I'm going to draw my vector symbols here.
5DUQ3-Y_gX4
I guess I'll put them here.
5DUQ3-Y_gX4
But I get tired of writing them all the time.
5DUQ3-Y_gX4
All my x's, and therefore my dx's, are vectors.
5DUQ3-Y_gX4
So think of it as an arbitrary small change
5DUQ3-Y_gX4
in an arbitrary direction.
5DUQ3-Y_gX4
We want it to be able to handle anything like that.
5DUQ3-Y_gX4
ALAN EDELMAN: And in case it wasn't
5DUQ3-Y_gX4
already obvious to everybody, what is the output of f?
5DUQ3-Y_gX4
Is it a scalar, a vector, a matrix?
5DUQ3-Y_gX4
It's a scalar, exactly.
5DUQ3-Y_gX4
Just wanted to make sure everybody realized--
5DUQ3-Y_gX4
STEVEN G. JOHNSON: Sorry, yes.
5DUQ3-Y_gX4
ALAN EDELMAN: --that this is a scalar function of a vector.
5DUQ3-Y_gX4
STEVEN G. JOHNSON: Yeah.
5DUQ3-Y_gX4
You could also write this as x dot product with ax.
5DUQ3-Y_gX4
That's the same thing.
5DUQ3-Y_gX4
OK.
5DUQ3-Y_gX4
So I'm just going to do this out.
5DUQ3-Y_gX4
I think it's still a little bit laboriously,
5DUQ3-Y_gX4
but we'll have a better rule for-- we'll do the product
5DUQ3-Y_gX4
rule in a minute.
5DUQ3-Y_gX4
But let's do it without the benefit of that.
5DUQ3-Y_gX4
ALAN EDELMAN: Because you're effectively
5DUQ3-Y_gX4
deriving the product rule in what's about to come.
5DUQ3-Y_gX4
STEVEN G. JOHNSON: Exactly, yes.
5DUQ3-Y_gX4
So what do we do?
5DUQ3-Y_gX4
So I'm going to plug--
5DUQ3-Y_gX4
I'm going to take f of x plus dx.
5DUQ3-Y_gX4
I'm going to subtract fx.
5DUQ3-Y_gX4
And I'm going to drop--
5DUQ3-Y_gX4
because it's d's, I'm going to drop anything
5DUQ3-Y_gX4
that looks like a d squared, a dx squared, something
5DUQ3-Y_gX4
that goes to 0 faster than dx.
5DUQ3-Y_gX4
So what's f of x plus dx?
5DUQ3-Y_gX4
Well, I take x.
5DUQ3-Y_gX4
I add dx, and I plug it into f.
5DUQ3-Y_gX4
So there's a transpose there.
5DUQ3-Y_gX4
There's an A. There's an x plus dx there.
5DUQ3-Y_gX4
And then I subtract x transpose Ax.
5DUQ3-Y_gX4
And then I can just multiply everything out.
5DUQ3-Y_gX4
Just think of dx as a really small vector.
5DUQ3-Y_gX4
And just use your ordinary rules from linear algebra.
5DUQ3-Y_gX4
I can just use the distributed whatever it's
5DUQ3-Y_gX4
called, the distributive rule.
5DUQ3-Y_gX4
I just have to make sure I don't change the orders of anything
5DUQ3-Y_gX4
since these are vectors and matrices.
5DUQ3-Y_gX4
So I can multiply.
5DUQ3-Y_gX4
There's this term times this term times this term.
5DUQ3-Y_gX4
That's the first term.
5DUQ3-Y_gX4
So there's an x transpose Ax.
5DUQ3-Y_gX4
There's also this term times this term times this term.
5DUQ3-Y_gX4
So that's a dx transpose Ax.
5DUQ3-Y_gX4
So dx is just a little vector.
5DUQ3-Y_gX4
It's perfectly fine to transpose it.
5DUQ3-Y_gX4
And then I also have this term times this term
5DUQ3-Y_gX4
times this term.
5DUQ3-Y_gX4
So that's x transpose Adx.
5DUQ3-Y_gX4
And then I have this term times this term times this term.
5DUQ3-Y_gX4
But that has a dx squared.
5DUQ3-Y_gX4
And so that I'm going to just--
5DUQ3-Y_gX4
let's see.
5DUQ3-Y_gX4
Well, I actually am going to [INAUDIBLE] here.
5DUQ3-Y_gX4
So that term is your dx transpose Adx.
5DUQ3-Y_gX4
This term is gone.
5DUQ3-Y_gX4
This is high order.
5DUQ3-Y_gX4
So in the limit as dx gets smaller and smaller
5DUQ3-Y_gX4
and smaller, this term is negligible
5DUQ3-Y_gX4
compared to these terms.
5DUQ3-Y_gX4
And then I still have this term over here, can't forget that.
5DUQ3-Y_gX4
Otherwise, these terms are negligible compared to this.
5DUQ3-Y_gX4
But it's OK because I'm going to subtract that.
5DUQ3-Y_gX4
And that term cancels.
5DUQ3-Y_gX4
And what's left is these two terms.
5DUQ3-Y_gX4
And this is a perfectly good linear operation on dx,
5DUQ3-Y_gX4
but it's not written in a very nice form.
5DUQ3-Y_gX4
So the trick is to-- since these are numbers,
5DUQ3-Y_gX4
I can transpose them.
5DUQ3-Y_gX4
And this is something I feel like in 18.06 people
5DUQ3-Y_gX4
get very confused by, that normally you're
5DUQ3-Y_gX4
not allowed to change the order of anything.
5DUQ3-Y_gX4
Normally, a matrix is not equal to its transpose.
5DUQ3-Y_gX4
But a number is always equal to its transpose.
5DUQ3-Y_gX4
So this is since--
5DUQ3-Y_gX4
let me do a note over here.
5DUQ3-Y_gX4
Since dx transpose Ax is a scalar,
5DUQ3-Y_gX4
a scalar is always equal to its transpose.
5DUQ3-Y_gX4
So then we can take this and set it
5DUQ3-Y_gX4
equal to its own transpose, which is the same thing as x
5DUQ3-Y_gX4
transpose A transpose dx.
5DUQ3-Y_gX4
And again, make sure you understand that.