- Hello, everyone. In these next few videos,

I'm going to be talking about something called,

the Jacobian,

and more specifically, it's the Jacobian matrix,

or sometimes the associated determinant,

and here, I just want to talk

about some of the background knowledge that I'm assuming,

because to understand the Jacobian,

you do have to have a little bit

of a background in linear algebra, and in particular,

I want to make sure that everyone here understands

how to think about matrices as transformations of space,

When I say transformations,

here, let me just get kind of a matrix on here.

I'll call it two one and negative three one.

You'll see why I'm coloring it like this in just a moment.

When I say, how to think about this

as a transformation of space,

I mean, you can multiply a matrix

by some kind of two-dimensional vector,

some kind of two-dimensional x y,

and this is going to give us a new two-dimensional vector.

This is going to bring us to, let's see in this case,

it'll be,

I'll write kind of two one negative three one,

where what it gives us

is two x plus negative three times y,

and then, one x plus one times y.

Right? This is a new two-dimensional vector

somewhere else in space,

and even if you know how to compute it,

there's still room for a deeper geometric understanding

of what it actually means to take a vector x y

to the vector two x plus negative three y

and one x plus one y.

There's also still a deeper understanding

in what we mean when we call this a linear transformation,

a linear transformation.

What I'm going to do is just show you

what this particular transformation looks like

on the left here, where every single point

on this blue grid, I'm going to tell the computer,

"Hey, if that point was x y, I want you to take it

to two x plus negative three y, one x plus one y.

Here's what it looks like.

Let me just kind of play it out here.

All of the points in space move,

and you end up in some final state here.

There are a couple important things to note.

First of all, all of the grid lines

remain parallel and evenly spaced,

and they're still lines.

They didn't get curved in some way,

and that's very, very special.

That is the geometric way that you can think

about this term, this idea of a linear transformation.

I kind of like to think about it that lines stay lines,

and in particular the grid lines here,

the ones that started off

as kind of vertical and horizontal,

they still remain parallel,

and they still remain evenly spaced.

The other thing to notice here is

I have these two vectors highlighted:

the green vector and the red vector.

These are the ones that started off,

if we kind of back things up,

these are the ones that started off

as the basis vectors, right?

Let me kind of make a little bit more room here.

The green vector is one zero,

one in the x-direction, zero in the y-direction,

and then that red vertical vector here,

is zero one,

zero one.

If we notice where they land,

under this transformation,

when the matrix is multiplied

by every single vector in space,

the place where the green vector lands,

the one that started off as one zero,

has coordinates two one,

and that corresponds very directly with the fact

that the first column of our matrix is two one.

Then, similarly, over here, the second vector,

the one that started off as zero one,

ends up at the coordinates negative three one,

and that's what corresponds with the fact

that the next column is negative three one.

It's actually relatively simple to see

why that's going to be true.

Here, I'll go ahead and multiply this matrix

that we had that was ...

See now it's kind of easy to remember

what the matrix is, right?

I can just kind of read it off here

as two one negative three one,

but just to see why it's actually taking the basis vectors

to the columns like this,

when you do the multiplication by one zero,

notice how it's going to take us to,

so it's two times one, that'll be two,

and then, negative three times zero,

so that'll just be zero,

and over here, it's one times one,

so that's one, and then, one times zero,

so again we're adding zero.

The only terms that actually matter

because of the zero down here,

was everything in that first column.

Similarly, if we take that same matrix,

two one negative three one,

and we multiply it by zero one over here,

by the second basis vector,

what you're going to get,

is two times zero, so zero,

plus that element in that second column,

and then, one times zero, so another zero,

plus one times one, plus that one.

Again, it's kind of like that zero knocks out

all of the terms in other columns.

Then, like I said, geometrically,

the meaning of a linear transformation

is that grid lines remain parallel and evenly spaced.

When you start to think about it a little bit,

if you can know where the screen vector lands

and where the spread vector lands,

that's going to lock into place

where the entire grid has to go.

Let me show you what I mean and how this corresponds with,

maybe, a different definition that you've heard

for what linear transformation means.

If we have some kind of function L,

and it's going to take in a vector and spit out a vector,

it's said to be linear if it satisfies the property

that when you take a constant times a vector,

what it produces is that same constant times

whatever would have happened if you applied

that transformation to the vector, not scaled, right,

so here you're applying that transformation

to a scaled vector,

and evidently, that's the same

as scaling the transformation of the vector.

Similarly, second property of linearity

is that if you add two vectors,

it doesn't really matter if you add them

before or after the transformation.

If you take the sum of the vectors

then apply the transformation,

that's the same as first applying the transformation

to each one separately,

and then adding up the results.

One of the most important consequences

of this formal definition of linearity,

is that it means if you take your function

and apply it to some vector x y,

I can split up that vector

as x times the first basis vector,

x times one zero

plus y,

let's see y, times that second basis vector,

zero one,

and because of these two properties of linearity,

if I can split it up like this,

it doesn't matter if I do the scaling and adding

before the transformation,

or if I do that scaling and adding

after the transformation,

and say that it's x times whatever the transformed version

of one zero is,

and I'll show you geometrically what this means

in just a moment, but I kind of want to get

all the algebra on the screen,

plus y times the transformed version

of zero one,

zero one.

To be concrete, let's actually put in

a value for x and y here,

and try to think about that specific vector geometrically.

Maybe I'll put in something like vector two one.

If we look over on the grid,

we're going to be focusing on the point that's over here

at two one, and this particular point.

I'm going to play the transformation,

and I want you to follow this point to see where it lands,

and it's going to end up over here.

Okay, so, in terms of the old grid, right,

the original one that we started with,

it's now at the point one three.

This is where we've ended up,

but importantly, I want you to notice how it's still

two times that green vector plus one times that red vector.

It's satisfying that property that it's still

x times whatever the transformed version

of that first basis vector is,

plus y times the transformed version

of that second basis vector.

That's all just a little overview,

and the upshot,

the main thing I want you to remember from all of this

is when you have some kind of matrix,

you can think of it as a transformation of space

that keeps grid lines parallel and evenly spaced.

That's a very special kind of transformation.

That is a very restrictive property to have

on a function from 2-D points to other 2-D points.

The convenient way to encode that,

is that the landing spot for that first basis vector,

the one that started off one unit to the right,

is represented with the first column of the matrix,

and the landing spot for the second basis vector,

the one that was pointing one unit up,

is encoded with that second column.

If this feels totally unfamiliar,

or you want to learn more about this,

it's something that I've made other videos on in the past,

but in terms of understanding the Jacobian matrix,

where we're going with this,

and kind of getting a geometric feel for it,

that short overview that I gave

should be enough to get us going.

With that, I will see you next video.