We ended the last segment with the Takens
theorem which concerns a delay coordinate

embedding of a time-series data set from a
dynamical system.

Today's task is to dig into one of the
really important pieces of this.

The words at the end of the second line
and the beginning of the third:

"diffeomorphic to" and "have the same 
topology as."

Now you remember the goal of this 
procedure. The goal is to undo a

projection. The procedure itself we 
covered in the previous section.

You plot delayed versions of the
measured quantity against itself.

If you have good data and you do the
embedding correctly, the results are

guaranteed to be topologically 
identical to the true dynamics.

The "good data" part of what I said is the
note at the bottom right on this slide.

The "do it right" part is the 
colored words.

We'll come back to all of that 
later in this unit.

Now since topology is not a prerequisite 
for this class, I'm going to do Topology

101 in two minutes.

Topology is the fundamental mathematics 
of shape, as I said last time.

What I mean by that: what the concept of 
shape becomes if you don't measure.

The word "geometry" has the word 
"metric" buried in it. A metric is a way

to measure something. The word "topology"
does not have that. It's not "topo-metry".

Size does not matter. Since size doesn't 
matter, all of these are the same thing.

When you're talking about the topology 
of an object, only the number of pieces,

or the number of holes matters.

The two objects at the top here, 
the coffee cup and the donut,

have the same topology because they both
have one hole and one piece.

Their geometry is very different however.

The two objects at the bottom have very
similar geometries.

The one on the left, which is a colander 
which you use to drain pasta however,

has lots and lots of holes whereas the 
bowl on the right doesn't have any holes.

So the two objects on the bottom have 
similar geometry and different topology.

The two objects on the top have the same 
topology and very different geometry.

Now think about a donut and a coffee mug 
made out of clay or dough.

You could deform one into the other with-
out destroying or creating pieces or

destroying or creating holes.

If you can do that, the two objects 
have the same topology.

And for those who know this object, 
obviously I'm referring to Betti numbers.

The mathematical form of transformations 
like that deforming of the dough that

don't make or break pieces or holes, 
is called a diffeomorphism.

A diffeomorphism is one to one, onto, 
invertible, and differentiable

in both directions.

And a correct embedding is related to the 
true dynamics by such a transformation

if the conditions of the 
theorem are met.

And what that means, is that the 
reconstructed dynamics here on the left

have the same topology as the true 
dynamics on the right.

The real and reconstructed attractors 
don't look alike to us because our eyes

respond to geometry, not topology. But 
they're identical in some very formal and

powerful mathematical ways. 
So why is all this useful?

You may remember my definition of 
bifurcation as "topological changes in

the attractor", for instance.
Now you know what that means.

Topology really matters. 
Many of the important properties of

dynamical systems, such as the Lyapunov 
exponent, are invariant under those

transformations, the diffeomorphisms, 
that preserve topology. And all of that

means, that you can measure one thing from
a very complex system, do the embedding,

compute the value of one of these 
dynamical invariants, and assert that your

answer holds for the true unobserved 
dynamics inside the black box.

Which is pretty darned amazing. 
Now taken to an extreme that means that I

could take a thermometer, stick it outside
my window, measure a time series of the

temperatures outside my window and from 
that time series I could reconstruct the

dynamics of weather of the Western
hemisphere. Now that does not work for a

couple of very important reasons having 
to do with how you would actually have

to do the embedding to get that to work
and how much data you would need

to get that to work properly.

And that is what we'll talk 
about in the next segment.