Nonlinear Dynamical Systems

Prof. Madhu N. Belur and Prof. Harish K. Pillai

Department of Electrical Engineering
Indian Institute of Technology, Bombay

Lecture - 35
Dynamical System on Manifolds-2

Welcome everyone to this next lecture on tangent spaces and manifolds. So, we had just
begun with the definition of tangent space and a manifold. We also said what is a tangent
bundle, and it helps to know little about some more examples of manifolds.

(Refer Slide Time: 00:33)

So, we saw some standard terms S^1, we said is a circle, circle meaning we just mean x
square plus y square minus 1 equal to 0, what is defined by this equation. S^2 was our
standard sphere which we prefer calling x square x_1 square plus x_2 square plus x_3
square minus 1 equal to 0. Similarly, we can have S^n, the sphere n dimensional m n
dimensional sphere embedded in R^(n + 1), R stands for the set of real numbers, super
script n plus 1 means n plus 1 components, which we will say x_1 up to x_(n + 1). So
this we will say S^n. Another importance object is so called torus T^1, T^2, etc; so, these
are called torus.

So what is a torus? This torus is also what can be thought as a donut; T^1, T^2, T^n are
more generalizations. But this turns out, here in this case this turns out to be S^1 cross
S^1. Any point is denoted as what angle along this torus and then if you cut the torus at
any particular angle, then we get another circle. In that sense, the torus with one hole is
embeddable in R^3.

(Refer Slide Time: 02:41)

So, this is a subset of R^3, let me write this in little more detail. Torus with one hole,
which we should think of as like our donut, donut embeddable in R^3 and this turns out
to be S^1 cross S^1; and why is that, look at it like this. There is, this is like a ring, which
we can hold with by putting our hand inside this. So, this can also be thought of like a
ring, a ring whose cross section itself is a circle. So, one can tell at any point, any point
on this ring can be told at what is the angle with respect to such a frame let us say, and
once it is in that cross section, once we have identified the cross section using this angle,
one can say which point on this cross section by another angle. That cross section
suppose it is like this, then with respect to radially outward of the outer of the bigger of
this frame, one can say what is the angle with respect to this circle. In that sense one can
see that it is indeed S^1 cross S^1.

So, there are different ways, this particular topic requires a good amount of imagination
and also a good amount of rigor to be able to do this systematically and not just keep
imagining various objects. So, it is easy to see for this example that this is indeed one to
one correspondence with the set S^1 cross S^1. In other words, one angle theta 1,
another angle theta 2, these two together describe the precise point on the torus. We are
not including the interior of this particular donut, we are just speaking about the torus,
but this is the torus with one hole.

So, this is also some standard manifold that people deal with. This particular thing is
very relevant for example, in robotics where we have two angles; two angles, one at one
joint, another at another joint and one can ask that we are speaking of the evolution on a
torus and what kind of dynamical system, what kind of equilibria do we have. This
brings us to the next topic that, once we are given with a manifold, one can ask, what can
we say about the equilibrium, what can we say about the singular points?

So, given a dynamical system f of, x dot is equal to f(x) in which x of t evolves on a
manifold. So, unless we specify otherwise, in this course we are going to be dealing with
only smooth manifolds, what we defined as regular manifolds. So, on this manifold, we
have this dynamics and if it turns out that at point a on the manifold M, if f of a equal to
0, then we will say this a is an equilibrium point. It is an equilibrium point, why because
if f of a is equal to 0, rate of change of x with respect to time 0.

So, x dot is 0 hence, x remains at that point, in that sense it is equilibrium. But if being
on, if being in equilibrium means, so that particular point a is also called a singularity,
its singular, there is some singularity, it does not mean that some particular matrix is
singular. Singularity just means that, something is different here and in this particular
case all the components of this function f are all 0, it is equal to 0 vector. Hence, the rate
of change of x at that particular point is 0, and in that sense it is in equilibrium point. One
can ask, what happens about a neighborhood. So, this is something that we have seen in
detail, so one would have to linearize at that particular point.
(Refer Slide Time: 06:51)

So, at an equilibrium point do nearby trajectories approach, approach which point

suppose that equilibrium point was called x_0, do nearby trajectories approach x_0. If all
nearby trajectories approach x_0, then we have decided to use the word stable. If all
nearby trajectories approach x_0 asymptotically, if they converge to x_1, we will call it
asymptotically stable. If they do not blow up, if they remain in that small neighborhood,
then we will call it just stable.

On the other hand, even if some, if some nearby trajectories go far, if some nearby initial
conditions this are things that we already saw in more detail when we were studying
Lyapunov stability. But if some nearby initial conditions go away, then we already called
it unstable.

So, these all studies are relevant only at an equilibrium point. Why is it relevant only at
an equilibrium point? Because let us consider R^2, so if this is a point which is not a
singularity, then that point itself is not going to remain there. As a function of time it is
going to evolve go further because the vector field at that point is not 0, it is nonzero,
hence it will move in that direction. So, nearby points also it is likely to be nonzero. If
this function f is continuous and if it is nonzero at a particular point, nearby it cannot
suddenly become 0. So, nearby also it will be nonzero.

So, all those points are going to also any way move. So, it does not ask do nearby
trajectories approach this point, this point itself is not even an equilibrium point. So, the
question about stability or instability automatically applies to only equilibrium points.
But consider this particular point where the trajectories are, let us say changing in a way
because so the neighborhood is indeed worth studying in detail because that point
happens to be in equilibrium point.

And one can ask that, can we stabilize this? If there were an input one can say that, we
will like to stabilize this. So stability is a question that we also apply only to equilibrium
points and it depends on the manifold. It is possible that certain manifolds, certain
manifolds allow no equilibrium points, certain others require at least one. And now,
speaking about a global property of this manifold, what is global about it? Not just
locally it is it turns out that certain manifolds force you to have some equilibrium point
at least. If you want the dynamical system f, x dot is equal to f(x), if the function f should
be continuous, then certain others require at least one. So, this is what we will see in little
more detail in this lecture.

(Refer Slide Time: 10:20)

So, let us take an example. So suppose we speak about R^2, the question is if somebody
tells us, can you draw a vector field R^2 in which there is no equilibrium point? Does
there exists f such that, x dot is equal to f(x), f is smooth, smooth meaning in this case it
is just continuous and differentiable. So more generally smooth word could also mean
infinitely often differential, any number of times the derivative exists... f is smooth and
no equilibrium point at all.
So, answer is yes. One can just make all one can say x dot is equal to 1 so that no point
is an equilibrium point. On the other hand, the same question answer is no, if you want
this f to be smooth, and x evolves on a sphere. On a sphere, it turns out on S^2 S^2
turns out the answer does not exist. So, we cannot have a vector field that is continuous
and there is no point where the vector field is 0. In other words, there is no equilibrium
point at all, such a situation cannot happen as this sphere S^2 is concerned, unless you let
this f to be discontinuous.

So, this turns out to be a very important result which we will see in more detail today;
that result is called Hairy ball theorem. So, this requires us to develop a little more
concept, but why I am trying to draw your attention to this is that, it is a property of
manifold, even the stability of the equilibrium point, the equilibrium point itself appears
to be of a very local nature. The fact that f has to be continuous and it has to eventually
cover the entire manifold, forces some properties on the manifolds itself. It requires
something on the manifold for existence of an f whether or not equilibrium points should
exist for that f

Let us ask about a circle. On the other hand, S^1 circle, is it possible to think of a
equilibrium point? Is it possible to think of a dynamical system in which there is no
equilibrium point? Yes, we can just have theta dot equal to 1 that continuously it is
rotating like this, and its going on rotating. So at no point there is an equilibrium point.
So, hairy ball theorem speaks about S^2, and it says that one is forced to have an
equilibrium point.

In fact if one requires only simple singularities, then one will in fact have at least two
singular points and this turns out to be a to a very celebrated and familiar result to all of
you which is that the number of faces, edges and vertices of any polyhedra satisfies a
relation.
(Refer Slide Time: 13:12)

So, let us comeback to classification of equilibrium points on a plane. On a plane we

already saw some examples: stable, unstable node. We saw a node, we saw a saddle
point, we also saw center, what where these? All trajectories are coming inwards, this
turn out to be the case, which turned out to be the case if the matrix A had eigenvalues
which were both real and negative; this was what we called stable node.

We saw another case where when both eigenvalues were real and positive, that time
they this one was an unstable node. On the other hand, a saddle point was where
eigenvalues are real, but one is positive one is negative. For example, this is an
eigenvector corresponding to positive eigenvalue, let us say, so everything goes away.
And if this is an eigenvector corresponding to negative eigenvalue, we are speaking of a
plane when we have linearized about the equilibrium point, and we are considering the
eigenvalues of the matrix, the linearization at that point. And at all other points, it is like
this.

So, we also saw center, we saw the situation where eigenvalues of A are complex, but
depending on whether the real part is positive or negative, we can have oscillations that
are coming inwards or going out. This was the situation where the eigenvalues are
complex, but on the imaginary axis.

So, the oscillations are neither coming in nor going out, that is as far as the linear system
is concerned. And in this case it turned out that the linearized system the real part is 0,
but the second order nonlinearity might cause that the oscillations come inward or go
outward. And that is why one cannot use the linearized systems conclusion about it being
a center for the origin nonlinear system also, for the equilibrium point of the nonlinear
system.

So, as I said for the case that eigenvalues of A are complex if the real part is nonzero, if
the real part is positive, then these oscillations are going out. On the other hand, if the
real part is negative, it comes inwards. So, these all we will like to classify as something
called the index of that vector field. So it will turn out that this is stable and unstable
node the index is 1. So, we are going to very soon define a notion of index of an
equilibrium point, which will turn out to be plus 1, plus 1 for stable node, unstable node.

For the saddle point it will turn out to be minus 1, we will verify it for a few examples.
For centers it will be plus 1 and for focus stable and unstable focus also it will turn out to
be plus 1; then only the saddle turns out to be a little special for which it will be minus 1.
So, we are going to see in more detail what is the meaning of index of a vector field is.
So, let us take a equilibrium point.

(Refer Slide Time: 16:41)

Suppose this is an equilibrium point and we have vectors all around, the vectors
themselves first we will consider an isolated equilibrium point, what is isolated about
it? There is some neighborhood within which this is the only equilibrium point. We are
able to find some circle, some curve such that that curve contains this equilibrium point
and that curve contains this equilibrium point in the interior and this is the only
equilibrium point inside it. If such a curve can be found, small enough, then we will call
that that equilibrium point is isolated in the sense that it is not sticking to any other
equilibrium point. In that sense it is isolated; at least some sufficiently small
neighborhood contains only this one.

Now, we will like that there is such a curve which does not contain any equilibrium
points on it. Choose suitable curve with no equilibrium points on curve, let us call this
curve so this curve is a closed curve, it is also a simple curve in the sense that we are
not going to allow this curve to have self-intersections. Starting point and end point are
the two only points that are common; no other intermediate points are repeated.

So, that is indeed the case and we can also give it an orientation, we can give it in either
clockwise or anti-clockwise; that is not the issue. Now, we will ask that, look as we go
along this curve, the way we can the vector field at every point on this curve has a
unique direction because there are no equilibrium points on the curve. If there are
equilibrium points on the curve, then the vector field has length 0 and hence it would
have no direction and that would cause the problem to us.

So, we chose a suitable curve which has no equilibrium points on the curve and we
choose that this curve has this equilibrium point only, and that is possible because this
equilibrium point is isolated. So, now this vector also goes through a rotation. Notice that
this vector is pointed like this, it is like this and as we go along that curve this vector is
also turning and if this vector if this curve is traverses in clockwise, it turns out that
this vector also has turned clockwise.

So, vector field on the curve has unique direction, has unique direction because the curve
has no equilibrium points on it; has unique direction as curve traverses one rotation, say
clockwise. As I said, the direction clockwise or anti-clockwise will not matter. Let us
choose this clockwise, then we can ask whether that particular vector also has rotated
how many number of times and whether it is a whether the orientation on that vector
also has remained the same or not.

So, if we have chosen the curve to be clockwise, has a vector rotated clockwise is the
first question and has it rotated say how many number of times once or more. So, index
is defined as plus 1, if rotation once in a same direction, plus 2 if it is rotation twice in
the same direction; minus 1 if it is rotation one, but in the opposite direction than the
curve. Then, the fact whether it is same or opposite is what decides whether sign and the
number of rotations of course is decided by how many times it has rotated. It remains the
question, do there exist equilibrium points where you have two rotations?

It seems unreasonable that the vector rotates in the opposite direction as we rotate along
the curve in a particular direction, let us say clockwise. So, most easy to think of is plus
1, where the vector field rotates in the same direction as a curve itself. So, notice that if
there is no equilibrium point inside, then the vector field does not rotate; any net rotation,
it might just change sings like this, but it may not complete a rotation.

So, zero is also possible. In fact if you have multiple equilibrium points, isolated
equilibrium points inside, then the curve will add all these indices and it will be an
algebraic sum, that turns out to be a extremely neat concept that all these indices of
equilibrium, isolated equilibrium points add up for a curve that contains all of these. So,
if the curve contains none, if it contains no equilibrium point, then the index of that curve
will also automatically be 0; that means, the vector field undergoes no net rotation. So,
let us see an example now, let us take for unstable equilibrium point, for unstable node
we have already verified.

(Refer Slide Time: 21:39)

Let us take a center and let us take that particular periodic orbit itself as a curve C. So,
we see that at every point the curve C, the vector field is tangential to the curve itself.
And hence, when the curve undergoes one rotation the vector also undergoes exactly one
rotation in same direction.

So, now notice that even if the even if we are considering sorry, this particular
arrow should have been like this. Even if the curve was chosen anti-clockwise positive
that does not change the directions of the vector field. So, vector field decided by f of x
dot is equal to f(x) is decided only by f. Curve chosen pretty arbitrarily, only important
constraint is that in order to decide the index of a equilibrium point one should chose a
curve that contains only this particular equilibrium point. And that curve should not have
any equilibrium point on it.

This curve should be, should contain precisely one equilibrium point inside it, precisely
the equilibrium point for which we are trying to find out the index. Moreover, this curve
should be a simple and closed curve. Except for this constraints, it is to be chosen
arbitrarily. One can ask it need not be the periodic orbit that is decided by because
this being a center, could be in fact the periodic orbit but in the opposite orientation also.
And still it turns out that the index of that equilibrium point is independent of which
curve has been chosen as long as it satisfies these conditions.

So, center one can verify that it will it is indeed index of a center equal to also plus1.
Now, let us verify a saddle point. Now, let us take a curve like this, let us orient it
positive, so this particular point is equilibrium point. So, we see that when we start at this
point on the curve, it is pointing upwards. So, this is how we should fill. So, now you can
check that as you go along this curve, in fact at this point it is tangential hence opposite
to the curve and this point it is inwards like this.

So, we see that as you go along that curve, this particular pen that this particular pen that
I was showing has rotated by one number, but in opposite direction. So, index of saddle
equal to minus 1. Why? because if you choose the curve clockwise, then you took the
vector field along this curve as we are traverse on a curve in clockwise direction, that
particular vector turned out to rotate by one number of times, but in the anti-clockwise
direction; that is why the index of the saddle point is equal to minus 1. So, now that
brings us to the question that, do there exist equilibrium points with index 2? So, this
particular question sure would have haunted the dynamical systems community for many
years and indeed there is, this was told to me by my teacher.
(Refer Slide Time: 25:19)

So, it requires some effort to construct one. One can check that this particular
equilibrium point of course, trajectories is for smooth vector field; unless we have non
Lipschitz properties, trajectories do not intersect. It is just that they are very close by and
they eventually separate like this. So, for such a vector field, we can draw a curve and
check that the vector rotates by 2 times when you go around this particular equilibrium
point once; so such a vector field has index equal to 2.

So, it turns out that this is a little a special side, this is a special vector field for it to have
such a index, and we will yet to see what is special about it, that it is not a simple
singularity. It is not simple, it is a singularity because it is a equilibrium point, what is
not simple about it, that we will see that the linearization if x dot is equal to f(x) of
this, then you take the derivative of f with respect to x, then it will be square matrix, then
you evaluate it at the equilibrium point, x equal to this particular equilibrium point, and
this matrix, eigenvalues is what decided everything.

So, you look at the determinant suppose you call this matrix A, square matrix, it is a 2
by 2 matrix. Determinant of A nonzero will decide that, that equilibrium point is simple.
What is the meaning that the determinant of A is 0? For the linearized system, if the
determinant of A is 0, it means that there is the equilibrium point is not isolated as far
as the linear system is concerned. The nonlinear system equilibrium point might be
isolated, but the linearization is suggesting that there is a continuum of equilibrium
points and that is indeed what happens for linear systems if the matrix A is singular.

So, if the determinant of A is nonzero, then the matrix A is what we will like to say as a
nonsingular matrix and for such A, for such a situation that isolated equilibrium point,
we will say is a simple singularity if the determinant of A is nonzero. If the determinant
of A is zero, then that equilibrium point even if it is isolated, we will say is not a simple
singularity. So, only with non-simple singularities one can have index more than 2, more
than 1.

So, index equal to 2 or more is possible only for non-simple singularities. One can check
that the saddle point, stable unstable focus they all have, they all are simple singularities
and hence the indexes are plus or minus 1 only. So, this brings us to the one of the last
topics of this, one of the last subtopic of this topic that is about the hairy ball theorem. So
one can ask now that suppose they are given with a sphere

(Refer Slide Time: 28:49)

Given a sphere, find f to have no singularity or if inevitable only simple singularities; so,
this is the question. Given a sphere, find a function f, what is this function f? Because we
are trying to construct dynamics like this see, if there are no singularities it means that
you can never stabilize the system at any point; there cannot be any equilibrium point
itself, that is the consequence if there is no singularity. If there are singularities, then we
will like that there are simple singularities because there linearized because we will
ideally like linear system, linearization and linear control to operate there.

So, if it is not a simple singularity, then we need more complex systems because the
eigenvalue at the origin suggests that at steady state there is a nonzero value, it is not
converging, but it is staying close by only. Asymptotic stability requires that all
equilibrium points are in the left half, linearization at every equilibrium point is in the
has eigenvalues in the left half complex plane. So, simple singularities are good in that
sense. So, it turns out that there is this person, not there is hairy ball theorem that tells
that no singularity is not possible, one requires at least two simple singularities. If you
allow the singularities to be non-simple, then one would suffice.

Hairy ball theorem says that, continuous combing of a hairy ball leaves at least one
singularity. What is combing of a hairy ball? Suppose we are given with the ball, and
suppose we are this ball has lot of hair along it. This hair, each hair is like our regular
hair. If this hair denotes a vector field at that point, at each point there is some hair that
starts at the origin of the of the tangent space at that point and it is, it defines the unique
direction and we are now asked that we are required to comb it in a continuous way.

So, this tells that the f has to also be continuous in x. So, what is combing about it?
Because we want that hairs are all in the tangent space, in that sense they are
tangential to the ball. They cannot be standing out, the hair cannot be sticking out like
this; they better get combed on that ball. So, combing means that particular vector is in
that tangent space. Continuous, the function f is continuous. Hairy ball meaning, the ball
has hairs and each of this hairs are nothing but vectors in that tangent space. Then,
tangents space is one that is getting forced because of this combing process.

Now, we will like that there is at every point there is one hair, that is nonzero length,
that is the meaning that there is no singularity. So, is it possible that we can continuously
comb without a singularity? So, it turns out that the answer is no. The hairy ball theorem
says that, there will at least be one singularity. This is the meaning that when we comb
there is a at least a point on the head, where all the hair are going away or going round
and round, these are what is well known when we comb the hair, comb our own heads
hair for example.
So, what is this one singularity? This one singularity is inevitable because the sphere, the
fact that we are given with a sphere. This sphere as a manifold it is forcing that the
manifold all the isolated singularities when we add the indices, we end up getting
number 2. That is the sum of all the indices of every equilibrium point, isolated
equilibrium point and we have defined index only for isolated equilibrium points.

(Refer Slide Time: 33:27)

So, let us now say, if simple, if simple singularities at least two are inevitable. At least
two required. So, this number 2 is extremely special. What is special about it? For
example, we can now think of a ball like this; let us say we have a north pole and a south
pole. So, we can think of vectors leaving form north pole and all converging towards the
south pole. This is an example of a vector field which is continuous, leaves a north pole
or comes towards the south pole.

So, two simple singularities, two singularities only, both are simple. Why? Because this
one, the north one is a unstable node while the south pole is a stable node. So they both
are simple singularities; they both have index 2, index 1 each. So, the total sum of all
indices for all equilibrium points is exactly 2. So, we can see that we can construct such a
case. Using that particular non-simple singularity, we can also think of this, on the sphere
and this goes all along like this.

So, one can think of a non-simple singularity or singularity itself of index 2 defined like
this on the sphere. So, now what is special about 2? So, that brings me to one of the
last very good relation with for a polyhedral, so for polyhedral the Eulers
theorem for polyhedra says that the number of faces minus number of edges plus number
of vertices is equal to 2. For any closed polyhedra with no holes, importantly with no
holes. So, for example, for a cube, so the faces, the number of faces is equal to 6. The
number of edges is equal to 4 on the top, 4 on the bottom and 4 vertical; that is 12. And
the number of vertices we have 4 on the top, 4 on the bottom; that is 8. So, we get that 6
minus 12, that is minus 4 6 minus 12, minus 6, plus 8; that is plus 2. So, we get 2.

So, what is the relation between this and singularities on a sphere? So, one can now think
of a sphere and we can try to sort of look at this having faces, edges in which we put this
points; all the edges are marked and the faces are like different regions on the sphere. So,
polyhedra we can think of is actually very similar to a sphere. In some topological sense
it is nothing but a sphere, with all this lines marked on the sphere. Now, inside each face
one can have, let us say for example, a stable node and at each node, at each vertex we
can have an unstable node, everything going away; and between two vertices it will turn
out that there will have to be a saddle point. Let me draw this figure little larger.

(Refer Slide Time: 37:18)

So, let us take this cube. So this is at the center of each face we decided to have a stable
node. At each vertex we will have a unstable node; at the center of each edge we will
have this saddle point. So, that sorry So, we see that at the center of every edge we
need to have a saddle point, if we have to be able to do this systematically, continuously
more precisely.

Sorry, this arrow, it should be inwards because all our vertices have been decided as
unstable nodes. So it is all going away from vertices; at if it goes away form every
vertex, then between two vertices there is exactly one edge, that center of that edge is
where both seem to be coming in. So, why do we make that stable as far as this edge is
concerned; but if we have the center of every face to also be stable node, then it will be
going away from this particular point. So fine, that way lets allow this to be a saddle
point.

So, notice that this is a systematic, continuous, consistent way of placing the saddle
points. How many saddle point would we have placed? E number, the number of edges.
How many stable nodes would we have put? F number of stable nodes. And, how many
unstable nodes would we have put? Number of vertices, V number of unstable nodes. So,
now we know that this we can add the indices for all of them. So, we know that
indices of saddle point is minus 1. So, the sum of overall equilibrium points will in fact
give us: F minus E plus V, because both stable and unstable nodes have plus 1 as their
indices. So, F and V both come with plus sign and E on the other hand because it
corresponds to the saddle point which has index minus 1. It corresponds to minus sign
here. So, this one is what we saw for polyhedra is equal to 2, and that is exactly what we
also saw for that north pole and south pole, there also it turn out to be 2.

So, what it says is that this sum being 2 is a property of the sphere, it is not a property of
whether you take a cube or a pyramid. The fact that every face you associate a stable
node and every vertex you can have an unstable node and then at the edge you are forced
to have a saddle point, and then sum of all vertices will sum of all indices of
equilibrium points will exactly turn out to be 2 by this particular formula. And the 2 is
like an invariant of this particular topological object called the sphere; and all this
polyhedra are in that sense homotopic to a sphere.

So, this brings us to the end of seeing how the hairy ball theorem says that it is inevitable
that for a sphere we cannot have a situation where for a continuous function f we do not
have any equilibrium points at all. We do not have any singularity, such a situation is not
possible.
One last question is somebody can ask: can you interchange the role of stable node and
unstable nodes? In other words, can you have a stable node at every vertex and unstable
node at every face? Yes, that is still possible; still at the edge you will require a saddle
point whereas with just some arrows reversed, but then the formula will still turn out
to be 2, that is because sum of all the indices of equilibrium points will still turn out to be
2, because we know that both stable and unstable node both have index plus 1.

So, this is one of the extremely important, extremely enchanting topics within nonlinear
dynamical systems about how it is related to Eulers polyhedra formula and through that
to something called algebraic topology. But then I do not work in this nor do I know
enough. I want just you to know about this; and tangent spaces on the other hand finds
wide applications in control when dealing with nonlinear systems, especially on
manifolds. With that we will end this lecture on tangent spaces and manifolds.

Thank you.