14 visualizações

Enviado por Ramesh Kadambi

Strogatz Non Linear Dynamics - Notes and Problem Solutions

Strogatz Non Linear Dynamics - Notes and Problem Solutions

© All Rights Reserved

- liyongmingIJMA1-4-2014
- 01338495
- Integral Equations -1
- Twu_GM_final_3232012
- Lyapunov Control
- General Theory of Elastic Stability_ Southwell
- Apiwat Etal IL03
- Control of Nonlinear Bilateral Teleoperation Systems Subject To
- C7e
- 03 Dynamical Systems Tutorial
- A Decomposition-Based Practical Approach to Transient Stability-Constrained Unit Commitment
- 22880_An Approximate Analytic Solution of the Blasius Problem
- International Journal for Numerical and Analytical Methods in Geomechanics Volume 17 issue 10 1993 [doi 10.1002%2Fnag.1610171002] Y. Sukirman; R. W. Lewis -- A finite element solution of a fully coupled implicit formulatio.pdf
- 30.18. IASS Paper Andres Harte N 37
- A Remark on Some Nonlinear Elliptic Problems
- Notes-PhasePlane
- Torque Control in Harmonic Drives with Nonlinear Dynamic
- CDM-2010-2010-0001-a005
- Synchronization & Anti-synchronization via Active Control of a chaotic Satellite: Nereid
- PDCass12_1

Você está na página 1de 48

Ramesh Kadambi

January 8, 2017

1

These are notes created from Strogatz and contain verbatim statements and figures from the book.

2

Chapter 1

1.1 Introduction

1. Nonlinear dynamics deals with systems of equations of the form,

x 1 = f1 (x1 , , xn )

x 2 = f2 (x1 , , xn )

x n = fn (x1 , , xn )

In contrast to linear systems, non-linear systems do not satisfy the condition of linearity.

2. Non-linear systems are divided into autonomous and non-autonomous systems. The autonomous systems do not explicitly

depend on time, non-autonomous systems do.

Example:

x = f (x) autonomous

x = f (x) + t non-autonomous

3. Geometric approach to analysing non-linear systems. Interpret the differential equation as a vector field.

Example:

x = sin(x)

dx

dt =

Zsinx

integrating both sides t = csc(x)dx = ln | csc(x) + cot(x)| + C

csc(x)+cot(x) |. The difficulty in interpreting the above

result is obvious. Despite being able to solve the system analytically, we are left with the following difficult unanswered

questions.

(a) Suppose x0 = 4 ; describe the qualitative features of the solution x(t) for all t > 0. In particular as t ?

(b) For an initial condition x0 , what is the behaviour of x(t) as t ?

The asymptotic behaviour of the system is unknown or difficult to answer despite the solution of the system.

4. The problem is visualized graphically with a plot of x vs x. We interpret x as the position of an imaginary particle moving

along the real line, and x as the velocity of the particle. The differential equation x = sin(x) represents a vector field on

the line; it is the velocity vector x at each x. The velocity vector is sketched by plotting x versus x and drawing arrows on the

x axis to indicate the corresponding velocity vector at each x. The arrows point to the right when x > 0 and to the left

when x < 0. It makes sense since a positive velocity increases x, a negative velocity decreases x.

3

4 CHAPTER 1. FLOWS ON THE LINE - STROGATZ

- 4

2

Figure 1.2: Plot of x versus t based on the graphical solution. This is only a qualitative indication and not a solution.

5. The points where x = 0 are called fixed points. The solid black dots are stable fixed points (also called attractors or sinks).

The unstable fixed points are the open circles (also knowns as repellers or sources).

6. Figure 1.1 shows that a partice starting at x0 = 4 moves to the right faster and faster until it crosses

2 (where sin(x)

reaches its peak). Then the particle decelerates to the stable fixed point at x = .

7. In general one can come to the conclusion that if x > 0 the particle moves in the direction of increasing x (to the right by

convention) and asymptotically approaches the nearest stable fixed point. Similarly if x < 0 the particle moves to a stable

fixed point to the left. If x = 0, then x remains constant.

1. The proposed graphical solution can be generalized to any first order ODE x = f (x). We just need to draw

the graph of f (x) and then use it to sketch the vector field on the real line. We imagine a fluid flowing

along the real line with a local velocity of f (x). The imaginary fluid is called phase fluid, and the real line

phase space. The flow is to the right where f(x) > 0 and to the left where f(x) < 0. A solution starting at

x0 is obtained by placing the imaginary particle (phase point) at x0 and watch how it is carried along the

flow. As time goes by the phase point follows a trajectory x(t) based on x0 , representing the solution to

the differential equation. The picture itself is called the phase portrait.

2. The appearance of the phase portrait is determined by the fixed points x defined by f (x ) = 0; these correspond to the

stagnation point of the flow.

3. Fixed points represent equillibrium solutions with respect to differential equation.

1.2. FIXED POINTS AND STABILITY 5

f (x) = x2 1

Sol Fixed points are determined by the equation f (x) = 0 x2 1 = 0 x = 1. The flow is to the right where x2 1 > 0

and to the left where x2 1 < 0.

Note that the definition of stable equillibrium is based on small disturbances; certain large disturbances

may fail to decay. In our example, all small disturbances around x = 1 will decay. However, any distur-

bance that takes the point beyond x = 1 will be repelled to . To lay emphasis on this fact we say the

point x = 1 is locally stable, but not globally stable.

4. The pictorial analysis does not provide quantitative insights. However, it does provide relevant and usefule qualitative

information about the non-linear system.

Ex Consider the following electrical circuit. A resistor R and a capacitor C in series with a batter of constant dc voltage Vo .

Suppose that the switch is closed at t = 0, and there is no charge on the capacitor initially. Let Q(t) denote the capacitor

at t 0. Sketch the graph of Q(t).

C

Vo

Sol The governing differential equation is as follows. Vo + RQ(t) + Q/C = 0 Q = Vo

Q

R RC

6 CHAPTER 1. FLOWS ON THE LINE - STROGATZ

Q

Q

Q

Q

t

The point Q = CVo is globally stable. No matter what the initial state the system returns to Q .

Ex Sketch the portrait corresponding to x = x cos(x), and determine the stability of all the fixed points.

Sol We will solve this graphically. The roots of x cos x are the points of interesection of the line y = x and y = cos x. We have

the following, x > 0 x cos x > 0 and similarly x < 0 x cos x < 0. We solve this equation graphically by plotting

y

y=x y = cos x

y = x and y = cos x.

1. A naive model of population growth is of the form N = rN (r is the growth rate and N is the population), results in

exponential population growth.

2. A more sophisticated model would decrease the growth rate beyond a certain level of population. The growth rate would

depend on the population. The equation would be of the form N = f (N )N , where r = f (N ).

N

3. Choosing r = 1 K, where K is the carrying capacity, we have the following equation for population growth.

N

N = N 1 , N 0

K

N

N 1

4. The fixed points of the equation are N = 0, N = K. 5

1.4. LINEAR STABILITY ANALYSIS 7

r = 2, No = 11, 6, K = 5

r

N = N (K N )

K

K

dN = rdt

N (K N )

Integrating both sides

Z Z

K

dN = rdt

N (K N )

Z Z Z

1 1

dN + dN = rdt

K N N

log(K N ) + log(N ) = rt + c

N

log( ) = rt + c

K N

N

= cert

K N

N (cert + 1) = Kcert

cKert

N=

(cert + 1)

K

N=

1 + cert

K

c= 1

No

1. We work around a valid fixed point of the system x = f (x). We consider a solution a small perturbation away from x .

We write the solution as (t) = x(t) x , we therefore have, = x = x = f (x) = f (x + ).

2. Assuming that f is sufficiently smooth, and using Taylor expansion. We have, f (x + ) = f (x ) + f 0 (x ) + O( 2 ). Since

x is a fixed point of f (x), we have f (x ) = 0 = f 0 (x ).

3. The equation (2) decays if f 0 (x ) < 0, grows exponentially if f 0 (x ) > 0. If f 0 (x ) = 0 the non-linear terms are significant

and we will need to perform non-linear analysis.

4. The magnitude of the slope f 0 (x ) determines how stable a given fixed point is. The magnitude plays the role of the growth

or decay rate. Its reciprocal is a characteristic time scale, it determines the time required for x(t) to variy significantly

in the neighborhood of x .

Ex Using linear stability analysis, determine the stability of the fixed points for x = sin x.

8 CHAPTER 1. FLOWS ON THE LINE - STROGATZ

Sol The fixed points of x = sin x are solutions to sin x = 0 x = k k {0, 1, }. Following in the footsteps of the linear

stability analysis, we look at

1 if k is odd

f 0 (x? ) = cos(k) =

1 if k is even

The stable points are k, k {1, 3, 5, } and unstable points are k, k {0, 2, 4, 6 }, which agrees with out earlier

graphical solution 1.1.

Ex Classify the fixed points of the logistic equation, using the linear stability analysis. Find the characteristic time scale in

each case.

N ). The fixed points are N = K, N = 0. f (N ) = r K , we have, f (0) =

0

0 1

r, f (K) = r. The point N = 0 is unstable and the point N = K is stable. The characteristic time scales, |f 0 (N )| = 1r .

Sol This is a case by case scenario. Let us look at, the following cases, (a) x = x3 (b) x = x3 (c) x = x2 and (d) x = 0.

x x

x3

x x

x3

x x

2

x

x x

(a) The point x = 0 is stable, globally at that, since x < 0, x > 0 and particles approach 0, x > 0, x < 0 and the

particles approach 0 again.

(b) The point x = 0 is unstable. In this case the opposite of (a) is true, x < 0, x < 0 and particles are repelled to ,

similarly, x > 0, x > 0 and particles are repelled to .

(c) The point x = 0 is semi-unstable, for points x < 0, x > 0 the particles move in the direction of the fixed point the

system is stable, however, for points x > 0, x > 0, the particles move away from the fixed point, it is unstable.

(d) Here every point x is a fixed point. The perturbations neither grow nor decay. They just stay there.

1.5. EXISTENCE AND UNIQUENESS 9

Remark 1. Note that all these stability analysis are based on small moves around the equilibrium position or fixed point.

The question of what is small is based on order of magnitude.

dx . The fixed points are points where the time derivative are

0

zero. The stable points are points where f (x ) < 0.

In this section we investigate the existence and uniqueness of solutions. We first see an example that establishes that solutions

need not be unique/exist.

1

Ex Show that the solution to x = x 3 starting from xo = 0 is not unique.

Sol This needs to be revisited. I do not quite understand the argument of uniqueness.

1.6 Potentials

The notion of potentials comes from the idea of potential energy. A particle is pictured sliding down the walls of a potential well,

where the potential V (x) is defined by

dV

f (x) =

dx

.

1. The negative sign in the definition of V follows the standard convention in physics; it implies that a particle always moves

down hill (from higher potential to lower).

dV 2

2. The change of potential w.r.t time is computed as dV (x(t)) = dV dx

= f (x) = dV dV

dt dx dt . Given x dx , we have dt = dx .

dV dV 2

3. The rate of change in potential dt = dx 0. This implies that V (t) decreases along trajectories.

dV dV

4. The potential is constant fixed points whenever dx = 0. This follows from the fact that x = f (x) = dx = 0 at fixed points.

6. Local maxima of V (x) correspond to unstable fixed points. Since the only trajectory possible from a local maxima is away

from it.

Ex Graph the potential of the system x = x, identify all the equllibrium points.

dx = x

dV

dx = x V (x) = 21 x2 + C. We choose arbitrarily C = 0. The only fixed point x is at 0

and is stable.

V (x) x

x2

x x

Ex Graph the potential for the system x = x x3 , and identify equillibrium points.

10 CHAPTER 1. FLOWS ON THE LINE - STROGATZ

x4 x2

Sol We have dV 3

dx = x x

dV

dx = x3 x V (x) = 4 2 + C. We again choose C = 0. The plots are shown below.

V (x) x

x x

x x3

There are 3 fixed points in all x {1, 0, 1}. The points x {1, 1} are stable. The point x = 0 is unstable. The

stable fixed points also form the local minima for V (x). The potential is often called the double well potential .

Chapter 2

Bifurcations - Strogatz

Qualitative aspects of a non-linear system can change as parameters change. Here we study the change in the nature of fixed

points when parameters change.

Definition 3. Bifurcation is a qualitative change in the system behaviour as a result of change in some parameter of the system.

1. Bifurcations provide models of transition and instabilities as some control parameter is varied.

The saddle point bifurcation is the basic mechanism by which fixed points are created and destroyed. The prototypical example

of a saddle node bifurcation is given by the first order system.

x = r + x2

.

1. The fixed poitns of the system are at x = r.

2. Clearly, for r < 0 we have two symmetric roots about the origin x = r.

3. As r increases and becomes positive the two roots approach each other and at r = 0 we have a semi-stable fixed point at

the origin.

4. As r becomes positive the fixed points move to the imaginary plane and disappear from the plot of x = f (x).

x x x r>0

r=0

r<0

x x x

5. A bifurcation diagram is a plot of the motion of fixed points with changing parameter value. Strogatz motivates the

bifurcation diagram very well. We will re-visit this another day to finish the notes.

6. In the biforcation diagram the stable node is represented by a solid line and the unstable by a dotted line.

11

12 CHAPTER 2. BIFURCATIONS - STROGATZ

7. The bifurcation diagram for the saddle node bifurcation is shown below.

x

unstable

stable

8. The saddle node bifurcation goes by many names, fold bifurcation, turning point bifurcation, blue sky bifurcation.

Ex Give linear stability analysis of the fixed points for the system x = r x2 .

Sol The fixed points of the system are x = r. The pictures are as given below. The analyisis steps are,

(a) The fixed points are x = 0 x2 r = 0 x = r.

(b) The stability of the fixed points implied by f 0 = d

dx (r x2 ) = 2x.

i. f 0 (x ) < 0 are stable points, f 0 (x ) < 0 2x < 0 x > 0

r is stable.

0 0

ii. f (x ) > 0 are unstable points, f (x ) > 0 2x > 0 x < 0 r is unstable.

iii. f 0 (x ) = 0 is the bifurcation point, f 0 (x ) = 0 2x = 0 x = 0 r = 0. This is the value of r where the

fixed points coalesce.

x

x x

x

x x

r>0

r=0 r<0

Ex Show that the first order system x = r x ex undergoes a saddle-node bifurcation as r is varied, and find the value of

r at the bifurcation point.

x = 0 r x ex = 0

x + ex = r

This is an equation that cannot be explicitly solved. We will plot the functions to see what the function looks like for

different values of r ie. r > 0, r < 0 and r = 0. We can then see the behaviour of the fixed points.

13

x

x

x

x

x

x

r>0

r=1

r<0

Clearly we see that there is a bifurcation point. In order to find it, we use a trick. The bifurcation point is the value

where the functions ex and r x are tangential to each other, ie. function values and their derivatives

are the same. We therefore have,

ex = r x (2.0.1)

x

e = 1 x = 0 r = 1 substituting the value of x into (2.0.1).

Remark 4 (Example Solution). Strogatz does the first part differently. He plots r x and ex and uses the intersection

points to figure out the stable points. I just plotted the function itself which is as easy to do.

Remark 5 (Normal Forms). In a certain sense the examples x = r x2 and x = r + x2 are representative of all saddle

node bifurcations. The idea is that close to the bifurcation pont, the dynamics typically look like x = r + x2 or x = r x2 .

Ex Consider the system x = r x ex . In order to investigate the system, we will linearize the system around its fixed

points as determined in 2.0.1 i.e. {0, 1}. We have,

1 1

ex = exo exo (x xo ) + exo (x xo )2 (1)n exo (x xo )n

2 n!

ex = 1 x + x2 + O(x)3

r x ex = r x 1 + x x2 + O(x)3

= r x2

the normal form for saddle node bifurcations.

To understand why saddle node bifurcations typically have this algebraic form. We just ask ourselves what should be

the local behavior for two roots/fixed points to coalesce and disappear.

i. The first observation is that there have to be atleast 2 fixed points.

ii. The function f (x, r) is locally parabolic (convex/concave).

x

f (x)

r > rc

r < rc

x

14 CHAPTER 2. BIFURCATIONS - STROGATZ

Expanding the function f (x, rc ) around the critical parameter value (x , rc ) we have,

x = f (x, r)

f f 1 2 f

= f (x , rc ) + (x x ) + (r rc ) + (x x ) +

x (x ,rc )

r (x ,rc ) 2

x2 (x ,rc )

f

We have f (x , rc ) = 0 by definition of a fixed point and x = 0 due to the tangency condition

(x ,rc )

of saddle point bifurcation. We therefore have,

x = a(r rc ) + b(x x )2 +

f 1 f

where a = x and b = 2 x2 .

(x ,rc ) (x ,rc )

It is clear from 8(a)ii that the dominant terms around the saddle point are of the form r x2 .

Transcritical bifurcations have a fixed point that exists for all values of a parameter and is never destroyed. However such a fixed

point may change its stability as the paramter changes value. The prototypical/normal form for a transcritical bifurcation is

x = rx x2

x

x x

. x

x x

r>0

r=0

r<0

r range x stability

r<0 0 stable

r unstable

r=0 0 bi-stable

r>0 0 unstable

r stable

3. As seen in the table, we see that the fixed point x = 0 changes its stability.

15

x stable

unstable

stable r

unstable

This bifurcation if commong to physical problems that have symmetry. In such problems fixed poits tend to appear and disappear

in pairs. There are different types of pitchfolr bifurcation, the simpler type is the supercritical bifurcation.

The normal form for supercritical pitchfork bifurcation is x = rx x3 .

1. The fixed points corresponding to x = 0 are, x {0, + r, r} with two roots at 0.

x

x x

x

x x

r>0

r=0

r<0

x

stable

stable r unstable

stable

16 CHAPTER 2. BIFURCATIONS - STROGATZ

In supercritical bifurcation the cubic term is stabilizing, it acts as a restoring force and pulls x(t) back toward x = 0.

If instead the cubic term were destabilizing, as in x = rx + x3 , we would have a subcritical pitchfork bifurcation. The

bifurcation diagram is shown below.

x x

stable

unstable

unstable

unstable

stable

4. The non-zero unstable bifurcation points x = r exist only below the bifurcation (r < 0) (motivates the term

subcritical).

5. The origin is stable for (r < 0) and is unstable for (r > 0).

6. As opposed to the supercritical pitchfork bifurcation the cubic term does not have a stabilizing effect for (r > 0). It aids

in driving the trajecteries to . This leads to the blow-up.

8. As with other bifurcations the supercritical bifurcation is sometimes called forward bifurcation, and subcritical bifurcation

is called inverted/backward bifurcation.

9. It may so happen that a higher order term may have a stabilizing effect. Consider x = rx + x3 x5 . The bifurcation plot

is shown above.

10. The bifurcation diagram displays hysterisis. The fixed point is unstable at r = 0, at which point a small push will cause

the system to jump to one of the stable wings. On the way down the wing stable points are stable to well below zero, so

the jump does not occur until r = rs .

11. Such layering of alternate stable vs unstable fixed points are fairly frequent occurrence.

12. The jump behavior does not usually happen with super critical bifurcation.

13. The origin is not a globally stable fixed point. It is only locally stable. It is stable to small perturbations.

The figure below shows FBD of a over damped bead on a rotating hoop. The equations of motion are as follows,

mr = mg sin b + m 2 cos

= r sin

mr = mg sin b + m 2 r sin cos

2.1. OVERDAMPED BEAD ON A ROTATING HOOP 17

m 2 cos

m 2

r

mg sin + b

mg cos + m 2 sin + m 2 r

m mg

We therefore have,

Since the system is assumed to be over damped. We can ignore the term mr.

= 0

2

m r sin cos mg sin = 0

g

sin = 0, cos = 2

r

1

Defining = 2 r

we have,

g

1 = n where n {0, 1, 2 }

1

cos 2 =

<1

=1

>1

cos

Graphical Solution to fixed points. Fixed points at < 1 on the hoop. Fixed points > 1 on the hoop.

2. The graphical solution shown in figure illustrates that < 1 the only fixed point is = 0 as the hoop starts spinning

faster and two fixed points appear on either side of 0 and move towards 2 . The bifurcation plot is given below.

18 CHAPTER 2. BIFURCATIONS - STROGATZ

The objective of dimensional analysis and scaling is to represent the physical system in non-dimensional terms. The process is

also used to find if the factors or parameters that were ignored is justified.

1. We will first look at non-dimensionalizing the time derivative. We therefore will use a normalized time variable = Tt , where

T is a characteristic time scale to be chosen later. The right choice of T will leave the derivatives , ,

dimensionless.

t

Rewriting the equations in terms of T we have,

d d 1 d

= =

d dt T d

d d 1 d d 1 1 d2

= + = 2 2

dt d T d dt T T d

substituting into your governing equation we have,

mr d2 b d

= mg sin + m 2 r sin cos

T 2 d 2 T d

We divide the equation by mg and we have,

r d2 b d 2 r

( ) = sin ( ) + ( ) sin cos

gT 2 d 2 mgT d g

r

All the terms in the parenthesis are dimensionless. In order for our assumption to hold we require that gT 2 << 1 and

b b b r m2 rg

mgT O(1). We choose T = mg , which will leave mgT = O(1). It is an interesting fact and as if magic gT 2 = b2 =

M 2 LLT 2 2

M 2 L2 T 4 T 2 is dimensionless b = MTLT

1 = M LT 11 and mg = M LT 2

.

Note the unit of T = secs (T ) since every coefficient is dimensionless, one can choose T to be any combination of parameters

from a given dimensionless group of which T is a part of.

No T The Dimensionless DE

b m2 rg 2 r

1 mg b2 + sin + g sin cos

2

1

+ g2 r sin + mr

b

sin cos

2

1

pg

3 n where n = r + mgr + sin n sin cos

b

Our justification for dropping is based on the following argument using the equation 1. If b2 >> m2 rg we could ignore

the second order term.

1 This N

is rotational damping coefficient, rad .

s

2.3. IMPROPER/IMPERFECT BIFURCATION 19

It is not true that one finds symmetry as we did earlier in pitchfork bifurcation scenarios. The normal form for improper

bifurcation is given by

x = h + rx x3

.

The fixed points are a tricky beast now. There are two parameters. We use our good old trick that x = 0 h = x3 rx. The

fixed points are points of intersection of h and x3 rx. The fixed points are intersections of horizontal lines h with the cubic

x3 rx.

Plot for values r < 0 Plot for values r > 0

x3 + 2x, r < 0

x3 2 x, r > 0

h=1

h < hc

x x

h > hc

h = 1

The point of bifurcation is where the line x = h and x3 rx are tangential to each other. Given that x = h is a constant, the

bifrucation points are the minima and maxima of x3 rx.

d 3

(x rx) = 3x2 r = 0 r = 3x2

dx r

r

xmax =

3

pr pr pr

= 3r r

3

pr

The critical value hc for bifurcation is the value of the function at xmax = 3. hc = ( 3) r 3 3 =

{ 23 r 3r , 23 r 3r }. Plotting hc vs r we have,

p p

hc x x x x

h=0 h 6= 0 r>0

r <= 0

r r r h 6= 0 h h

r>0

2. The bifurcations meet tangentially at (r, h) = (0, 0).

3. The bifurcation in co-dimension2 ie: there are 2 parameters.

4. Read Strogatz for the details.

20 CHAPTER 2. BIFURCATIONS - STROGATZ

The proposed model for the budworm population is given below,

N

N = RN 1 p(N )

K

1. The budworm population grows according a logistic equation with growth rate of R and carrying capacity of K.

2. After the population of the bud worms hits a particular level there is predation by birds.

2. As the population increases an reaches a threshold or critical level, the predation by birds rises.

3. As the population grows beyond a certain level, the predation saturates as the birds are eating as much as they can without

impact on the population growth.

BN 2

p(N ) =

A2

+ N2

BN 2

N

N = RN 1 2 (2.3.1)

K A + N2

The model (2.3.1) has 4 parameters, {R, K, A, B}. We would like to reduce the parameters via non-dimensionalizing (2.3.1).

We divide (2.3.1) by B and let x = N

A . Clearly N and A are of the same dimension (population levels).

N2

1 R N

N = N 1 2

B B K A + N2

N

Let x = N = Ax

A

dN dx

=A

dt dt

A dx RA A2 x2

= x x

B dt B BK 1 + x2

A . Clearly from the predation equation we have B is

A

the number of bud worms eaten per unit time. The dimension of B is units of time (the time required to reach the critical level

of predation). We now have,

B A

= tt=

A B

dx dx d

=

dt d dt

B dx

=

A d

RA K

choosing r = , k= and substituting into (2.3.1) we have,

B A

dx x x2

= rx(1 ) (2.3.2)

d k 1 + x2

2.3. IMPROPER/IMPERFECT BIFURCATION 21

dx

We have x = 0 d = 0. Applying to (2.3.2), we have:

x x2

rx(1 ) =0

k 1 + x2

x = 0 is always a fixed point. The remaining fixed points satisfy the equation,

x x

r 1 =

k 1 + x2

kx

r(k x) =

1 + x2

First we will investigate points where r = 0 and k = 0. r = 0 has only one root, x = 0. On the other hand of k = 0, the solution

is uninteresting, x = 0. Given r = 1 we have,

k

(1 + x2 )(k x) =

x

r

k

k + kx2 x x3 = x

r

3 2 k

x kx + (1 + )x k = 0 (2.3.3)

r

kx

The above does not lead us anywhere. We will resort to our conventional approach of graphing the two functions r(kx) and 1+x2 .

f (x)

x

22 CHAPTER 2. BIFURCATIONS - STROGATZ

Plot x vs x

x

0.5

0

12 3 4

x

0 5 10 15 20

Some observations,

1. The non-dimensionalization was clever to facilitate the graphical solution of the roots. One side of the solution is fixed

while the other side changes.

Point Stable?

origin (1) unstable

2. The stability of the points is clear from the plot of x vs x. 2 stable

3 unstable

4 stable

3. Depending on the slope for a given fixed value of k, there can be 1, 2 or 3 fixed points. It is clear from the cubic form of

the equation (2.3.3).

4. In the configuration of r and k with 3 fixed points (shown in figure), the behaviors are classified as below,

point description

1 unstable

2 refuge This is the point one would like to keep the bud worms for pest controls.

3 unstable point Threshold, if the budworm population rises above this level, it may jump to the outbreak leve.

4 outbreak level This is where the budworm population catastrophically increases.

5. Once k is fixed, the line r(1 xk ) rotates about k as r changes. The number of fixed points change decreasing from 3. The

fixed points coalesce in a saddle node bifurcation.

Figure illustrating saddle point bifurcation

f (x)

0.5

0 x

0 5 10 15 20

2.4. WHAT IS LARGE? 23

The bifurcation happens when the line r(1 xk ) is tangential to x

1+x2 . We there fore have,

r 1 1 x2

= 2 2

(1 + x2 2x2 ) =

k (1 + x ) (1 + x2 )2

k(1 x2 )

r=

(1 + x2 )2

2

k(1 x ) x x

2 2

(1 ) =

(1 + x ) k 1 + x2

2x3

k= 2 (2.3.4)

x 1

2x3

r= (2.3.5)

(1 + x2 )2

0.8

0.6

0.4

r

0.2

0

0 10 20 30 40

k

In most fields problems do not have exact solutions. Specialists in the field approximate the real-life system to facilitate approx-

imate solutions. The approximations may be purely numerical, purely analytical or a combination of the two.

The first step towards solving a problem is mathematical modeling. A means by which the chosen problem is posed in mathe-

matical terms. The process of mathematical modeling involves keeping certain elements, neglecting some and approximating yet

others. Dimensional analysis facilitates such approximations by doing the following

1. Reduce the number of parameters in the problem.

2. Bring to fore the relative importance of the parameters and their impact on system behavior.

24 CHAPTER 2. BIFURCATIONS - STROGATZ

Chapter 3

1. One dimensional phase spaces flow is extremely confined and all trajectories are forced to move monotonically or remain

constant.

2. The objective is to study linear systems and make connections to classification of non-linear systems.

Definition 6. A two dimensional linear system if a system of the form.

x = ax + by

y = cx + dy

2. x = 0 x = 0 if A is of full rank, so x = 0 is always a fixed point for any choice of A.

3. The solutions of x = Ax can be visualized as trajectories moving on the (x, y) plane (phase plane).

Ex 5.1.1 Consider a spring mass system given by, m

x + kx = 0.

4. Our objective is to develop methods for deducing behavior without actually solving the differential equations.

5. We can write it in matrix form as follows:

Let v = x

k

v 0 m v

= (3.1.1)

x 1 0 x

k

6. The system (3.1.1) assigns a vector (x, = (v, 2 x), where

v) m = 2 .

7. We will follow the same approach we did previously in solving one dimensional

systems. We will look at the vector fields.

v

As before the direction of the vector field depends on the sign of . Positive values increase the value of the function

x

and negative values decrease.

8. In the phase space (v, x), the direction is the vector sum when (v, x)

are not zero. On the x-axis, we have (x, v)

=

(0, 2 (x)) v < 0, the arrow is pointing downwards (negative v-direction) for +ve, x and pointing upwards for ve, x

(positive v-direction). Similarly on the y-axis we have, x = 0 and (x,

v)

= (v, 0), and as before the vectors point to the

right/left (+ve x, -ve x) depending on the sign of v = x.

9. At any other location the direction of the vector is such that the (x, v) components of the vector are (x, = (v, 2 x).

v)

This is equivalent to a fluid particle flowing with a local velocity (x, = (v, 3 x) starting with an initial velocity vector

v)

(xo , vo ).

10. The imaginary phase point starts at (xo , vo ) and flows along the vector field.

25

26 CHAPTER 3. LINEAR SYSTEMS - STROGATZ

11. The flow shown in figure 3.1a swirls around the origin.

12. The origin is special, a phase point placed at there would remain motionless since (x,

v)

= (0, 0). The origin is hence a

fixed point.

13. Phase points placed anywhere else will circulate around the origin. These trajectories are called closed orbites.

Vector field

1.0

1.0

0.5

0.5

disp

0.0

0.0

-0.5

-0.5

-1.0

-1.0

-1.0 -0.5 0.0 0.5 1.0 -1.0 -0.5 0.0 0.5 1.0

vel

(a) Vector field for spring mass (b) Orbits for spring mass

15. The fixed point (0, 0) corresponds to static equilibrium of the spring mass. The orbits correspond to the oscillatory motion

of the spring mass. The point on the orbit where the displacement is maximum the velocity is zero and vice versa. The

shapes of the orbits satisfy the equation 2 x2 + v 2 = C where C 0 is a constant.

a 0

Ex: 5.1.2 Solve the linear system . Graph the phase portrait as a varies from to , showing qualitatively different

0 1

cases.

x a 0 x

=

y 0 1 y

x = ax, y = y

x = x0 eat y = yo et

3.2. LANGUAGE OF STABILITY 27

16. The figure illustrates the different phase diagrams for varying a.

a Phase plot Behaviour

< 1 For a < 1 both solutions are stable and (0, 0) is a stable fixed point. Both (x, y) decay to zero with

time no matter the initial condition. x decays faster than y and solutions approach the fixed point along

y axis. The slower of the two decays decides the approach to the fixed point.

= 1 The solution are both stable with the origin as a stable fixed point. Both (x, y) = (xo et , yo et ) decay

at the same rate no matter where they start. The phase plot is a system of straight lines whose slope is

determined by the initial condition ( xy = xy00 ).

1 < a < 0 Both solutions are stable with fixed point at origin (0, 0). y decays faster than x. The solutions approach

the fixed point along x axis.

=0 Here x = xo is a constant solution and becomes the fixed point. The coordinates (x, y) = (xo , yo et )

(xo , 0) as t .

a>0 The solution for x is unstable and exponentially increases. The coordinates (x, y) = (xo eat , yo et )

(, 0) as t . A strange effect is when y0 starts on the y axis, the solution (x, y) = (0, y0 et ). The

solution decays along the y-axis to the saddle point x = 0. Similarly if the solution starts on x-axis, it just

goes along the x-axis to . The y-axis is the stable manifold and x-axis is the unstable manifold

of the saddle point x .

Definition 7 (Stable Manifold). The stable manifold is defined as the set of initial conditions x0 such that x(t) x as

t .

Definition 8 (Unstable Manifold). The unstable manifold is defined as the set of initial conditions such that x(t) x

as t .

17. Trajectories : Typically trajectories of unstable solutions, asymptotically approach the unstable manifold as t and

the stable manifold as t . Similarly the trajectories of stable solutions, asymptotically approach the stable manifold

as t and the unstable manifold as t .

Definition 9 (Attracting Fixed Point). A fixed point x is an attracting fixed point if all trajectories that start near x approach

x as t . If all trajectories are attracted to the fixed point x it is called globally attracting. Mathematically for each

there exists a time t > 0 such that x(t) B(x , ).

28 CHAPTER 3. LINEAR SYSTEMS - STROGATZ

Definition 10 (Liapunov Stable Fixed Point). A fixed point x is liapunov stable if all trajectories that start sufficiently close

to x remain close to it for all time. Mathematically this would mean if a trajectory starts inside an open ball remains inside

the ball, x(t0 ) B(x , r ) t x(t) B(x , r ).

Definition 11. A fixed point that is Liapunov stable but not attracting is neutrally stable.

Example: In figure 3.2(d) the fixed points are Liapunov stable but not attracting and hence neutrally stable. Nearby points

are neither attracted not repelled by neutrally stable fixed points.

Note that attracting fixed points and liapunov stability do no imply each other. They are orthogonal ideas.

stable (neutrally stable) non-attracting

unstable attracting

stable attracting / globally attracting

unstable non-attracting

Definition 12 (Stable Fixed Point). A fixed point is said to be stable if it is both Liapunov stable and attracting.

Definition 13 (Unstable Fixed Point). A fixed point is said to be unstable if it is neither attracting or Liapunov stable.

The objective of this section is to classify a systems described by a full rank coupled matrix A. Our objective is to classify all

possible phase portraits that occur. It was found in the examples there possibly exist manifolds tthat played a crucial geometric

role. These manifolds determined the direction of the trajectories as t . They determined the direction of the trajectories

as t . These manifolds also contained straight line trajectories that started and stayed in the manifold forever. The

trajectories exhibited exponential growth or decay.

x = Ax

x(t) = et v (3.3.1)

2. The objective is to determine the solutions corresponding to v 6= 0. If solutions exist they correspond to exponential

motion spanned by v.

x = vet

x = Ax vet = Aet v

cancelling the non-zero scalar factor et we have

Av = v

Av = Iv (A I)v = 0 (3.3.2)

3.3. CLASSIFICATION OF LINEAR SYSTEMS 29

The solution to the (3.3.2) is obtained by finding the eigen values of the matrix A given by the roots of the characteristic

equaiton |A I| = 0. For a 2x2 matrix A we have,

a b

A=

c d

The characteristic equation is given by

a b

=0

c d

2 (a + d) + ad bc = 0

2 + = 0

where

= trace(A) = a + d

= |A| = ad bc

2 4

1,2 = (3.3.3)

2

1. If (3.3.3) are two distinct values (1 6= 2 ), then the corresponding vectors v1 , v2 are linearly independent and span the

plane. Any initial value x0 = c1 v1 + c2 v2 .

2. The solutions

x(t) = c1 e1 t v1 + c2 e2 t v2

.

Figure 3.3: x0 as the sum of eigen vectors v1 , v2

(a) it is a linear combination of solutions to x = Ax.

(b) it satisfies the initial condition x0 = x(0) (Figure 3.3).

(c) the existence and uniqueness theorems imply it is the only solution.

Ex: 5.2.1 Solve the initial value problem x = x + y, y = 4x 2y, subject to the initial condition (x0 , y0 ) = (2, 3).

Sol: In matrix form we have,

x 1 1 x

=

y 4 2 y

We find the eigen values using the equation (3.3.3). We have

= 1, = 6

1 1 + 24

1,2 = = {3, 2}

2

30 CHAPTER 3. LINEAR SYSTEMS - STROGATZ

1+3 1 v1

= 0 0

4 2 + 3 v2

4 1 v1 1 1

= 0 0 v1 = similarly v2 =

4 1 v2 4 1

1 1 2t

x(t) = c1 e3t + c2 e

4 1

2 1 1

= c1 + c2 c1 = 1, c2 = 1. (3.3.4)

3 4 1

The equation (3.3.4) expresses the initial condition as the linear combination of the eigen vectors. In example 5.1.2, we

classified the system based on the parameter value a. Here we will look at characterizing the system using the eigen

values/vectors.

3.3. CLASSIFICATION OF LINEAR SYSTEMS 31

Eigen Values,

Example Phase Portrait Observations

Vectors

5.2.2 1,2 = {2, 3}, This is similar to solution of Example 5.1.2 where a > 0. The

2 4 > 0 difference is the direction vector of the stable and unstable

manifolds, corresponding to the two distinct real eigen val-

1

ues. As observed before we have the vector corresponding

1

to the +ve eigen value 2 = 2 is the unstable manifold, where

1

as the eigen vector is the stable manifold. The origin

4

is a saddle point since one solution is unstable while the

other is stable. As with all saddle points the trajectory

approaches the unstable manifold as t and the

stable manifold as t .

2 < 1 < 0,

5.2.3 This is the scenario where the origin is a stable fixed point.

2 4 > 0

Both solutions are negative and real. As previously observed

in example 5.1.2 case (a). The solution approaches the fixed

point tangentially along the slower eigen direction, spanned

by the eigenvector with the smaller ||. The eigen vectors are

not mutually perpendicular. In negative time t the

trajectories become parallel to the fast eigen-direction. The

unstable version is obtained by reversing thedirection

of the arrows and making the origin unstable fixed

point.

5.2.5 2 4 0 This is the case of two complex conjugate roots. The cases

are

(a) Two imaginary roots. In this case figure (a) the trajec-

tories are ellipses around a neutrally stable origin.

(b) Tow complex conjugate roots with stable and unstable

real roots. The phase portrait is a spiral figure(b). The

stable portrait spirals towards the origin, while the un-

stable portrait spirals out.

The case of equal roots. This is the case where we have two real roots.

(a) Suppose 1,2 = and there exist two independent vectors corresponding to the eigenvalue, then every vector spanned

by the eigenvectors is an eigenvector with this same eigenvalue .-

It can be shown that if there are two equal roots with two linearly-independent eigen vectors, then the matrix

32 CHAPTER 3. LINEAR SYSTEMS - STROGATZ

0

A= . If 6= 0 then all trajectories are straight lines through the origin x(t) = et x0 (since both solutions are

0

the same). This is the same situation as 5.1.2b. The direction of the trajectories (towards or away from origin) will

depend on 1,2 < 0 or 1,2 > 0.

b

(b) The case of equal roots but a single eigen vector solution. This happens when the matrix A = . It can be

0

clearly seen that there is only one eigen-vector.

b v1

(A I)v = 0 A = =0

0 v2

0 b v1

= 0 v1 = 0, v2 = c c R

0 0 v2

Figure: Degenerate node Phase Portrait Figure: Equal roots and distinct eigen vectors

When there is only one eigendirection the fixed point is a degenerate node. The phase portrait is shown in Figure.

As t the trajectories become parallel to the sole eigendirection. The degenerate node can be seen as a con-

tinuous transformation via rotation of two distinct eigendirections till they line up as one. Strogatz refers to this as

scissoring them together, in reality the degenerate node is a boundary between spiral and a node.

4. The figure shows the classification of Fixed points of linear system via (, )

Figure S: Regions of stability

3.3. CLASSIFICATION OF LINEAR SYSTEMS 33

Regions Behavior

<00< < 2 4 This implies that the system has two real roots of opposite signs. This would mean that

the fixed points are all saddle points.

>0

(a) 2 4 > 0 (a) We have two real roots since > 0, 2 4 > 0 > 0. The stability is determined

by the sign of . These are the regions marked as stable/unstable nodes.

(b) 2 4 < 0

(b) We have two complex conjugate roots. The stability is again determined by the sign

(c) 2 = 4

of . We see stable/unstable spirals.

(c) This is the border. We have two equal real roots. Degenerate nodes, stars live

here. The stability of the nodes are again determined by the sign of .

=0 This is the region of 0 eigen value (we have at least one zero eigen value). The origin is

not the only fixed point here. We either have a line of fixed points or an entire plane if

A = 0. This corresponds to the Figure 3.2d.

"The Figure S shows that saddle points, nodes and spirals are major types of fixed points; they occur in the large

open regions of (, ) plane. Centers, stars, degenerate nodes, and non-isolated fixed points are borderline cases that

occur along curves in the (, ) plane. Of these borderline cases, centers are by far the most important. Centers

occur very commonly in frictionless mechanical systems".1

saddle point <0 This has both stable and unstable paths leading from it. This is a result of real

solutions that have opposite signs.

node > 0, 2 4 > 0 A node is a result of real solutions with the same sign. The node is either stable

or unstable.

center =0 A center results from two purely imaginary conjugate roots.

spiral 6= 0, 2 4 < 0 A spiral is a result of complex conjugate roots.

star 2 4 = 0 A star is a result of two equal eigen values.

degenerate 2 4 = 0 A degenerate occurs when there is only one eigen-direction resulting from a single

eigen value.

non isolated =0 One of the eigen values is zero.

In this section we will look at a model of Love Affairs. Our first model is very simple. Quoting Strogatz, "Romeo is in love

with Juliet, but Juliet is a fickle lover. The more Romeo loves her, the more Juliet wants to run away from Romeo and hide.

But when Romeo gets discouraged and backs off, Juliet begins to find him strangely attractive. Romeo on the other hand tends

to echo her: he warms up when she loves him, and grows cold when she hates him". The system of equations is given below,

R = aJ

J = bR

where R(t) = Romeos love/hate for Juliet at time t, J(t) = Juliets love/hate for Romeo at time t, a, b are positive parameters.

Clearly,

R

0 a R

=

b 0 J J

1 Quoting from Strogatz.

34 CHAPTER 3. LINEAR SYSTEMS - STROGATZ

From our table we have = 0 and = ab > 0. We have a complex conjugate pair whose real part is zero. The trajectories are

essentially elliptical orbits around the neutrally stable fixed point at the origin (Figure 3.4). There is only one quadrant when

both feel attracted/love for each other.

a b

A more general lovers model is given by a full matrix A = , leading to the governing equations,

c d

R

a b R

=

J c d J

Ex: 5.3.1 What happens when two identically cautious lovers get together?

R

a b R

Sol The system is = . Where a is a measure of cautiousness and b is a measure of responsiveness. Our solution

J b a J

is as follows,

= 2a, = (a b)(a + b)

p

2a 4a2 4(a2 b2 )

1,2 = =ab

2

1

v1 =

1

1

v2 =

1

1 (ab)t 1

v(t) = c1 e + c2 e(a+b)t

1 1

Scenario {a, b, , } The love Effect

(a) This is the case where they are more responsive than cau-

tious. We therefore have an unstable fixed point at (0, 0).

The love affair just explodes either into mutual hate or mu-

a < 0, b > 0, a 6= b = 2a < 0 tual love. As seen all trajectories approach R = J, so their

(a) < 0, < 0 a2 < b2 feelings are mutual. This is a consequence of our model,

(b) < 0, > 0 a2 > b2 where the amount of like/dislike is the same for both ie a/b.

(b) Here they are more cautious than responsive. This is the

case of a stable fixed point of (0, 0). The love affair fizzles

out by mutual consent.

3.3. CLASSIFICATION OF LINEAR SYSTEMS 35

36 CHAPTER 3. LINEAR SYSTEMS - STROGATZ

Chapter 4

The study of non-linear systems begins in the earnest. The objective is to build on the linear system theory that has been

developed thus far. The study will progress as below,

Our objective is to look at the general form of a vector field on the phase plane given by,

x 1 = f1 (x1 , x2 )

x 2 = f2 (x1 , x2 )

where f1 , f2 are given functions. In vector notation we have x = f (x). Where, x = (x1 , x2 ) and f (x) = (f1 (x), f2 (x)).

2. The solution x(t) is the path of a phase point along the vector field, corresponding to a trajectory on the phase plane.

3. For nonlinear systems, there is typically no hope of finding trajectories analytically. Even with explicit formulas available,

they are often complex to provide insight.

(b) The closed orbits. These correspond to solutions for which x(t + T ) = x(t) t, for some T > 0.

(c) The arrangement of trajectories near fixed points and closed orbits.

(d) The stability and instability of fixed points and closed orbits.

The system of equations 4.0.1 can be solved by numerical integration. The standard method used is the Runge Kutta Method in

vectorized form. The integration is followed by plotting the velocity vectors at each point in the phase plane.

Ex 6.1.1 Consider the system x = x + ey , y = y. First use qualitative arguments to obtain information about the phase portrait.

Then, using a computer, plot the direction field. Finally use, the Runge Kutta method to compute several trjectories, and

plot them on the phase plane.

37

38 CHAPTER 4. PHASE PORTRAITS - STROGATZ

y = 0 y = 0 (4.0.1)

x = 0 x + ey = 0 x + 1 = 0 x = 1 (4.0.2)

t t

y = y y(t) = yo e y = yo e

lim y(t) = 0 lim x = x + 1

t t

From equations (4.0.1), (4.0.2) tells us that the fixed point is (1, 0). The paths can start in any of the 4 quadrants.

We first sketch the nullclines where either x = 0 or y = 0. This kind of follows from our linear system analysis. The

nullclines serve the same purpose as the eigen directions of our linear problem? The line y = 0 y = 0 so one of the

nullclines is the x-axis. Similarly if x = 0 x = ey y = log(x) We also have, x > 0 x + ey > 0 x >

3 2 1 1 2 3

1 x

1. The nullclines are analogous to eigen-directions of linear systems. Flows do not cross the nullclines when the system of

equations have a unique solution.

2. The nullclines partition the region into quadrants where the derivatives x,

y have different signs.

Theorem 14 (Existence and Uniqueness Theorem n dimensions). Consider the initial value problem x = f (x), x = x0 . Suppose

fi

f is continuous and all its partial derivatives xj

, i, j = {1, , n}, are continuous for x in some open connected set D Rn .

Then for x0 D, the initial value problem has a solution x(t) on some interval (, ) about t = 0 and the solution is unique.

There are some interesting consequences of the existence and uniqueness theorem (14).

2. Since the trajectories do not intersect the phase portrait has a clear structure to it.

3. In two dimensional phase spaces. In the presence of a closed orbit C in the phase plane any trajectory starting inside C

remains in C.

4. Trajectories starting within closed orbits will approach fixed points inside the orbit if any. If there are none then the

trajectory will approach the closed orbit.

4.2. FIXED POINTS AND LINEARIZATION 39

Theorem 15 (Poincare-Bendixson Theorem). For vector fields on a plane, if a trajectory is confined to a closed, bounded region

and here are no fixed points in the region, then the trajectory must eventually approach a closed orbit.

Consider the nonlinear system

x = f (x)

. Let (x1 , x2 ) be a fixed point of the system,

f1 (x1 , x2 ) = 0, f2 (x1 , x2 ) = 0

. Consider a small disturbance around the fixed point

u = x1 x1 , v = x2 x2

. Our objective is to answer the question will the disturbance decay or grow?. In order to answer the question we derive

the differential equations for u, v. We therefore have,

u = x1 since x1 is a constant

= f1 (x1 + u, x2 + v) by substitution

f1 f1

v + O(u2 , v 2 , uv) Multi dimensional Taylor series expansion.

= f1 (x1 , x2 ) + u+

x1

(x ,x ) x

2 (x ,x )

1 2 1 2

f1 f1

=u + v + O(u2 , v 2 , uv) since f1 (x1 , x2 ) = 0

x1 (x ,x ) x2 (x ,x )

1 2 1 2

similarly we have

v = x2

f2 f2

=u +v + O(u2 , v 2 , uv)

x1 (x ,x ) x2 (x ,x )

1 2 1 2

" f1 f1 #

u u

= x 1

f2

x2

f2 + O(u2 , v 2 , uv)

v v

x1 x2

(x1 ,x2 )

" #

f1 f1

x1 x2

The matrix f2 f2 is called the Jacobian.

x1 x2

40 CHAPTER 4. PHASE PORTRAITS - STROGATZ

It is tempting to just drop the non-linear terms and proceed with the linear analysis. However, the nonlinear terms can be

ignored as long as the fixed point for the linearized system is not one of the borderline cases. If the linearized system

predicts a saddle, node or a spiral, then the fixed point is really a saddle, a node, or a spiral of the original non-linear problem.

The borderline cases (centers, degenerate nodes, stars, or non-isolated fixed points) are much more delicate. These can be altered

by small non-linear terms.

Ex: 6.3.1 Find all the fixed points of the system x = x + x3 , y = 2y, and use linearization to classify them. Then check the

conclusion by deriving the phase portrait for the full nonlinear system.

fixed points:

x = 0 = x + x3 x = 0, x = 1

y = 0 = 2y y = 0

Pf = {(0, 0), (1, 0), (1, 0)}

1 + 3x2 0

J=

0 2

1 0

J(0,0) =

0 2

2 0

J(1,0) =

0 2

2 0

J(1,0) =

0 2

(b) (1, 0), (1, 0) are saddle points.

(b) y = 0 y = 0, the invariant horizontal line.

(c) The point (0, 0) attracting stable point, since x3 < x, 1 x 1 x > 0 x (1, 0), x < 0 x (0, 1) and

y < 0 y > 0, y > 0 y < 0 .

(d) The points (1, 0), (1, 0) are unstable since x3 > x x < 0 x < 1, x > 0 x > 1.

(e) The phase portrait is symmetric since x x and y y the equations are invariant.

v = x v = x = x x3 = v + v 3

u = y u = y = 2y = 2u

4.3. IS IT SAFE TO IGNORE THE NONLINEAR TERMS? 41

x = y + ax(x2 + y 2 )

y = x + ay(x2 + y 2 )

where a is a parameter. Show that the linearized system incorrectly predicts that the origin is a center for all values of a,

whereas in fact the origin is a stable spiral if a < 0 and unstable spiral if a > 0.

Sol: We proceed with the solution as usual, out steps start with the fixed points

x = 0 y + ax(x2 + y 2 ) = 0

y = 0 x + ay(x2 + y 2 ) = 0

It can be clearly seen that the origin (0, 0) is a valid solution. A thing to note about the original problem is that it is a

polynomial in x, y and hence resembles a Taylor series. We can just drop the higher order terms to linearize it. We now

compute the jacobian the usual way,

3ax2 + ay 2 1 + 2ay

J=

1 + 2ay ax2 + 3ay 2

0 1

J( 0, 0) =

1 0

the eigen values are

1 2

1 = 0 + 1 = 0 = i

A thing to note in the linearization process is that, we have lost the parameter a in the process. We now switch to polar

co-ordinates to continue the analysis,

r 2 = x2 + y 2

x = r cos

y = r sin

x = r cos r sin = r sin + ar3 cos

y = r sin + r cos = r cos + ar3 sin

rr = xx + y y = r cos (r sin + ar3 cos ) + r sin (r cos + ar3 sin ) = ar4

r = ar3

xy y x

similarly we can solve for = =1

r2

42 CHAPTER 4. PHASE PORTRAITS - STROGATZ

We now analyze the stability of the fixed points in the polar coordinates. Clearly from 4.3 we have r < 0 a < 0, r >

0 a > 0, we have all trajectories rotating about the origin at the same angular velocity, but for a < 0 it is a stable spiral

and a > 0 it is an unstable spiral, and a center for a = 0.

The problem illustrates why centers are delicate. All trajectories are required to close perfectly after one cycle. The

slightest miss converts the center into a spiral.

Similarly stars and degenerate nodes can be altered by nonlinearities, but unlike centers their stability does not change. A stable

star may be changed into a stable spiral but not an unstable spiral. The figure ?? shows why this is the case. Centers are on the

border of stable and unstable spirals. However, stars and degenerate nodes are at the boundary of stable/unstable nodes and

stable/unstable spirals

If our interest was only in stability we can ignore the the degenerate nodes as their stability does not change. We can classify

the nonlinear system as follows,

4.4. HYPERBOLIC FIXED POINTS,TOPOLOGICAL EQUIVALENCE AND STRUCTURAL STABILITY 43

Robust Cases

Repellers Both eigen values have positive real part.

Saddles The eigen values are mixed with one negative and one positive.

Marginal Cases:

Higher-order and isolated fixed At least one of the eigen values is zero.

points:

In essence, from the point of view of stability, the marginal cases are those where at least one eigenvalue satisfies Re() = 0.

Definition 16 (Hyperbolic Fixed Points). Hyperbolic fixed points are fixed points where both eigen values of the linearized system

have non-zero real part. These are points that either exponentially grow or decay. Hyperbolic fixed point stability is not affected

by small non-linear terms. Non hyperbolic fixed points are fragile.

The model

x = x(3 x 2y)

y = y(2 x y)

where x(t) = population of rabbits

y(t) = population of sheep

x, y 0

y = 0 = y(2 x y) 2y xy y 2 = 0

x = 0, y = 0 : 3 x 2y = 0, 2 x y = 0 3 x 2(2 x) = 3 x 4 + 2x = 0 x = 1

x = 0 2y y 2 = 0 y = 0, y = 2;

y = 0 3x x2 = 0 x = 0, x = 3;

x=1y=1

Fp = {(0, 0), (0, 2), (3, 0), (1, 1)}

Compute the Jacobian:

3 2x 2y 2x

J=

y 2 x 2y

44 CHAPTER 4. PHASE PORTRAITS - STROGATZ

3 0

J(0,0) = 1,2 = {3, 2}unstable node

0 2

1 0

J(0,2) = 1,2 = {2, 1}stable node

2 2

3 6

J(3,0) = 1,2 = {3, 1}stable node

0 1

1 2

J(1,1) = 1,2 = {1 + 2, 1 2}saddle point

1 1

The construction of the phase portrait from the results of the linear system is a process of piecing the trajectories based on the

characteristics of the nodes. The table below outlines the trajectories at each eigen-value.

4.4. HYPERBOLIC FIXED POINTS,TOPOLOGICAL EQUIVALENCE AND STRUCTURAL STABILITY 45

Fixed

Eigenvalues Description

Point

ries leave the origin tangential to the eigen-direction

0

corresponding to the eigen value = 2, v = ,

1

which spans the y-axis. The phase portrait looks

0 1

(0, 0) 1,2 = {2, 3} v1,2 = ,

1 0

like,

1

the fixed point tangential to = 1, v = . The

2

1 1

(0, 2) 1,2 = {2, 1} v1,2 = ,

0 2

approach

1

the fixed point tangential to v = . The phase

3

0 1

(3, 0) 1,2 = {2, 1} v1,2 = ,

2 3

1 1

(1, 1) 1,2 = 1 2 v1,2 = ,

2 2

looks like,

46 CHAPTER 4. PHASE PORTRAITS - STROGATZ

In the real world once the linear problem is solve and the fixed points obtained one can use intuition to complete the picture

using the local information at the fixed points. In our case the different pieces from the table 4.3 put together leads to the 4.6

(a). Here we see there are trajectories leaving the unstable origin (0, 0) and the saddle point. Trajectories arriving at the saddle

points, stable nodes. There are trajectories that do not go to either of the stable nodes by end up in the saddle point. These

trajectories are a part of the stable manifold of the saddle.

(a) The different fixed points and Trajectories (b) Intuitive Completion of Trajectories

There is an interesting observation to be made of the phase portrait 4.5 (b). There is no equilibrium state where there are

both rabbits and sheep to be found. There are either rabbits or sheep. This phenomenon is named the principle of competitive

exclusion.

Definition 17. Given an attracting fixed point x , we define its basin of attraction to be the set of initial conditions x(t) x

as t .

Definition 18. A stable manifold that separates the basins of two nodes is called the basin boundary.

4.5. CONSERVATIVE SYSTEMS 47

Consider the age old consequnce of Newtons Second Law m x = F (x) where F (x) is some nonlinear force. Under the as-

sumption that F is independent of both x,

t, i.e there is no damping forces or a time-dependent driving force. We can show that

energy is conserved.

Let V (x) be the potential energy of the system. We define F (x) = dVdx(x) . Then,

dV

m

x+ =0 (multiplying both sides by x)

dx

dV d 1 2

mx

x + x =0 mx + V (x) = 0

dx dt 2

integrating we have for any solution x(t)

1

Ex(t) = mx 2 + V (x) (The conservation of energy)

2

Definition 19. A conserved quantity is a real valued continuous function E(x) that is constant on trajectories (x(t)).

Definition 20. Systems for which a conserved quantity exists are called conservative systems.

1. To avoid trivial solution/values one requires that E(x) be non-constant over every open set. If not a constant function

E(x) = 0 would qualify as a conserved quantity and every system would be conservative.

2. The conservation is along a path x(t) and not the entire space.

Ex 6.5.1 Show that conservative systems cannot have attracting fixed points.

Proof. Let x be a fixed point of a conservative system. Since x is a fixed point, x| x = 0 V (x ) = E(x ). Essentially

this implies the energy be constant on all points of its basin of attraction, since all trajectories in the basin of attraction

lead to x . This contradicts the definition of a conservative system. One needs dissipation for the system under no

external influences to come to rest (fixed point).

3. Since attracting fixed points cannot occur, only saddles and centers are the only feasible fixed points in conservative

sytems.

Ex 6.5.2 Consder a particle of mass m = 1 moving in a double well potential, V (x) = 21 x2 + 14 x4 . Find and classify all the

equilibrium points for the system. Plot the phase portrait and interpret the results physically.

Sol. First order of business fixed points.

dV

= 0 x + x3 = 0

dx

x = 0, x2 1 = 0

x = {0, 1}

Note that the higher order term is a cubic and not a quadtratic term. This is mostly due to the fact that it is a model of

a physical system such as a spring/mass. The force vs displacement is either an odd or an even function in such cases. As

pointed out in class by Dr. Nat with his anecdote of Mr Ali, trying to fit data from a biological system to a spring mass

equivalent model. Given that we have a conservative system,

dV

m

x+ =0

dx

x + x3 = 0

x

= x x3

x

defining the phase space as (v = x,

x) we have

v = x x3

x = v

48 CHAPTER 4. PHASE PORTRAITS - STROGATZ

0 1 3x2

J=

1 0

0 1

J(0,0) =

1 0

0 2

J(1,0) =

1 0

= {1, 2i}

Values

Description

0 1

x = (x, x)

= (0, 0) = 1, J0 = Saddle node. Here = 0, = 1 < 0, from 3.2 we have a saddle point.

1 0

0 2

x = (x, x)

= (1, 0) = 2i, J1 = A center. In this scenario, = 0, = 2 > 0 resulting in a center.

1 0

0 2

x = (x, x)

= (1, 0) = 2i, J1 = A center. In this scenario, = 0, = 2 > 0 resulting in a center.

1 0

, x are zero in the above table. The trajectories are shown below.

(a) The directions of the flow are determined by y = x. Given that x = x the x-axis splits the plane on the top where x

is +ve and the plane below the x-axis where x is -ve. The flow in the top half is to the right. The flow on the bottom

half is to the left. The orientation of the neighboring trajectories follow from continuity.

(b) The fact that we have two centers should cause concern about the accuracy of the linear analysis. However, since the

system is conservative, all trajectories follow the conservation law,

1 2 1 2 1 4

E= x x + x = C

2 2 4

The trajectories are therefore closed curves defined by contours of constant energy 3b.

(c) Each of the neutrally stable centers are surrounded by small closed orbits. There are also large orbits that encircle

the three fixed points.

(d) The solutions are typically periodic, except for the equilibrium solutions and two very special trajectories: these are

trajectories that appear to start and end at the origin, ie: they approach the origin as t . Orbits of this type

that start and end at the same fixed point are called homoclinic orbits. These are common in conservative systems.

- liyongmingIJMA1-4-2014Enviado porFirman Nurrobi
- 01338495Enviado porapi-3697505
- Integral Equations -1Enviado porM S Mostafa
- Twu_GM_final_3232012Enviado pormatmean
- Lyapunov ControlEnviado porBharat Mahajan
- General Theory of Elastic Stability_ SouthwellEnviado portpadhy
- Apiwat Etal IL03Enviado porGeof180
- Control of Nonlinear Bilateral Teleoperation Systems Subject ToEnviado porsamim_kh
- C7eEnviado pormorpheus_unknown
- 03 Dynamical Systems TutorialEnviado porPrashanth Mohan
- A Decomposition-Based Practical Approach to Transient Stability-Constrained Unit CommitmentEnviado porVijaya Chintala
- 22880_An Approximate Analytic Solution of the Blasius ProblemEnviado porfrankkqqzz
- International Journal for Numerical and Analytical Methods in Geomechanics Volume 17 issue 10 1993 [doi 10.1002%2Fnag.1610171002] Y. Sukirman; R. W. Lewis -- A finite element solution of a fully coupled implicit formulatio.pdfEnviado porIlan Gargon
- 30.18. IASS Paper Andres Harte N 37Enviado porZeeshan Riaz
- A Remark on Some Nonlinear Elliptic ProblemsEnviado porPatricio Cerda Loyola
- Notes-PhasePlaneEnviado porMinu Priya
- Torque Control in Harmonic Drives with Nonlinear DynamicEnviado porRafaelMC
- CDM-2010-2010-0001-a005Enviado porDevendra Talekar
- Synchronization & Anti-synchronization via Active Control of a chaotic Satellite: NereidEnviado porSEP-Publisher
- PDCass12_1Enviado porAnonymous WkbmWCa8M
- Transient Stability Series Compensators MibEnviado poralfredo quiroga
- lecture8.pdfEnviado porJason Chiang
- Trabajo de Pseudo-metodosEnviado porDiego Mendoza Yerren
- 605Enviado porAnonymous bB5jfj9
- 1306.6672Enviado porCroco Ali
- MATH2011PS08Enviado porJohn Chan
- A novel approach for load flowEnviado porsf111
- GateEnviado porPugazh Arasan
- Slides Chap3Enviado pormemrah2955
- 11Enviado porMoloyBanerjee

- Solutions to StrogatzEnviado porRamesh Kadambi
- Milner-damas Principal TypesEnviado porRamesh Kadambi
- fremlin measure theory volume 3Enviado porRamesh Kadambi
- Deprecating Observer Str 2010Enviado porRamesh Kadambi
- Finite Di Mension Al Vector SpacesEnviado porRamesh Kadambi
- fremlin measure theory book 2Enviado porRamesh Kadambi
- Regression Models NotesEnviado porRamesh Kadambi
- Advanced Engineering AnalysisEnviado porRamesh Kadambi
- OPF Haug DerivationsEnviado porRamesh Kadambi
- Barrier OptionsEnviado porabhishek210585
- Stochastic Processes With Mesure TheoryEnviado porRigoberto Escudero

- Excel Specifications and LimitsEnviado porMaaz Ahmed
- Study of Cavitation in Hydro Turbines - A Review184Enviado porAlberto Aliaga
- Ring TheoryEnviado porIyappan Alagappan
- freemathpreassessmentofbasicskillsgrades68Enviado porapi-327883841
- ciip09_1564_1571.2683Enviado pormuse
- PPEnviado porBookMaggot
- Microphone HandbookEnviado porDiegorila Kong
- The Assessment of Solar Panel in QuezonEnviado porJhon Christian
- Chest X-ray and Coronary Computed Tomography AngiogramEnviado porkristine_silang
- Ee-445-- Electrical Machine Theory & Design_f2013Enviado porSylviaHofer
- Mirco A. Mannucci and Rose M. Cherubin- Model Theory of Ultrafinitism I: Fuzzy Initial Segments of ArithmeticEnviado porNine000
- MDU B.tech Computer Science 4th Sem SyllabusEnviado porVaibhav Agochiya
- woo et alEnviado porSabina Bancila
- perfil aerodinamico matlabEnviado porManuel Huaman
- Rothera Test JurnalEnviado porAzra Zahrah Nadhirah Ikhwani
- tcp fundaEnviado porRakesh Rath
- 01-vectoresEnviado porClaudia Reolid Pérez
- PDMS basic CommandsEnviado porSuhas Mv
- Fuel Gas System DescriptionEnviado porParmeshwar Nath Tripathi
- Data Communication and NetworkingEnviado porsrikant.singh62583
- Artificial Bee Colony Algorithm Based Optimal Reactive Power Flow of Two-terminalEnviado porJaol1976
- math lesson 4-3-17Enviado porapi-354124545
- IRJET-Alpha-Beta Pruning in Mini-Max Algorithm –An Optimized Approach for a Connect-4 GameEnviado porIRJET Journal
- raid linux.docxEnviado poralecsander_008
- DMA MEnviado porDoctorr Camacho
- A Century of Developments in the Chemistry of Flotation ProcessingEnviado porditende
- Starmicronics C# APIEnviado porJd Torres Martínez
- 036 SIGA-UM,Enviado porzezohome
- Ipl1 1xx Xx Xxx d Xx x MktEnviado poryamaha640
- JN0-102Enviado porRamon Pirbux