Differential Equations (AIMS 2015)
Differential Equations (AIMS 2015)
Differential Equations (AIMS 2015)
November 2015
Contents
1
1.1
1.2
Definition of terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3
1.4
Separation of variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5
Exact equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.6
1.7
1.6.1
1.6.2
Homogeneous functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.7.2
The substitution y = vx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
24
2.1
2.2
2.3
2.4
2.5
The Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.6
Phase Portrait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1
2.7
Bifurcations of equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.7.1
Saddle-node bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.7.2
Transcritical bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.7.3
Pitchfork bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.8
Flow on a circle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.9
Oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Fixed points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.1.1
3.2
3.4
4
3.3
50
Routh-Hurwitz criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3.2
Degenerate node . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
64
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2
4.3
4.1.1
4.1.2
4.1.3
Polar coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.1.4
4.1.5
Topological equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.1.6
Conservative Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.2.2
4.2.3
Linard System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.2.4
References
You will find the following books particularly useful for this module.
Bernd J. Schroers. Ordinary Differential Equations: A Practical Guide. Cambridge 2011
Zill, D. G and Cullen M. R. Differential Equations with Boundary-Value Problems, Sixth or later
edition.
Nagle, R. K and Saff, E. B. Fundamentals of Differential Equations and Boundary Value Problems.
1993 or later editions.
Steven Strogatz. Nonlinear Dynamics and Chaos.
Jordan, D. Nonlinear ordinary differential equations: an introduction to dynamical systems.
Chapter 1
Analytic Techniques for Solving First Order
Differential Equations:
1.1
A differential equation is an equation that involves derivatives of some unknown function. If the unknown function is a function of a single independent variable then the differential equation is called an
ordinary differential equation. If the unknown function depends on two or more independent variables
then the differential equation is called a partial differential equation. Equations
2
u
t
= 3 u
x
In equations (1)-(3) the unknown function is y and is assumed to be a function of the single independent
variable x, (i.e y = y(x)). The functions y0 and y00 are the first and second derivatives of y(x) with respect
2
dy d y
to x ( usually denoted by dx
, dx2 ). the equations (1)- (3) have ordinary derivatives depending on one
single variable x and are called ordinary differential equations.
In equations (4) and (5) the unknown function u is a function of two independent variables t and x (and
2
u
is usually written as (u(t, x) or u(x,t)). We define u
x and x2 as the first and second partial derivative of u
with respect to x respectively. And equations (4) and (5) are called partial differential equations because
they involve partial derivatives.
1.2
Definition of terms
(a) The order of a differential equation is the order of the highest derivative which occurs in the
equation. For example
dy
dx
= x+2
d2y
dy
+ 2 dx
+ 3y = 0
dx2
xy0 + 3y = 4
(y00 )2 (y0 )3 + 2y = 5x
z
x
2 z
x2
z
= z + x y
2
z
2
+ y
2 = x +y
second order
(1.1)
is called an nth order ordinary differential equation, for the unknown function y.
An nth order differential equation is said to be linear differential equation if it is of the special
form
a0 (x)y(n) + a1 (x)y(n1) + . . . + an1 (x)y0 + an (x)y = f (x).
(1.2)
where the functions ai (x); i = 0, 1, 2, . . . , n and f (x) are given functions independent of the variable
y, with a0 (x) 6= 0. In general, the nth-order equation (1.2) is linear if the function F(x, y0 , y00 . . . y(n) )
is a first degree polynomial in y, y0 , y00 , . . . , y(n) . The functions ai (x), i = 0, 1, 2, . . . , n are known as
the coefficients of the differential equation.
(c) If any one of the functions ai (x), i = 0, 1, 2, . . . , n is a function of y the differential equation (1.2) is
nonlinear.
When the function f (x) in the differential equation (1.2) is identically zero, we say that the differential equation (1.2) is homogeneous, and when f (x) is not identically zero then (1.2) is nonhomogeneous.
If all the coefficients ai (x); i = 0, 1, 2, . . . , n are constants, then the equation (1.2) is known as a
linear differential equation with constant coefficient. Otherwise if only one of the coefficients
ai (x), i = 0, 1, 2, . . . , n is a variable then the differential equation is variable coefficient.
The term linear refers to the fact that each expression in the differential equation is of degree one
or zero in the variables y, y0 , . . . , y(n) .
1.3
p
x(x 1) is a solution of y0 =
1
2y
The differential equation y00 y = 0 given in (i) above, the function y = ex is a solution too and moreover,
y = c1 ex + c2 ex is again a solution to y00 y = 0 for some constants c1 and c2 .
Therefore the process of finding a solution to a differential equation is to focus ones mind to a unique
solution (called a primitive) that was differentiated to have the combination terms of the differential
equation satisfied.
Such a primitive function of a differential equation is called the general solution of the differential
equation.
Definition 2 A solution to a differential equation that comes from particular qualities specified together
with the equation is called the particular solution. Such a differential equation that has initial conditions
that will result into a particular solution is called an initial value problem (IVP).
Given any primitive we can in some way obtain a differential equation that is associated with it by
eliminating the constants involved.
Exercise
Classify the following differential equations as to linearity, nature of coefficients, order and homogeneity
5
a xy00 2ex y0 5x = 0
b y00 4y0 3 cos x = 5y
c 3y000 4y00 + xy = 11
d y000 6y00 17y0 = xey
e (y00 )2 4y0 + 5y = 6x
f y(4) 5y00 + 11y = 0
g y000 + 5y00 6y0 = cos xy
h y(4) + 3(cos x)y000 + (y0 )2 = 0
i y000 = 5
j y000 5x = 0
1.4
Separation of variables
dy g(x)
=
dx
f (y)
(If can separate xs on one side and ys on the other.R dx and dyR should always be the Numerators)
such that f (y)dy = g(x)dx and on integration gives f (y)dy = g(x)dx F(y) + c1 = G(x) + c2
F(y) = G(x) +C. The method used to solve simple separable differential equations is known as the
method of separation of variables.
Example 1.4.1 Solve the differential equation
Solution: By separation of variables
dy
= 2x dx
y
dy
=
y
dy
dx
= 2xy
2x dx ln y = x2 + c ln y = x2 + ln A
y = Aex
Example 1.4.2
dy
y = 1
dx
dy ydx = dx
dy = (1 + y)dx
dy
= dx
(1 + y)
dy
=
(1 + y)
dx
ln(1 + y) = x + ln A
y = A ln ex 1
The constant A can be got out if given initial conditions.
Example 1.4.3
1. Solve the differential equation:
dy
dx
y1
x+3
y1 x+3
dy
=
y1
dx
ln (y 1) = ln k(x + 3) y = 1 + k(x + 3)
x+3
6x5 2x+1
cos y+ey
dy
dx
(cos y + ey )dy =
(6x5 2x + 1)dx
sin y + ey = x6 x2 + x + c
1
dy =
y
7
dx 2 y = ln (1 + x) + c
1+x
and y(1) = 1 2 = ln 2 + c c = 2 ln 2
x+1
Thus 2 y = ln(1 + x) + 2 ln 2 2 y = 2 ln
y = 1 ln
2
1.5
x + 1 2
.
2
Exact equations
dy
dx
(1.3)
Example 1.5.1
1.
y
= 3xx1
(y 3x2 )dx + (x 1)dy = 0 in which M(x, y) = y 3x2 and
N(x, y) = x 1.
dy
dx
Definition 3 A first-order differential equation whose LHS expressions can be easily regrouped into
total differentials is said to be exact differential equation
Theorem 1.1 The first-order differential equation M(x, y)dx + N(x, y)dy = 0 is exact iff
M(x, y) N(x, y)
=
y
x
(The opposites) where the partial derivatives are continuous functions.
Proof:
Assume M(x, y)dx + N(x, y)dy = 0 is exact. Then there exists a function F(x, y) satisfying
F
F
= M(x, y) and
= N(x, y)
x
y
(Since for any function F(x, y), dF(x, y) =
Then
=
y
y
F
x
F
F
x dx + y dy)
2 F
N
=
and
=
xy
x
x
8
F
y
2 F
=
.
xy
Hence
M
2 F
2 F
N
=
=
=
y
yx xy
x
F
y
F
x
= M(x, y),
Then F(x, y) =
(1.4)
(b) To determine g(y) we take the partial derivatives with respect to y of both sides of (1.4) to give
F
=
y
y
Example 1.5.2
Solve the differential equation: (x2 + y2 )dx + (2xy)dy = 0.
Solution: Let M(x, y) = x2 + y2 then
M
y
N
x
= 2y. Since
M
y
then the differential equation is exact. Let the solution be of the form F(x, y) = c in which
and
F
y
F
x
= M(x, y) = x2 + y2
F(x, y) =
And
F
y
1
(x2 + y2 )dx = x3 + xy2 + g(y)
3
F
x
= 2y =
N
x
= M(x, y)
1
1
Then F(x, y) = x3 + xy2 + g(y) gives the solution F(x, y) = x3 + xy2 +C
3
3
F=
x3
+ x2 y + y2 x y
3
Example 1.5.4 Solve the differential equation: (2xy sec2 x)dx + (x2 + 2y)dy = 0.
Solution: Let M(x, y) = 2xy sec2 x, then
M
y
= 2x =
F
x
= M(x, y) and
N
x
F
y
then
N
x
= 2x. Since
then the differential equation is exact. Let the solution be of the form F(x, y) = c in which
F
y
F
x
F(x, y) =
And
M
y
Example 1.5.5 Solve the differential equation: (1 + ex y + xex y)dx + (xex + 2)dy = 0.
Solution: Let M(x, y) = 1 + ex y + xex y; then
N(x, y) = xex + 2; then
N
x
M
y
= xex + ex . Since
F
x
M
y
= ex + xex =
= M(x, y) and
F
= N(x, y) = xex + 2 F(x, y) =
y
F
y
N
x
x3
3
+ x2 y + y2 x y 43 ]
Example 1.5.7 Solve the differential equation:(3x2 2xy + 2)dx + (6y2 x2 + 3)dy = 0
Solution: M(x, y) = 3x2 2xy + 2,
N(x, y) = 6y2 x2 + 3,
Since
M
y
= 2x =
N
x ,
N
x
M
y
= 2x
= 2x.
(1.5)
F
= N(x, y) = 6y2 x2 + 3
y
(1.6)
(1.7)
(1.8)
(a)
dy
dx
(b)
dy
dx
= y(2 + sin x)
(c)
dy
dx
= 3x2 (1 + y2 )
(d)
dy
dx
+ y2 = y
(e)
dy
dx
sec2 y
1+x2
12
1.6
An integrating factor is a term which multiplied through, it makes the DE, exact.
Definition 4 A first order differential equation is said to be linear if it takes the form
a1 (x)
dy
+ a0 (x)y = b(x)
dx
(1.9)
where a1 (x), a0 (x), b(x) are continuous functions that depend only on the independent variable x.
Example 1.6.1
dy
1. x2 sin (x cos x)y = (sin x) dx
is linear since it can be written in the form
dy
(sin x) dx
+ (cos x)y = x2 sin x
dy
dy
2. y dx
+ (sin x)y3 = ex + 1 is not linear due to the presence of terms y3 and y dx
(1.10)
1.6.1
M
y
= P(x) and
(1.11)
N
x
is not exact. Let by multiplying through equation (1.11) by (x) make it exact. Then [((x)P(x).y
(x)Q(x)]dx + (x)dy = 0 implies that
M
y
N
x
= (x)P(x)
=
d(x)
dx
13
)
(1.12)
M
y
N
x
d(x)
dx
(x)P(x) =
(x) = e
P(x)dx
(1.13)
dy
Then (1.13) is our desired integrating factor; and clearly since (x)P(x) = d(x)
dx then (x) dx +P(x)(x)y =
dy
(x)Q(x) (x) dx
+ d(x)
dx y = (x)Q(x)
d
dx ((x).y) = (x)Q(x)
P(x)dx .
Example 1.6.2
dy
dx
+ 2y = 3ex
P(x)dx
=e
d
2x
3x
dx (e y) = 3e
2dx
Therefore y = ex + ce2x .
2. Solve the differential equation:
1 dy
x dx
= x cos x;
2y
x2
y( 2 ) = 3
2x dx
= e2 ln x =
1
.
x2
Thus
1 dy
x2 dx
x23 y = cos x
d 1
dx
( x2 y) = cos x x12 y = sin x+c. And y( 2 ) = 3 c = ( 12
1) gives the solution y = x2 sin x+
2
( 12
1)x2 .
2
14
1.6.2
M
y
N
x
(1.14)
(1.15)
(1.16)
+ (x, y)
= N + (x, y)
y
y
x
x
M N =
y
x
N M
(x, y)
x
y
(1.17)
= 0 and
d
N
=
dx
d
dx .
N M
(x)
x
y
or
d
1
=
dx N
M N
y
x
15
(x)
(1.18)
And if
1 M
N ( y
N
x ) is a function of x only, say f (x) or a constant, (1.18), would reduce to
d
dx
d
=
(x)
f (x)dx ln (x) =
= 0 and
f (x)dx (x) = e
d
dy .
f (x)dx
(1.19)
d
M N
M
=
(y)
dy
y
x
or
1
d
=
dy
M
And if
1 M
M ( y
M N
y
x
(1.20)
(y)
N
x ) is a function of y only, say g(y) or a constant; we get
d
= g(y)(y)
dy
(1.21)
d
=
(y)
g(y)dy ln (y) =
g(y)dy (y) = e
g(y)dy
(1.22)
(a) If
M
y
N
x
(b) Compute
1 M
N ( y
M
y
6=
N
x
M
y
and
N
x .
N
x ). If this is a function of x only, say f (x); or C (The opposites) then the
R
f (x)dx .
(c) Compute
1 M
M ( y
N
x ). If this is a function of y only, say g(y); or C (The opposites) then the
R
(1.23)
M(x, y) = 2x2 + y,
Since
M
y
= 1 6= 2xy 1 =
N
x ,
M
N
= 1 N(x, y) = x2 y x,
= 2xy 1
y
x
1 2xy + 1
2(1 xy)
2
1 M N
(
)=
=
=
(a function of x only).
N y
x
x2 y x
x(1 xy)
x
R
2x dx
= e2 ln x =
1
.
x2
M
y
= 2y + 2x 6= 2x =
N
x
M
y
= 2y + 2x N(x, y) = x2 ,
N
x
= 2x
We check
1
N
M N
y
x
=
[2y + 2x (2x)] 2y + 4x
=
x2
x2
17
1
M
M N
y
x
[2y + 2x (2x)]
y2 + 2xy
2y + 4x
2(y + 2x) 2
=
= = g(y), (a function of y only)
2
y + 2xy y(y + 2x) y
1 2
x2
(y
+
2xy)dx
dy = 0
2
y
y2
g(y)dy
= e
R 2
y dy
= e2 ln y =
1
.
y2
is exact.
d
Thus dx + 2xy1 dx x2 y(2) dy = 0 dx + dx
(y1 x2 ) = 0
M(x, y) = 2xy,
Since
M
y
= 2x 6= 6x =
N
x ,
1
N
1
M
M
N
= 2x N(x, y) = y2 3x2 ,
= 6x
y
x
M N
y
x
M N
y
x
8x
4
=
(not a function of x only).
2xy y
We compute
=
8x
y2 3x2
(1.24)
(1.25)
M(x, y) = 2xy y2 ,
Since
M
y
6=
N
x ,
N
M
= 2x 2y N(x, y) = 2y3 x2 ,
= 2x
y
x
1
N
M N
y
x
=
4x 2y
2y3 x2
We compute
1
M
M N
y
x
=
2(2x y) 2
4x 2y
=
( a function of y only).
=
2
2xy y
y(2x y) y
2
g(y)dy = e y dy == 1 .
y2
1
2
(7) y0 + y = xex + 1
(8)
dx
dy
= ey x;
y(1) = 0
1.7
1.7.1
Consider a first order differential equation M(x, y)dx + N(x, y)dy = 0. Then the functions M(x, y) and
N(x, y) are said to be homogeneous if they both contain expressions of the same degree.
More generally, a function f (x, y) is said to be a homogeneous function of degree n if f (ax, ay) =
dy
= f (x, y) if the righthand side function f (x, y) can
an f (x, y). And to a first order differential equation dx
be written as a function of the ratio xy alone then the equation is homogenous.
Example 1.7.1
1. The function f (x, y) = 2x2 + y2 is homogeneous with degree 2 because
f (ax, ay) = 2(ax)2 + (ay)2 = a2 (2x2 + y2 ) = a2 f (x, y).
p
x3 + y3 is homogeneous with degree
(ax)3 + (ay)3 =
3
2
because
p
3p
3
a3 x3 + a3 y3 = a 2 x3 + y3 = a 2 f (x, y).
20
ax
2ay
+4 =
x
2y
ax
a(2y)
5.
dy
dx
xy
x
6.
dy
dx
y2 +2xy
x2
7.
dy
dx
= ln x ln y + x+y
xy = ln
dy
dx
8.
x
+ 4) = a0 f (x, y).
+ 4 = a0 ( 2y
= 1 ( xy ) is homogeneous
x3 4xy
x2
= ( xy )2 + 2( xy ) is homogeneous
1
y
x
1+ y
+ 1 xy is homogeneous
x
= x 4( xy ) is not homogeneous.
1.7.2
The substitution y = vx
If the degree of all the functions in M(x, y) is the same as the degree of the functions in N(x, y) then the
RHS of the differential equation
dy
M(x, y)
=
= f (x, y)
dx
N(x, y)
can clearly be written as a ratio of
y
x
21
Summary 1.7.1 If M(x, y) is the same as the degree of the functions in N(x, y), then use the substitution
y = vx = dy = vdx + xdv which will become a separable DE in x and v. and later substitute y back
in the answer.
But first check if the differential equation is not separable, exact or linear as these are easily solved with
methods mentioned earlier.
Example 1.7.2
1. Solve the differential equation (x y)dx + xdy = 0
c
x
Solution: Both M(x, y) = xy + y2 + x2 and N(x, y) = x2 have functions of the same degree.
And y = vx
dy
dx
dy y2 + 2xy
y
y
=
= ( )2 + 2 .
2
dx
x
x
x
dv
dv
dv
= v + x dx
gives v + x dx
= v2 + 2v x dx
= v2 + v
1
1x dx = ( 1v v+1
)dv. Then
v
ln Ax = ln v+1
Ax =
v
v+1
R 1
R 1
1
x dx = ( v v+1 )dv ln x = ln v ln (v + 1) + c.
Ax =
y
x
y
x +1
22
y
y+x .
Therefore y =
Ax2
1Ax .
Section B
23
Chapter 2
Introduction to dynamical systems
Expected Learning Outcomes
My expectations are that when we are done with this chapter, you should be able to:
Convert higher order equations into a system of first order autonomous equations.
Calculate equilibrium points (sometimes also called fixed points or steady states).
Use the stability theory to decide whether the equilibrium points are stable, neutrally stable or
unstable.
Sketch phase portraits for one-dimensional system.
1 Introduction
t, s R: additivity
to Dynamical Systems, Lecture Notes for MAS424/MTHM021 (Version 1.2, 18/04/2008), Rainer Klages.
24
In other words, a dynamical system is a mathematical formula for evolving the state of a system in time.
Other possible answers to the above question may include the following:
1. Any system whose state changes when an independent variable (such as time) changes is a
dynamical system.
2. Any system that allows us to determine the future states of a system given its present or past state.
The initial value problem
dx
= 2x,
dt
x(t0 ) = x0 ,
1.
The basic goal of dynamical systems is to determine or characterize the long term behaviour of a system
(long term means as t ).
For example, the logistic equation
p
dp
= rp 1
,
dt
k
defines a dynamical system p(t), the population with growth rate r(t).
In studying this system we want to know how the population would shape up in the future. We want to
answer some or all of the following questions;
1. would the population numbers eventually explode (i.e grow to infinite)?
2. would the population eventually collapse (i.e would the human race become extinct in future)?
3. would the population stabilize? and if so, at what value? Its about 6 billion now, how far can we
keep own growing in numbers before we become a danger to ourselves, the numbers threaten our
own existence?
25
Dynamical Systems
Continuous Time DS
Discrete Time DS
Odes, Pdes
Maps
Difference Equations
Figure 2.1: Characterization of Dynamical Systems
2.
In this module we consider only deterministic dynamical systems which are defined by rules,
formulas or functions to indicate how the future states are obtained from the present or the past
states of the system. Differential and difference equations define deterministic dynamical systems.
xn+1 = M(xn ).
This last equation is sometimes called the equation of motion of the dynamical system.
2.1
x R, f : R R,
where there is only one dependent (and one independent) variable defining the system. Typical examples
include the following population growth models;
The exponential population growth model
dN
= rN,
dt
where N is the population at time t and r is the growth rate parameter. This equation has the
solution
N(t) = N0 ert ,
where N0 is the population at time t = t0 .
What happens to this population?
The logistic growth
dN
N
= rN 1
,
dt
k
where N(t) is the population at time t, and r and k are positive constants. (NB. In reality both r
and k could be functions of t).
27
2.2
x = 2x
where x =
dx
,
dt
x = 2xt
(2.1)
(2.2)
Equation (2.1) does not depend explicitly on t, that is, there is no t on the right-hand side of the equation!
Such equations are said to be AUTONOMOUS. On the other hand, equation (2.2) is multiplied by t on the
right hand side. It is NON - AUTONOMOUS.
NB : Any non-autonomous system can always be converted into a system of autonomous equations by
treating the time t as an unknown variable. We shall see show this is done shortly.
2.2.1
Higher order odes can be written as a system of first order equations. As an example, consider the
second order equation of the damped harmonic oscillator:
mx + bx + kx = 0,
where m is the mass, b the damping and k the stiffness of the system. Write this equation as a system of
two first order equations:
Solution: Let x1 = x and x2 = x x = x2 to get:
x1 = x2
x2 = mk x1 mb x2
28
2 linear equations
(2.3)
The two first order equations (2.3) can be written in matrix form x = Ax;
x =
mk
mb
where x =
x1
x2
(2.4)
Using equation (2.4) as an example, we note that the general form of a system of linear equations is;
x = Ax + c,
where x Rn , A Rnn (a matrix) and c Rn (aconstant vector). We note the following:
If c 0 the system is homogeneous
If A and c are both independent of t, then the system is autonomous
If A or c are functions of t then the system is nonautonomous
Example 1 Convert the one-dimensional non-autonomous equation of the forced damped oscillator;
mx + bx + kx = F cost,
to an autonomous system of first order equations.
S OLUTION: Proceed in two steps as follows:
1. S TEP 1: Convert the system into a 2-dimensional non-autonomous system. Let x1 = x and x2 =
x x = x1 . The system now becomes;
x1 = x2
F
x2 = mk x1 mb x2 + m
cost
d
dt
x1
x2
mk mb
x1
x2
0
F
m cost
x1
x1
d
k
b
F
k
b
x2 +
x2 =
dt
= m x1 m x2 + m cos x3
0
m
x3
x3
= 1
0
0 0
F
m cos x3
29
2.3
(2.5)
Definition 6 (Equilibrium Points) The equilibrium points of equation (2.5) are all values of x for which
f (x) = 0.
Example 2 Find the equilibrium points of the logistic equation
dx
= x(k x),
dt
where and k are positive constants.
SOLUTION :
and x(t) = k.
The equilibrium or fixed points x = 0 and x = k are fixed for all time, but there may be other solutions
such that f (x) = 0. How can we find the long term behaviour of these other (unknown) solutions?
First we note that the fixed points x(t) = 0 and x(t) = k divides the real number line into three intervals;
see Figure 2.2. Now let us consider what happens in each of the three regions.
x=0
x=k
x<0
x0 < x < k
x>k
Figure 2.2: The three subintervals of the number line and the equilibrium points of the logistic equation
Region 1: x < 0. Notice that in this region f (x) = x(k x) < 0 which means that;
dx
<0
dt
30
x=k
x<0
0<x<k
x>k
Figure 2.3: We indicate the fact that x is decreasing for x < 0 by an arrow pointing to the left of 0 on the
x-axis
I MPLICATION: The implication of the above result is that a solution of the logistic equation with
an initial condition near x = 0 but to the left of 0 will decrease with time (i.e. move away towards
x as t = ).
Region 2: 0 < x < k. In this region f (x) = x(k x) > 0 which means that;
dx
>0
dt
is a strictly increasing function of t 0 < x < k
We indicate this fact by an arrow pointing to the right between 0 and k , see Fig. 2.4.
x=0
x<0
x=k
-
0<x<k
x>k
Figure 2.4: We indicate the fact that x is increasing in 0 < x < k by an arrow pointing to the right of 0 on
the x-axis
I MPLICATION: A solution of the logistic equation with an initial condition x(0) lying between 0
and x will increase monotonically with t and converge k k as t = .
Region 3: x > k. In this region f (x) = x(k x) < 0 which means that;
dx
<0
dt
is a strictly decreasing function of t x > k
We indicate this fact by an arrow pointing to the left in the region x > k , see Fig. 2.5.
I MPLICATION: A solution of the logistic equation with initial condition x(0) > k will decrease
monotonically with t and converge at k as t = ).
31
x=0
x=k
-
x<0
0<x<k
x>k
Figure 2.5: We indicate the fact that x is decreasing for x > k by an arrow pointing to the left in the region
x > k.
Fig. 2.5 is called the Phase Portrait of the logistic equation:
dx
= x(k x) for all , k > 0.
dt
The fixed point x = 0 in (Fig. 2.5) is called a REPELLER. This fixed point is unstable.
The fixed point x = k is called an ATTRACTOR. This fixed point is stable.
E XERCISE. Consider the one-dimensional system:
dx
= x x3 .
dt
Find the equilibrium points and classify these equilibrium points as repellers or attractors. Sketch the
phase portrait.
2.4
dx
= f (x) with a fixed point x = x so that f (x ) = 0.
dt
Definition 7 (Stable Equilibrium) The fixed point x is asymptotically stable (or just stable) if values
of x near the fixed point move closer to the fixed point with time.
Mathematically; the fixed point x is stable if
> 0 such that x0 with |x0 x | < lim |x(t) x0 | = 0.
t
2.4.1
We have seen that we can use the phase portrait to deduce whether an equilibrium point is stable or
unstable, alternatively, we can use the derivative test for stability.
Definition 9 (Derivative Test for Stability) Given the differential equation
dx
= f (x),
dt
suppose that f (x) and f 0 (x) are continuous and that f (x ) = 0 (where x is an equilibrium point). If
f 0 (x ) 6= 0 then:
1. If f 0 (x ) < 0 then x is an attractor (a stable equilibrium point).
2. If f 0 (x ) > 0 then x is a repeller (an unstable equilibrium point).
3. If f 0 (x ) = 0 then the stability test is inconclusive.
2. Find the fixed points and classify them. Give the phase portraits and solution curves;
(a) x = x2 1
(b) x = x cos x
2.5
The Potential
dV dx
dx dt
dV
=
dx
2
dV
dV
=
0,
dx
dx
The implication of the statement
dV
dx
(2.6)
2
0 is that V (x) decreases monotonically with the flow. In
34
2.6
Phase Portrait
2.7
Bifurcations of equilibria
This is a qualitative change in the character (or behaviour) of solutions of a nonlinear system as a parameter is varied. By varying a parameter, we can talk of
2.7.1
Saddle-node bifurcation
This type of bifurcation results in the creation and destruction of fixed points (meaning that new fixed
points appear and disappear as one varies a certain parameter)
Saddle-Node Bifurcations is also known as;
Turning point bifurcation
Fold Bifurcation
Blue sky bifurcation (the inventors of this name thought the bifurcation just appeared from
nowhere!)
Saddle-Node Bifurcation is best portrayed by the differential equations;
x = r + x2
x = r x2
where r is a parameter that can be positive, zero or negative.
The above equations are called the normal or generic form of Saddle-Node bifurcation.
NB. The two equations above are equivalent under the change of variables;
x x
and r r.
35
(2.7)
(2.8)
2.7.2
Transcritical bifurcation
(2.9)
near x = 1 and show that the system undergoes a transcritical bifurcation at a certain value of r.
2.7.3
Pitchfork bifurcation
This is bifurcation in the presence of symmetry e.g. consider an upright beam buckling under a load, it
can either go left or right. And if there is an equilibrium point on the left there will be a corresponding
equilibrium point on the right. Pitchfork bifurcation has two sub-categories shown below;
Pitchfork Bifurcation
Supercritical
pitchfork bifurcation
Subcritical
pitchfork bifurcation
Normal Form
Normal Form
x = rx + x3
x = rx x3
36
2.8
Flow on a circle
One-dimensional flow on a line x = f (x) is non-reversible, a particle always moves monotonically either
towards a fixed point or towards . Flow on a circle is still one-dimensional except that by moving
along a circle, a particle may eventually return to its starting point. We say that the solutions are periodic.
The generic (or standard) form of a vector field on a circle is
= f (),
where is a point on the circle and is the uniquely defined velocity vector at that point. Here f is
2-periodic with f ( + 2) = f ().
Example 4 Consider the one-dimensional system x = sin x. Plot the phase portrait if the system is
considered as flow on a line.
2.9
Oscillations
For one-dimensional flow on a line, x = f (x), the direction of the flow never reverses. Trajectories must
increase or decrease monotonically or stay constant. The only possibilities are that:
the trajectory approaches a fixed point as t
the trajectory diverges to
For flow on a circle, periodic or oscillatory solutions are possible.
37
W ORKSHEET 1 T UTORIAL
1. Convert the second order equations to a system of autonomous first order equations and write these
equations in matrix form (provide also the appropriate boundary conditions);
(a) x = x2 + (1 + cost)x,
x(0) = 1,
x(0)
=0
(b) x + x + x = g sint,
(c) xx
t x + = 0,
x(0) = 1,
x(0)
=0
2. In the following exercises, (i) find all the fixed points and determine their stability, (ii) sketch the
phase portrait and (iii) discuss the long-term behaviour of all solutions.
(a)
dx
= (x 1)(x + 2),
dt
(b)
dx
= (x 1)(x + 2)(x + 3),
dt
(c)
dx
= (x 1)2 (x + 2),
dt
(d)
dx (3 x)(1 + x)
=
,
dt
x2 x + 2
(e)
dx
= (x2 9)(sin x 2)
dt
(a)
dx
= 2x7 3x6 + x5 2x + 1,
dt
(b)
dx
= xex + sin x,
dt
x = 0.5.
x = 0.
d
x(t) = cos x ex .
dt
NB. Plot the curves of cos x and ex separately. The fixed points are the intersections of these two
curves. Find at least the first 3 or 4.
5. A population subjected to harvesting (with a constant harvesting effort, h(x) = ex) can be modelled
by the equation
dx
x
= r0 1
x ex,
dt
k
where r0 and k are strictly positive and e is non-negative.
(a) Sketch graphs of the right hand side as a function of x in the cases e < r0 and e > r0 .
(b) Sketch a graph of the equilibrium points as functions of e (these are called bifurcation diagrams). Indicate on your diagram which equilibrium points are asymptotically stable and
which are unstable.
6.
(a) For which values of a does the equation = sin(a) give a well- defined vector field on the
circle?
(b) Find and classify all the fixed points of = sin 2 and sketch the phase portrait on the circle.
2.10
to Dynamical Systems, Lecture Notes for MAS424/MTHM021 (Version 1.2, 18/04/2008), Rainer Klages
39
Note that xn+1 = f (xn ) is the equation of motion of the dynamical system. Given initial condition x0 the
equation determines the outcome after n discrete time steps as follows:
x1 = f (x0 ) = f 1 (x0),
x2 = f (x1 ) = f ( f (x0 )) = f 2 (x0 )
..
.
xn
..
.
= f n (x0 ) =
f f f f (x0 )
|
{z
}
n-fold composed map
0 xn 1, 0 < r 4.
To get information about the long term behaviour of this map (given an initial condition x0 ), we look for
all iterates {x0 , x1 , x2 , . . . , xn }, which define the trajectory or orbit of f (x0 ) using cobwebs.
2.11
2.11.1
Stability analysis
Suppose that x is a fixed point of xn+1 = f (xn ), and consider a small perturbation of x : n = xn x .
Then (by Taylors theorem);
xn+1 = f (xn ) = f (x + n )
= f (x ) + f 0 (x )n +
40
1 00 2
f (x )n +
2!
= x + f 0 (x )n +
1 00 2
f (x )n +
2!
x + n+1 = x + f 0 (x )n +
1 00 2
f (x )n + .
2!
(2.10)
We call = f 0 (x ) a multiplier or eigenvalue. From equation (2.10) we get the following results;
1 = f 0 (x )0 = 0
2 = 1 = 2 0
3 = 2 = 3 0
..
. =
n =
..
.
n1 = n 0
3. marginally stable if: Linearly stable if: || = | f 0 (x )| = 1 (NB. you need a cobweb diagram to
prove stability)
4. superstable if: || = | f 0 (x )| = 0 (here we have quadratic convergence with n+1 2 ).
41
2.12
Graph of a function
The goal of this section is to understand the behaviour of the orbits of a given map without having to
compute the orbits, but simply by looking at the graph of the function f (x). To plot the graph of a
function in Matlab, we can use the command fplot. For instance, to plot the function
f (x) = 2.5x(1 x)
in the range of x values from 0 to 1, we may type
>> fplot(0 2.5 x (1 x)0 , [0 1],0 r0 );
(The string r specifies that the graph should be a red line, the default is a blue line).
Alternatively, we can plot the function using plot command: first specify the range of x values:
>> x = 0 : 0.01 : 1;
Next, compute the values of the function and plot the result:
>> f = 2.5. x. (1 x);
>> plot(x, f ,0 r0 );
(Note that we use .* for multiplication instead of just *. The reason is that * is interpreted by Matlab
as a vector multiplication. The .* symbol tells Matlab to multiply vectors element by element.)
2.13
There is a simple geometric procedure for describing the behaviour of orbits using only the graph of
f (x). It is sometimes referred to as the cobweb diagram. Lets say we want to show on the function
graph an orbit starting from x0 . To help us do that, we first draw the diagonal line y = x, which makes a
45 angle with the x- and y-axes.
>> hold on; plot([0 1], [0 1],0 k0 );
(The command hold on holds the current plot of the function, so that the black line from (0, 0) to (1, 1)
is added to it).
The next point on the orbit is the number f (x0 ). The graph y = f (x) allows us to read off this point, since
(x0 , f (x0 )) is the point on the function graph directly over x0 .
>> x0 = 0.1;
>> x = x0; f = 2.5.*x.*(1-x);
>> plot([x x], [x f]);
42
2.14
A defining characteristic of a chaotic orbit is that it displays exponentially sensitive dependence on initial
conditions.
Consider the map
xn+1 = f (xn ).
(2.11)
For a chaotic system, two orbits x1 and x2 originating from very close initial conditions (such as x1 (0) =
0.3 and x2 (0) = 0.300000001), the orbits initially stay close to one another for a number of iterates, but
eventually drift apart. The separation grows exponentially fast with the number of iterates (observe this
by plotting the natural logarithm of the separation between the orbits (that is, log(abs(x2 x1 ))) versus
the number of iterates.This quantity grows approximately linearly until it reaches a plateau. In the linear
growth region
log |x2 x1 | a + n,
where a is the logarithm of the initial separation between the orbits and n is the number of iterates.
Alternatively, this may be written as
|x2 x1 | A exp( n) where A = exp(a),
which shows that the separation indeed grows exponentially with n. The coefficient is called the
Lyapunov exponent of the dynamical system. The Lyapunov exponent characterizes the rate at which
nearby trajectories diverge away from each other.
For the map (2.11), consider two orbits starting at x0 and x0 + 0 , where 0 is infinitesimally small. The
next points of the two orbits are
x1 = f (x0 ),
x1 + 1 = f (x0 + 0 ).
(2.12)
(2.13)
Note that, since 0 is very small, we can use the Taylor series for f (x0 + 0 ) at x0 to write the last
equation
x1 + 1 = f (x0 ) + f 0 (x0 )0
1 = f 0 (x0 )0 ,
where f 0 (x0 ) is the derivative of f evaluated at x0 .
Similarly, for the nth points along the orbits, xn = f (xn1 ), xn + n = f (xn1 + n1 ), so that
n = f 0 (xn1 )n1 .
(2.14)
Equation (2.14) is useful in deriving a formula for computing the Lyapunov exponent of the map f (x).
First, note that the logarithm of the separation between the two orbits starting at x0 and x0 + 0 changes
approximately linearly with n, that is,
log |n | log |0 | + n,
44
1
log |n /0 |,
n
for some large n. Now, we use equation (2.14) to rewrite the ratio n /0 as
n
= (n /n1 )(n1 /n2 ) . . . (2 /1 )(1 /0 ) = f 0 (xn1 ) f 0 (xn2 ) . . . f 0 (x1 ) f 0 (x0 ).
0
Finally, the formula for computing the Lyapunov exponent is given by
1
1 n1
log | f 0 (xn1 )| + log | f 0 (xn2 )| + . . . + log | f 0 (x1 )| + log | f 0 (x0 )| = lim log | f 0 (xi )|,
n n
n
i=0
where depends on x0 .
For chaotic orbits > 0, the Lyapunov exponent is positive, because the separation increases. However,
if we consider parameter values where the orbits converge to a fixed point or a periodic orbit, then the
separation between the orbits will decrease. The rate of decrease will also be exponential.
(a) S TABLE FIXED
< 0).
rx
r(1 x)
0 x 12
1
2 < x 1,
(2.15)
so | f 0 (x)| = r for all x [0, 1]. The fixed points are stable for r < 1 and unstable for r > 1. The Liapunov
exponent is:
1 n1
ln | f 0(xi)|
n n
i=0
= lim
45
1 n1
ln r = ln r.
n n
i=0
= lim
The Liapunov exponent is independent of the initial conditions x0 . The tent map has chaotic solutions
for r > 1.
o 0 o
2.15
Bifurcation diagrams
The behaviour of a dynamical system can change quite dramatically with the change of system parameters. These changes are not only quantitative, such as, for example, change in the location of a fixed
point, but also qualitative: fixed points can be created or destroyed, their stability can change, the system
behaviour can change from regular (stationary or periodic) to irregular - chaotic. It is the qualitative
changes in the system dynamics that are the subject of investigation in the theory of dynamical systems.
Definition 13 Qualitative changes in the system dynamics are called bifurcations, and the parameter
values at which they occur are called bifurcation points.
Consider the so-called logistic map:
xn+1 = rxn (1 xn ).
For the values of parameter r just below 3.0 the orbits converge to a stable fixed point. When the value
of r exceeds 3.0, the fixed point becomes unstable, and the orbits converge to a stable period-2 orbit,
which is created at r = 3.0. Therefore, we say that r = 3.0 is a bifurcation point of the logistic map. The
bifurcation that occurs at r = 3.0 is called a period-doubling bifurcation, which is one of many types of
bifurcations that can occur in dynamical systems. One of the goals of the dynamical systems theory is to
classify different types of bifurcations and investigate their properties.
2.15.1
In order to study bifurcations in dynamical systems, it is convenient to visualize the bifurcations that
happen at different parameter values. A good way to see the general behaviour of the system at different
parameter values is to plot the orbits as a function of the parameter. That is, we will plot the orbit points
xn along the vertical axis against the values of parameter r along the horizontal axis. Such a plot is
called a bifurcation diagram.
Below is a Matlab script file that constructs a bifurcation diagram (see Figure 2.6) for the logistic map
with parameter r in the range from 2.5 to 4.
Npre = 200; Nplot = 100; x = zeros(Nplot,1); for r = 2.5:0.005:4.0,
46
x(1) = 0.5;
for n = 1:Npre,
x(1) = r*x(1)*(1 - x(1));
end,
for n = 1:Nplot-1,
x(n+1) = r*x(n)*(1 - x(n));
end,
plot(r*ones(Nplot,1), x, ., markersize, 2);
hold on;
end, title(Bifurcation diagram of the logistic map); xlabel(r);
ylabel(x_n); set(gca, xlim, [2.5 4.0]); hold off;
Bifurcation diagram of the logistic map
1
0.9
0.8
0.7
xn
0.6
0.5
0.4
0.3
0.2
0.1
0
2.5
3.5
2.16
Brief Summary
and
47
f (q) = p.
p, q =
r+1
p
(r 3)(r + 1)
,
2r
which are real for r > 3. Thus a 2-cycle exits for r > 3.
3. A NALYZING THE STABILITY OF A CYCLE: The 2-cycle is stable if p and q are stable fixed points
for f 2 . Hence the 2-cycle is stable if || < 1 where;
=
2.17
d
[ f ( f (x))]x=p = f 0 ( f (p)) f 0 (p) = f 0 (q) f 0 (p).
dx
1. F IXED POINTS AND C OBWEBS: Find the fixed points and decide whether they are stable or not
(in addition to the mathematically argument, use a cobweb also).
(i) xn+1 = xn ,
(ii) xn+1 = xn3 .
(iii) xn+1 = ln xn .
2. Consider the map xn+1 = 3xn xn3 .
(i) Find all the fixed points and classify their stability.
(ii) Draw a cobweb starting at x0 = 1.9 and another starting at 2.1. Try to explain the dramatic
differences between the two orbits.
3. Suppose you want to find the roots of an equation g(x) = 0. Then Newtons method says you should
consider the map xn+1 = f (xn ) where
f (xn ) = xn
g(xn )
.
g0 (xn )
(a) Write down the Newton Map xn+1 = f (xn ) for the equation g(x) = x2 4 = 0.
48
(c) Show that the fixed points are superstable. Show that the modified Newton method converges
quadratically!]
4. L OGISTIC M AP:
(a) Find the value of r at which the logistic map has a superstable fixed point.
(b) Let p and q be points in a 2-cycle for the logistic map.
(i) Show that if the cycle is superstable, then either p =
1
1
or q = . (In other words, the
2
2
point where the map takes on its maximum must be one of the points in the 2-cycle.
(ii) Find the value of r at which the logistic map has a superstable 2-cycle.
5. Given the quadratic map xn+1 = f (xn ) = r xn2 , where r and x can take on any real values,
(a) Find all the fixed points as a function of r.
(b) Find any values of r at which the fixed points bifurcate and classify those bifurcations.
(c) For which values of r is the 2-cycle superstable?
(d) For which values of r is there a stable 2-cycle?
6. Consider the cubic map xn+1 = f (xn ) = rxn xn3 .
(a) Find the fixed points. For which values of r do they exists? And stable?
(b) To find the 2-cycles of the map, suppose that f (p) = q and f (q) = p. Show that p, q are roots
of the equation x(x2 r + 1)(x2 r 1)(x4 rx2 + 1) = 0 and use this to find all the 2-cycles.
7. Calculate the Liapunov exponent for the linear map xn+1 = rxn .
49
Chapter 3
Second-order linear systems
The equation
d
dt
x
y
=A
x
y
+ F(t) where
x
y
R2 ,
F R2 ,
and A is an nn matrix with constant entries is homogeneous if F(t) 0, and inhomogeneous otherwise.
Hence homogeneous second order linear systems are systems of the form
x = f1 (x, y) = ax + by
y = f2 (x, y) = cx + dy.
(3.1)
In matrix form
x = Ax
where A =
a b
c d
and x =
x
y
R2 .
2. Neutral stability,
3. Liapunov stability.
Construct phase portraits for a given linear planar system.
3.1
Fixed points
3.1.1
a b
c d
,
we define the trace of the matrix tr(A) = = a+d and the determinant by = ad bc. The characteristic
polynomial is
P() = 2 + = 0.
51
Hence the eigenvalues are obtained from the formulas (compare with lecture notes):
2 4
1,2 =
,
2
= 1 2 ,
= 1 + 2 ,
3.2
We have already seen that the eigenvalues of a matrix A are the roots of the characteristic polynomial
P() = det(I A) = 0.
Solving the equation P() = 0 is non-trivial for 3 3 matrices or higher dimensions. A possible way to
find the eigenvalues is to note the following;
1. A general polynomial P() can be represented in the form
P() = n + an1 n1 + . . . + a1 + a0 .
(3.2)
2. If the n roots of P() are 1 , 2 , , n , then P() can also be written in the form
P() = ( 1 )( 2 ) . . . ( n ).
(3.3)
Expanding equation (3.3) and equation coefficients of like powers of in equations (3.2) and (3.3) gives
the following relationships between the coefficients a j and the roots j ;
an1 = i
i
an2 =
i j
i< j
an3 =
i< j<k
..
.
52
i j k
a0 = (1)n 1 2 n .
Now, the important results to note are the following;
a0
= P(0) = ()n det(A),
an1 = tr(A),
where tr(A) is the sum of the diagonal elements of A.
NB. If all the eigenvalues 1 , 2 , , n of A have strictly negative real parts, then
5 3 1
A = 0 4 2 .
0 0 4
1 2 1
0
1 .
A= 1
4 4
5
53
= 3 62 + 11 6.
To find the roots of P() we try the integer factors of a0 = 6, namely 1, 2, 3 and 6. Since
P(1) = 0, then = 1 is a root of P() = 3 62 + 11 6.
Hence ( 1) is a factor of P(). Dividing P() by ( 1) gives
P() = ( 1)(2 5 + 6 = ( 1)( 2)( 3).
3.2.1
Routh-Hurwitz criteria
The Routh-Hurwitz criteria provides necessary and sufficient conditions for the stability of higher order
systems (n 3).
Given any polynomial P() = n + an1 n1 + . . . + a1 + a0 , define the matrices
M1 = (an1 ),
M2 =
an1 an3
1
an2
,
M4
an1 an3
1
an2
0
an1
0
1
an5
an4
an3
an2
an7
an6
an5
an4
, , Mn =
an1 an3
1
an2
0
an1
0
1
0
0
0
0
0
0
an5
an4
an3
an2
an1
..
.
an+1
an+2
an+3
an+4
an+5
0
0
a1
a0
2. For each of the following matrices A, determine if the system x = Ax is stable or not.
(i)
3
1
0 2
(ii)
2
1
7 3
1 1
7
(iii) 4 1 8
0
0
1
(vi)
3.3
1 1
2 3
1
0 2 4 0
4
(vii)
0
8
0
0 5
0
0
0 1
2
1 1 5
(iv) 4 0 2
6 2 0
1 2 3
2 0
0 3
2 2 4 2
3 4 0
(vii)
3 1
0 0 5
0 5
7 0 1
0 0
0 1
1. S ADDLE
signs.
POINTS :
These occur if < 0. For this case the eigenvalues are real and have opposite
2. S PIRALS AND C ENTERS: These occur when > 0. With > 0, eigenvalues are either real with
the same sign (NODES), or complex conjugates (SPIRALS AND CENTERS)
3. N ODES: These satisfy 4 > 0.
4. S PIRALS: These satisfy 4 < 0.
5. The parabola 2 4 = 0 is the border line between nodes and spirals; star nodes and degenerate
nodes live on this parabola.
3.3.1
55
If = 0, at least one of the eigenvalues is zero. Then the origin is not an isolated fixed point. There is
either a whole line of fixed points or a plane of fixed points if A = 0.
3.3.2
Degenerate node
Degenerate nodes are obtained when there is only one eigen direction. The degenerate node is on the
borderline between a spiral and a node.
3.4
x0 (t) = ax + by
y0 (t) = cx + dy
d
dt
x(t)
y(t)
=
a b
c d
x
y
.
(3.4)
x = 0
y = 0
i.e.,
f (x, y) = 0
g(x, y) = 0
The only equilibrium point is the origin (0, 0). It is worth noting that (0, 0) is the only critical point if
ad bc 6= 0, that is, if the determinant of the coefficient matrix is non-zero.
The system has solutions of the form
x(t) = uert
y(t) = vert .
(3.5)
(r a)u bv = 0
c + (r d)v = 0
r a b
c r d
u
v
= 0.
This has a non-trivial solution if the determinant of the matrix is non-zero, that is
56
(3.6)
r a b
c r d
=0
(3.7)
Equation (3.7) is called the characteristic equation for the system above. It has the roots
r1,2 =
(a + d)
p
(a + d)2 + 4(ad bc)
.
2
The asymptotic (or long-term)behaviour of the solutions to (3.4) depend on the roots r1 and r2 , namely
whether the roots are;
1. real, distinct and positive,
2. real, distinct and negative
3. real and of opposite sign,
4. equal,
5. complex.
Case 1: r1 , r2 are real, distinct and positive
Consider the simple decoupled system
x(t)
= x,
y(t)
= 3y
1 0
0 3
r2 4r + 3 = 0.
A=
This has roots r1 = 1 and r2 = 3 (both real, distinct and positive). This gives the general solution
x(t) = Aet ,
y(t) = Be3t ,
where A and B are constants. The integral or solution curves for this system are a family of curves given
by the equation
y = Cx3 ,
57
for different values of C obtained by combining the previous two equations or by solving the equation
dy 3y
= .
dx
x
N B. When the roots r1 , r2 are real, distinct and positive, the fixed point (0, 0) is an unstable improper
node. All the trajectories move away from the origin with time.
Case 2: r1 , r2 are real, distinct and negative
A single illustrative example is
x(t)
= 2x,
y(t)
= y
A=
which has roots r1 = 2 and r2 = 1 (both real, distinct and negative). This gives the general solution
x(t) = Ae2t ,
y(t) = Bet ,
The integral curves
x = A/B2 y2
x = Cy2 .
The origin is asymptotically stable since the trajectories approach this point in the long term. The fixed
point (0, 0) is a stable improper node.
Example 7 Classify the critical point at the origin and sketch a plausible phase portrait of the system
x = 5x + 2y,
y = x 4y.
(3.8)
By setting x = 0 and y = 0, it is clear that the fixed point is (0, 0), that is, the origin. The system above
gives the coefficient matrix
A=
1 4
r2 + 9r + 18 = 0 : r1 = 6, r2 = 3.
58
m=
x 4mx
1 4m
=m=
5x + 2mx
5 + 2m
1
m = , 1.
2
and
y = x.
Sketch these on your diagram and show the direction of flow along each line.
59
x(t)
= x,
y(t)
= y
A=
0 1
r1 = 1, r2 = 1.
In this case the fixed point (0, 0) is a saddle point. It is unstable and the solution of the equations is
x(t) = Aet ,
y(t) = Bet ,
The integral curves
x=
C
.
y
Example 8 Classify the critical point at the origin and sketch a plausible phase portrait of the system
x = 5x 3y,
y = 4x 3y.
(3.9)
By setting x = 0 and y = 0, it is clear that the fixed point is (0, 0), that is, the origin. The coefficient
matrix leads to the characteristic equation
r2 4r 3 = 0
r1 = 1, r2 = 3.
Hence the origin is a saddle point. Re-write the system of equations in the form
4x 3y
dy
=m=
,
dx
5x 3y
and substitute y = mx to get the transformed axis
y = 2x
2
and y = x.
3
Sketch these on your diagram and show the direction of flow along each line.
60
Case 4: r1 = r2
A single illustrative example is
x(t)
= sx,
y(t)
= sy
s 0
A=
0 s
r1 = s, r2 = s.
In this case the fixed point (0, 0) is a proper node. It is unstable if s > 0 and stable if s < 0. The solution
of the equations is
x(t) = Aest ,
y(t) = Best ,
The integral curves x = Cy, that is, a family of straight lines through the origin.
Case 5: Complex roots
If the roots are complex, then they are of the form
r1 , r2 = i,
where , R. We distinguish the following cases;
1. If 6= 0, the fixed point is called a spiral or focus. It is a stable focus if < 0 and an unstable
focus if > 0.
2. If = 0, then the roots are purely imaginary and the fixed point is called a centre. A centre is an
example of a fixed point that is neutrally stable.
Example 9 Find the fixed points, and sketch the phase portrait of the system
x = 2x y,
y = x + 2y.
(3.10)
Setting x = y = 0 gives the fixed point (0, 0). Note that the characteristic polynomial is
(r 2)2 + 1
r = 2 i.
W EEK 2 T UTORIAL
P ROBLEMS
1. Write the following systems in matrix form and obtain the eigenvalues and eigenvectors of the
coefficient matrix.
(a) x = y, y = x.
(b) x = 3x 2y, y = 2y x.
2. Consider the system x = y, y = x.
(a) Sketch the vector field.
(b) Show that the trajectories of the system are hyperbolas of the form: x2 y2 = C
(c) The origin is a saddle point; find equations for its stable and unstable manifolds.
(d) The system can be decoupled and solved as follows. Introduce new variables u and v where
u = x + y, v = x y. Then rewrite the system in terms of u and v. Solve for u(t) and v(t),
starting from an arbitrary initial condition u0 , v0 ).
(e) What are the equations for stable and unstable manifolds in terms of u and v?
3. For each of the following systems, decide whether the origin is attracting, Liapunov stable, asymptotically stable or none of the above.
(a) x = 0, y = y
(b) x = x, y = 5y.
(c) x = 2y, y = x.
62
(b) The general solution is x(t) = c1 e1t v1 + c2 e2t v2 . This way of writing x(t) involves complex
coefficients and looks unfamiliar. Express x(t) purely in terms of real-valued functions. (Hint:
use eit = cos t + i sin t to write x(t) in terms of sines and cosines, and then separate the
terms that have a prefactor of i from those that dont.)
6. Plot the phase portrait and classify the fixed point of the following linear systems.
(a) x = y, y = 2x 3y
63
Chapter 4
Phase plane analysis
In this chapter we want to extend our discussion of systems of equations started in the previous chapter
to nonlinear equations. A typical example of such systems that we shall be concerned about are the
Lotka-Volterra equations:
x =
Ax Bxy
y = Cy + Dxy
(= f (x, y))
(= g(x, y)).
These equations model the dynamics of the interaction of competing species, in this case the predators
x(t) and the prey population y(t). The first thing to note is that nonlinear equations are notoriously
difficult to solve and invariably, the tendency is to resort to numerical methods such as the Runge - Kutta
methods. The basic aim here remains unchanged: we want to find the long-term behaviour of the system.
Routh-Hurwitz conditions
Hopf bifurcation theorem
4.1
Summary
Theorem 4 (Existence and uniqueness Theorem) The initial value problem x = f (x), x(0) = x0 has a
unique solution x(t) on some interval (, ) if f is;
1. continuous, and
2. all its partial derivatives are continuous for all x in the interval.
Note that the system is stable if x(t) 0 as t . It is unstable otherwise. In terms of the eigenvalues, 1 , 2 , . . ., of the system, the real parts of all the s must be strictly negative for stability to be
guaranteed.
C OROLLARY
1. Different trajectories never intersect (if they did, the uniqueness theorem would be violated).
2. In a closed orbit C, any trajectory starting inside C is trapped forever inside C.
3. If there are fixed points inside C, then the trajectory might eventually approach one of them.
4.1.1
Given
dx
= f (x, y)
dt
dy
= g(x, y),
dt
the fixed points are x , y such that
x
y
=
f
x
g
x
65
f
y
g
y
!
x
y
.
Let
J=
f
x
g
x
f
y
g
y
=
a11 a12
a21 a22
,
4.1.2
The fixed points as predicted by the linearized system are the same as those of the nonlinear system, if
they are;
1. S ADDLE POINTS,
2. S PIRALS,
3. N ODES.
If the linearized system predicts that the fixed points are borderline cases, that is centers, star nodes,
degenerate nodes or non-isolated fixed points, then we cannot guarantee that the nonlinear system really
has this type of fixed point. These may be altered by small nonlinear terms. The stability is established
as in Worksheet 2.
4.1.3
Polar coordinates
Sometimes it is easier to deal with a seemingly complicated nonlinear system by first transforming it into
polar coordinates (IMPORTANT: I expect you to be able to do this). Let x = r cos , y = r sin so that
x2 + y2 = r. Then using = tan1 (y/x), show that = (xy yx)/r
2.
Alternatively, we could proceed by using complex numbers. In particular, let
i .
x + iy = rei = r(cos + i sin ) x + iy = (r + ir)e
66
4.1.4
If Re() 6= 0 for both eigenvalues, the fixed point is often called hyperbolic (Notice that all saddle points
are hyperbolic!). The stability type of hyperbolic fixed points is not affected by small nonlinear terms.
Non-hyperbolic fixed points are more fragile.
4.1.5
Topological equivalence
Two phase portraits are topologically equivalent if one is a distorted version of the other, obtained e.g.,
by bending or warping. Ripping is not allowed, e.g., a closed orbit must remain a closed orbit even after
distortion.
T OPOLOGICAL equivalence implies homeomorphism (a continuous deformation with a continuous inverse) that maps one local phase portrait onto the other while preserving the direction of the arrows.
Theorem 5 (Hartman-Grobman Theorem) The local phase portrait near a hyperbolic fixed point is
topologically equivalent to the phase portrait of the linearization, in particular, the stability of the fixed
point is preserved by linearization.
4.1.6
Conservative Systems
These are system for which a conserved quantity exists. Usually, energy is the conserved quantity. If
x = F(x) (Newtons law), set
F(x) =
Integrate to get the energy E(x).
67
dV
.
dx
1. P HASE P ORTRAITS: For each of the following systems, find the fixed points. Sketch the nullclines
and a plausible phase portrait.
(a) x = x y, y = 1 ex ,
2. F IXED P OINTS AND L INEARIZATION: Find the fixed points, classify them, sketch the neighbouring trajectories, and try to fill in the rest of the phase portrait.
(a) x = x y, y = x2 4,
(b) x = 1 + y ex , y = x3 y,
(c) x = xy 1, y = x y3 .
(c) Show that |x(t) y(t)| 0 as t for all trajectories. (Hint: form a differential equation
for x y.)
4. N ONLINEAR TERMS CAN CHANGE A STAR INTO A SPIRAL: This example shows that borderline fixed points (stars, centers, degenerate nodes) are sensitive to nonlinear terms. Consider the
system in polar coordinates given by r = r, = 1/ ln r.
(a) Find r(t) and (t) explicitly (brush up your first year integration!) given an initial condition
(r0 , 0 ).
(b) Show that r(t) 0 and |(t)| as t . Therefore the origin is a stable spiral for the
nonlinear system.
(c) Write the system in x, y coordinates.
(d) Show that the linearized system about the origin is x = x, y = y. Thus the origin is a
stable star for the linearized system.
5. L OTKA -VOLTERRA E QUATIONS:
(a) Find the fixed points, investigate their stability, draw the nullclines, and sketch the plausible
phase portraits. Indicate the basins of attraction of any stable fixed points where x = x(3
2x y), y = y(2 x y) with x, y 0.
68
(b) The following system of differential equations models an interaction between two populations:
x = x(1 x) xy
y = xy y,
where , and are strictly positive constants.
i. Briefly describe the type of interaction that these equations might be modelling.
ii. Find the equilibrium points of these equations and determine the values of the parameters for which they satisfy x 0, y 0 and the values for which they are asymptotically
stable.
iii. Sketch phase portraits of the system for the cases (a) > and (b) < .
6. C ONSERVATIVE S YSTEMS. Consider the system x = x3 x.
(a) Find all the equilibrium points and classify them.
(b) Find a conserved quantity.
(c) Sketch the phase portrait.
7. R EVERSIBLE S YSTEMS: Show that the system x = y(1 x2 ), y = 1 y2 is reversible.
8. S IMPLE P ENDULUM: Consider a simple pendulum with no damping and no external driving
force:
mL
d2
d2 g
=
mg
sin
+ sin = 0.
dt 2
dt 2 L
Show that
(a) the origin is a nonlinear centre,
(b) the system is conservative,
(c) find a conserved quantity,
(d) Sketch a phase portrait for the system.
9. P ENDULUM WITH DAMPING: Now consider + b + sin = 0:
(a) Show that
= v
v = sin bv.
(b) Find and classify the fixed points.
(c) Show that the energy of the system, E decreases monotonically along the trajectories.
(d) Complete the phase portrait.
o 0 o
69
4.2
4.2.1
Theorem 7 (Non-existence of closed orbits) A dynamical system would not have closed orbits if:
1. it is a gradient system, that is, x = V where V (x) is a single valued scalar function.
2. a Liapunov function exits (this is an energy-like function that decreases along the trajectories.
3. Dulacs or Bendixson Negative Criterion applies.
P ROOF OF (1): (By contradiction) Suppose on the contrary that there is a closed orbit C with period T
such that x(T ) = x(0). Then
x(T ) = x(0) V (x(T )) = V (x(0)),
(i.e. there are 2 times 0 and T when the potential is the same. However, by chain rule
2
dV dx
dV
dV
=
=
0.
dt
dx dt
dx
So V (t) is strictly decreasing function of time and either x(t) is a fixed point for all time or the assumption
is not correct. There is thus no trajectory x(t) such that x(T ) = x(0).
NB :
dV
,
dx
70
2V
2V
x y
=
xy yx
y x
(4.1)
Definition 16 (Liapunov function) A function V (x) is a Liapunov function if the two conditions below
hold:
1. V (x) > 0 for all x 6= x and V (x ) = 0, (i.e, V must be positive definite),
2.
dV
dV
< 0 for all x 6= x , that is,
is negative definite, meaning all trajectories flow downwards
dt
dt
towards x .
If
dV
0,
dt
P ROBLEM: There is no systematic way to construct V (x), often try sums of squares.
Theorem 8 (Bendixson Negative Criterion) Let x = f (x) where f is continuously differentiable on a
simply connected region R. if
f =
x y
+
x y
has one sign in R, that is, either f > 0 or f < 0, then there are no closed orbits lying entirely in R.
Bendixsons negative criterion can be generalized to D ULAC S C RITERION:
Theorem 9 (Dulacs Criterion) If there exists a real valued function g(x) such that
(gx) = (g f ),
has one sign in R, then there are no closed orbits in R.
71
y = g(x, y),
has no fixed points in R. If there exists a trajectory C that is confined within R, then either C is a closed
orbit or it spirals toward a closed orbit as t In either case, the system contains a closed orbit.
Alternatively, the Poincar-Bendixson Theorem may be stated in the simpler form: If there are no fixed
points within C, then the trajectory must eventually approach a closed orbit.
The Poincar-Bendixson Theorem is one of the central results of nonlinear dynamics.
4.2.2
This an oscillator whose motion is governed by the van der Pol equation
x + (x2 1)x + x = 0
where > 0.
The damping term is (x2 1)x so that when |x| < 1 amplification occurs and when |x| > 1 decay occurs.
Alternatively;
x = y
y = (x2 1)y x.
4.2.3
Linard System
This is a generalization of the van der Pol equation and models oscillating circuits with the equation;
x + f (x)x + g(x) = 0,
or equivalently;
x = y
y = g(x) f (x)y.
Theorem 11 (Linards Theorem) Let f (x) and g(x) be continuous functions and let
F(x) =
Z x
f (s)ds,
G(x) =
Z x
g(s) ds.
0
72
In some books Linards Theorem is referred to as the Levinson and Smiths Theorem
4.2.4
where
0 << 1,
and h(x, x)
is an arbitrary smooth function. Weakly nonlinear oscillators represent small perturbations
of the linear oscillator x + x = 0. Two important examples are the:
1. VAN DER P OL EQUATION: x + x + (x2 1)x = 0 and the
2. D UFFING E QUATION: x + x + x3 = 0.
73
1. Sketch the phase portraits and each limit cycle and fixed point.
(a) r = r3 4r, = 1,
y = x + y y(x2 + y2 ),
(c)
ii. x = 2xex
2 +y2
, y = 2yex
2 +y2
75
1. (a) To get you start, let x = y and z = t, then the system x = x2 + (1 + cost)x reduces to
x = y
y = y2 + (1 + cos z)x
z = 1.
Complete the rest of the solutions
(b)
(c)
2. These are simple second year problems, but to help you practice your Matlab skills, here are simple
programs you could use to plot the curves:
(a) Given x0 = x(t), x(0) = 1, write a program (and save it as func1.m):
function dxdt=func1(t,x)
dxdt=-x;
Note that the function func1 calculates the right-hand side of the differential equation you
want to solve. Now, at the command prompt, type the following commands:
clear
[t,x]=ode45(func1,[0,5], 1);
plot(t,x)
xlabel(time)
ylabel(function values)
The above would give the output in Fig. 4.1. In the line [t,x]=ode45(func1,[0,5], 1), the 0
and the 5 are the initial and final values of t. The 1 is the initial value of x. Since our initial
condition was x(0) = 1, we started at y at 1 and t at 0.
(b) To sketch the graph of the IVP: x0 = x(4 x), x(0) = 1, you could write a program func2.m
as;
76
1
0.9
0.8
function values
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
time
3.5
function values
2.5
x = x(4x), x(0) = 1
2
1.5
0.5
1
time
1.5
1.4
1.2
x = 1
function values
1
x,
1 + t
x(0) = 0
0.8
0.6
0.4
0.2
0
0
0.5
1
time
1.5
(d) x =
(3 x)(1 + x)
: Fixed points x = 0 x = 1, 3.
x2 x + 2
4. (a) f (x) = 2x7 3x6 + x5 2x + 1 f 0 (x) = 14x6 18x5 + 5x4 2 f 0 (0.5) = 2.031 < 0:
Attractor.
(b) f (x) = xex + sin x f 0 (x) = (1 x)ex + cos x f 0 (0.5) = 2 > 0: Repeller.
5. To find the fixed points of
d
x(t) = cos x ex , you need to sketch the graphs of cos x and ex on the
dt
78
same axis. The intersection points are the fixed points, see Fig. 2. Read these off from the graph.
Now use the stability analysis to determine the stability of these intersection points.
NB. There are an infinite number of fixed points for this system.
25
20
f(x)
15
10
5
8
2
x
4.3
Plot of f(x)=2(1x)xx
0.4
0.2
f(x)
0.2
0.4
0.6
0.8
1
0.5
0.5
0.8
0.6
0.4
0.2
0.2
(iii) xn+1 = ln xn = ln x = x. This does not have a solution. You can see this if you plot y = ln x
and y = x on the same axes.
80
7.5
5
2.5
-4
-2
2
-2.5
-5
-7.5
g(xn )
where g(x) = x2 4.
g0 (xn )
(c) f 0 (x) =
x2 + 4
xn2 4
= xn+1 = n
.
2xn
2xn
x2 + 4
= x = x2 4 = 0 = x = 2.
2x
2x2 8
= f 0 (2) = 0. Hence the fixed points are superstable.
4x2
4. L OGISTIC M AP:
1
(a) Logistic map: f (x) = rx(1x) so fixed points are x1 = 0 and x2 = 1 . So, f 0 (x) = r(12x)
r
and f 0 (0) = r. Thus x1 = 0 is superstable if and only if r = 0.
Similarly, | f 0 (1 1/r)| = |r(2/r 1)| = 0 if and only if r = 0 or r = 2.
[NB. taking r = 0 is obviously not a smart idea since this produces f (x) = 0!. So the only
viable option for superstability is r = 2.]
(b) Let p and q be points in a 2-cycle for the logistic map.
(i) Then, since f 0 (x) = r(1 2x), we want to find p and q such that
( f 2 (p))0 = f 0 ( f (p)) f 0 (p) = f 0 (p) f 0 (q) = r(1 2p) r(1 2q) = 0.
81
1
1
or q = .
2
2
(ii) We want r such that p = 1/2 or q = 1/2. Let us choose q = 1/2. Then
1
(but q = )
2
f (p) = q
and
f (q) = p
f (p) =
1
2
and
f (1/2) = p
rp(1 p) =
1
2
and
1
r = p r3 4r2 + 8 = (r 2)(r2 2r 4) = 0,
4
r=2
or r = 1 + 5
(since r 0.
1 + 4r
1
for r .
2
4
r = x 2 = p2 .
x = 0 = q,
Then proceed as before.
(d) Given that f 0 (x) = 2x, the objective is to find a p and an q such that
|( f 2 (p))0 | = | f 0 (p) f 0 (q)| = |4pq| < 1.
This p (and q)is a solution of the second generation map f 2 (p) = p. Thus
f 2 (x) =
r (r x2 )2 = x
r (r2 2rx2 + x4 ) = x
x4 2rx2 + x + r2 r = 0
now factorize
4
2
2
2
x 2rx + x + r r = (x + x r)(x2 x + 1 r) = 0.
The first term comprises the fixed points of the map f (xn ). The points p and q on the 2-cycle
are solutions of (x2 x + 1 r) = 0. So,
p=
1 1
+
4r 3
2 2
and q =
1 1
4r 3.
2 2
Hence
1 1
+
= f (p) f (q) = 4
4r 3
2 2
0
1 1
4r 3 = 4r + 4.
2 2
3
5
<r< .
4
4
Thus at r = 5/4 the 2-cycle loses its stability. Check for flip bifurcation at r = 3/4.
6. C UBIC M AP: xn+1 = f (xn ) = rxn xn3 .
0
| f ( r 1)| = | 2r + 3| < 1 if 1 < r < 2.
(b) Suppose that f (p) = q and f (q) = p. Then p, q are roots of the second generation map
f 2 (x) = x. Now;
f 2 (x) = f ( f (x)) = x
r(rx x3 ) (rx x3 )3 = x
83
provided r2 4 0
1
1p 2
r 4
r
2
2
7. L IAPUNOV E XPONENT: Calculate the Liapunov exponent for the linear map xn+1 = rxn .
o 0 o
84
B IFURCATION OF M APS
(A) T RANSCRITICAL B IFURCATION: The normal form is xn+1 = f (xn ) = (1 + r)xn xn2 .
(a) Fixed points: f (x) = (1 + r)x x2 = x x = 0, r.
(b) Stability: f 0 (0) = 1 + r so x = 0 is stable if |1 + r| < 1 1 < 1 + r < 1 2 < r < 0 and
x = r is stable if | f 0 (r)| = |1 r| < 1 0 < r < 2.
Transcritical bifurcation (change of stability) occurs at r = 0.
(B) S ADDLE -N ODE B IFURCATION (aka tangent or fold bifurcation):The NORMAL FORM is xn+1 =
f (xn ) = r + xn + xn2 .
1. (a) x (t) =
0 1
1 0
x(t),
eigenvectors we need to solve (A I)v = 0 for each where v = (v1 , v2 )T . Thus for the
1 You
should always check the solutions (and let me know of errors, typos, etc)
85
1 1
1 1
v1
v2
=
0
0
v1 =
1
1
s,
where s is arbitrary. Now (for simplicity) take s = 1, so that v1 = (1, 1)T . Find the eigenvector corresponding to 2 = 1 in the same way to be v2 = (1, 1)T .
(b) x (t) =
3 2
1 2
(a)
x
dy dy dt
=
= xdx ydy = 0 x2 y2 = C.
dx dt dx y
dy x
= = m (the gradient). Now substidx y
v = x y = y + x = v.
To solve this system, not that the characteristic polynomial is
P() = |A I| = ( + 1)( 1) = 0 = 1.
These are the same as in 2(a).
(e) Check how 2(c) was done!
3. (a) Note that x(t) = x0 is constant for all time. Also A =
0 0
0 1
, so 1 = 0,
2 = 1.
Since = 1 + 2 = 1 < 0 we expect the origin to be stable. We also expect a whole line of
fixed points on the x-axis and decay along the y-axis. The origin is Liapunov stable.
(b) A =
, so 1 = 1, 2 = 5. Since = 1 2 = 5 > 0 we expect a node and
1 0
0 5
0 2
1 0
4 1
2 1
4
x. Now, |A I| =
2
1
1
= 2 5 + 6.
1
2
2t
e + c2
1
1
e3t
5. (a)
87
(b) x(t) = c1 e
cost
sint
+ c2 e
sint
cost
.
6. (a)
(b)
x
y
=
0
1
2
x
v
k
b
, = . So,
m
m
.
1p 2
42 . The three possible scenarios are:
(b) The eigenvalues are; 1,2 =
2 2
(i) 2 > 42 : distinct real eigenvalues with = 1 2 = 2 > 0 : origin must be a node.
(ii) 2 = 42 : repeated (real) eigenvalue. If v0 is the eigenvector corresponding to =
/2 then the second solution is y(t) = tet v0 +et where is a solution of (AI) =
v0 . [You should do the calculations to find y(t).]
(iii) 2 < 42 : complex eigenvalues.
88
1. P HASE P ORTRAITS: For each of the following systems, find the fixed points. Sketch the nullclines
and a plausible phase portrait.
(a) N ULLCLINES x = 0 x = y and y = 0 ex = 1. F IXED POINTS: Not that ex = 1 iff x =
0 y = 0 So the only fixed point is (x , y ) = (0, 0). To classify the fixed points, we first
find the Jacobian: J(x, y) =
1 1
ex 0
J(0, 0) =
1 1
1 0
. The characteristic
polynomial is:
1 1
= 2 1,
P() =
1
with eigenvalues 1,2 = (1 5)/2 (which are real, distinct and with opposite sign), = 1 >
0, = 1 < 0 so the point (0, 0) is a saddle point. To plot the phase portrait, you must first
find the stable and unstable manifolds (which are not the same thing as the nullclines!). Find
the eigenvectors and proceed as in Worksheet 3.
(b) N ULLCLINES:
x = 0 x = 0 or y = 2 x
y = 0 y = x.
F IXED POINTS; (x , y ) = (0, 0) and (x , y ) = (1, 1) and the Jacobian;
J(x, y) =
2 2x y x
1
1
J(0, 0) =
2 0
1 1
.
1 1
1 1
,
89
BASINS OF ATTRACTIONS: We can identify a basin of attraction for the stable point
(1, 1): The line x = 0 (i.e., the y-axis) is an invariant manifold (a nullcline which is
parallel to the flow lines) and so acts as a separatrix (separating those points being
attracted and those moving away). All points in the right half-plane (x > 0) flow to
(1, 1).
M ANIFOLDS: The stable manifold is the y-axis. The unstable eigenvector at (0, 0) is
(3, 1)T , so the unstable manifold is a sheet stretching from the origin, in the direction
(3, 1) which goes to the stable spiral, and continues in the direction (3, 1) to infinity.
2x y
x
2y
2x 2y
J(0, 0) =
0 0
0 0
= 0.
1 1
2x 0
J(2, 2) =
1 1
4 0
,
J(2, 2) =
1 1
4 0
ex 1
3x2 1
J(0, 0) =
1 1
0 1
P() = 2 1.
(c) You know the drill! (1, 1), stable node and (1, 1) saddle point.
3. For the system x = y3 4x, y = y3 y 3x.
(a) The fixed points would lie on the line y x = 0 (subtracting y3 y 3x = 0 from y3 4x = 0).
Stable node at (0, 0) and saddle points at (2, 2).
90
dx dy
d
= (x y)
(x y) = (x y)
dt dt
dt
1
d(x y) = dt
xy
ln |x(t) y(t)| = t +C
|x(t) y(t)| = Aet 0
as t .
1
t (having conveniently {but perhaps wrongly??} assumed
ln r
91
y
3\
\
\
\
\
He
2 HH \
HH\
H\
u
H\H
\HH
\ HH
HH
\
HH
\e
e
H
\
H
(0, 0)
1.5
Figure 4.8: Nullclines for 5(a) with fixed points, (1,1) is a stable node.
Equilibrium points; (x , y ) = (0, 0), (1, 0) and (x = /, y = (1 /)/) with x 0,
y 0 iff < . Jacobian:
J(x, y) =
J(0, 0) =
J(1, 0) =
1 2x y x
y
x
1 0
0
1
0
J(/, (1 /)/) =
,
,
saddle point,
,
x x
y
0
,
= x < 0, = x y ,
92
1
a bb
b
b
@
@
b
b
b
@
@
b
b
b
@
@
@
b
b
bb
r/b
r/b
(ii)
(i)
0
1
3x2 1 0
= J(0, 0) =
0 1
1 0
= = i.
Hence (0, 0) is a linear centre. Now check that it is a nonlinear centre by showing that the
system is reversible, that is, it is invariant under the change of variables t t and y y.
Thus we have a centre at (0, 0). Sketch the phase portrait near the origin.
(b) The conserved quantity (energy)is given by
E(x) =
=
dV
1 2
x +V (x) where y =
= x3 x
2
dx
1 2 1 2 1 4
y + x x .
2
2
4
0
1
3x2 1 0
= J(1, 0) =
93
0 1
2 0
= = 2.
Hence (1, 0) are saddle points. Sketch the phase portrait around each point and then
combine these plots with the plot in (a) to obtain the phase portrait for the system.
7. R EVERSIBLE S YSTEMS: The system x = y(1 x2 ), y = 1 y2 is reversible since the equations
are invariant under t t, y y.
8. S IMPLE P ENDULUM. Given
mL
d2 g
d2
=
mg
sin
+ sin = 0,
dt 2
dt 2 L
t
non-dimensionalize the system by setting: = = t, =
T
+ sin = 0
( =
g
, so that
L
d
).
d
0
1
cos 0
. If n is zero
or even;
J(0, 0) = J(2k, 0) =
0 1
1 0
,
= i.
Hence (0, 0) is a linear centre. Now note that the system is invariant under the transformation
, v = v. Hence the system is reversible, implying that (0, 0) is a nonlinear centre. To
show that the system is conservative, note that
+ sin ) = 0
+ sin = 0 (
1
E(, v) = ( 2 cos )
2
94
d 1 2
( cos ) = 0
dt 2
1 2
1
1 2
v 1 + . . . ' (2 + v2 ) 1,
2
2
2
remember here that near (0, 0), is small and so we have no problem expanding the cosine term.
Hence E has a local minimum at (0, 0) and so the origin is a nonlinear centre. If n = 2k + 1, then
J(, 0) = J((2k + 1), 0) =
1
1
1 = 1 : v1 =
0 1
1 0
,
= 1,
2 = 1 : v2 =
1
1
.
0
1
cos b
0 1
1 b
,
saddle points.
95
1. P HASE P ORTRAITS:
(a) Given: r = r3 4r, = 1; then r = 0 r (r2 4) = 0 r = 0 or r = 2. The fixed
points are the origin (0, 0) and the circle r = 2. The flow is counterclockwise (since = 1 >
0).
For 0 < r < 2, r < 0 r is decreasing, so trajectories in this region spiral away from
the circle towards the origin.
For r > 2, r > 0 r is increasing, so trajectories in this region spiral towards infinity.
The limit cycle r = 2 is unstable.
(b) r = r(1 r2 )(9 r2 ), = 1, so r = 0 r = 0, 1 or r = 3. The fixed points are then the
origin and 2 limit cycles r = 1 and r = 3. The flow is counterclockwise (since = 1 > 0).
For 0 < r < 1, r < 0 r is decreasing, so trajectories in this region spiral away from
the circle towards the origin.
For 1 < r < 3, r < 0 r is decreasing, so trajectories in this region spiral towards the
limit cycle r = 1. The limit cycle r = 1 is semi-stable.
For r > 3, r > 0 r is increasing, so trajectories in this region spiral away towards
infinity. The limit cycle r = 3 is unstable.
(c) Given: x = x y x(2r r2 ), y = x + y y(2r r2 ), the 1st thing is to rewrite the system
1
in polar coordinates. Since r2 = x2 + y2 and = 2 (xy yx)
(you can work this out from
r
y
tan = ) then;
x
rr = xx + yy = x[x y x(2r r2 )] + y[x + y y(2r r2 )]
= (x2 + y2 ) (x2 + y2 )(2r r2 ) = r2 r2 (2r r2 ) = r2 (r 1)2
r = r(r 1)2 .
1
1
= 2 (xy yx)
= 2 [x x + y y(2r r2 ) y x y x(2r r2 ) ]
r
r
=
1 2
(x + y2 ) = 1.
2
r
So, the fixed points are the origin r = 0 and the limit cycle r = 1. From here, you know the
drill!
3 You
96
2. T O SHOW:
r = r(1 r2 )
= 1
x = x y x(x2 + y2 )
y = x + y y(x2 + y2 )
0
1
2
1 2axy ax 3ay2 + a
J(0, 0) =
0 1
1 a
p
1
1,2 = (a a2 4),
2
if a = 0 then 1,2 = 2i origin is a centre [cant really happen since a > 0]
if |a| > 2 then a2 4 > 0 origin is a saddle node
if |a| = 2 then a2 4 = 0 origin is a degenerate node
if 0 < a < 2 then a2 4 < 0 1,2 = i origin is an unstable spiral.
[In general, if we didnt know a priori that a > 0, we would have to consider all the above
possibilities.]
(b) In polar coordinates:
rr = xx + yy = xy + y[x ay(x2 + y2 1)] = ay2 (x2 + y2 1)
= ar2 sin2 (r2 1)
r = ar(r2 1) sin2 .
Similarly,
r2 = xy yx = (x2 + y2 ) axy(x2 + y2 1)
= r2 ar2 sin cos (r2 1)
97
x y
=
y x
(8y)
(2x)
=0=
: gradient system
y
x
2
2
2
2
2
2
2
2
so, again we have a gradient system. This rules out any possibility of a periodic orbit for the
system.
(c) If V = ax2 + by2 is a Liapunov function, it must satisfy two conditions; (1) V (x) must be
positive definite, i.e. V (x) > 0 and (2) V (x) must be negative definite or semi-definite, i.e.
V (x) 0. So, as a starting point, assume a, b > 0 to satisfy condition (1). Next:
dV
= 2axx + 2byy = 2ax(y x3 ) + 2by(x y3 ) = 2(a b)xy 2(ax4 + by4 ).
dt
dV
= 2(x4 + y4 ) < 0 the derivative is negative defidt
J(x, y) =
J(0, 0) =
1 1
1 1
1,2 = 1 i,
origin is an unstable spiral.
(ii) Polar coordinates:
rr
r
=
=
=
=
1
(iii) r < 0 when r 2r3 < 0 r1 > .
2
(iv) r > 0 when r r3 > 0 r2 < 1.
1
(v) The trapping region is the annulus < r < 1. Now consider (which you must find)
2
and then use the Poincar-Bendixson theorem.
(b) We use rr = xx + yy and r2 = xy yx:
= a + r cos .
So, for the system r = r(1 r2 (1 + 3 sin2 )), = a + r cos , lets see if we can find a trapping
region,
(i) We first find a circle of maximum radius R1 such that the flow is radially outwards:
1
r = r(1 r2 (1 + 3 sin2 )) > r(1 4r2 ) > 0 r < R1 = .
2
99
(ii) Next find a circle of minimum radius R2 centered at the origin so that the flow is radially
inwards along the circle:
r = r(1 r2 (1 + 3 sin2 )) < r(1 r2 ) < 0 r < R2 = 1.
The region
1
< r < 1 is the trapping region and we can only apply the Poincar-Bendixson
2
theorem if there are no fixed points within this region. To find the location of the fixed points
(if any), lets consider r = r(1 x2 4y2 ) = 0 r = 0 and 1 = x2 + 4y2 which is an ellipse.
And from = a + x = 0 x = a, which is a straight line. The fixed point is the intersection
of this line with the ellipse. Clearly, there is no intersection for |a| > 1 implying that there are
no fixed points for |a| > 1. Hence by the Poincar-Bendixson theorem there exists a closed
orbit (its a lot harder to say anything about the case |a| < 1).
(c) Consider x = y + ax(1 2b r2 ), y = x + ay(1 r2 ) where a and b are parameters 0 < a
1, 0 b < 1 and r2 = x2 + y2 .
(i) You can proceed as in the previous examples using rr = xx + yy and r2 = xy yx.
Alternatively, in order to illustrate the use of complex numbers, let us proceed in the
following manner.
i , where i = 1. Now
Consider: x + iy = rei = r(cos + i sin ) x + iy = (r + ir)e
add the first equation to the second multiplied by i, and combine with these results;
= y ix + a(1 r2 )(x + iy) 2abx
x + iy = ei (r + ir)
= irei + a(1 r2 )rei 2abr cos
r + ir = ir + a(1 r2 )r 2abr cos ei
= ir + a(1 r2 )r 2abr cos (cos i sin ).
Now separate real and imaginary parts to get
r = ar(1 r2 2b cos2 ),
1 2b,
R1 =