Professional Documents
Culture Documents
AND
ORDINARY DIFFERENTIAL
EQUATIONS
I-Liang Chern
Department of Mathematics
National Taiwan University
Contents
1 Introduction
1.1 Modeling population of single species . . . . . . . .
1.1.1 Simple population growth model . . . . . .
1.1.2 Logistic population model . . . . . . . . . .
1.1.3 Exact solutions for the logistic equation . .
1.1.4 Qualitative analysis for the logistic equation
1.1.5 Logistic population model with harvesting .
1.1.6 Abstract logistic population models . . . . .
1.2 First-order equations . . . . . . . . . . . . . . . . .
1.2.1 Separation of variables . . . . . . . . . . . .
1.2.2 Linear first-order equation . . . . . . . . . .
1.2.3 Bernoulli equation . . . . . . . . . . . . . .
1.2.4 * Riccati equation . . . . . . . . . . . . . .
1.3 Existence, uniqueness . . . . . . . . . . . . . . . . .
1.3.1 Local existence theorem . . . . . . . . . . .
1.3.2 Uniqueness theorem . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
1
1
2
2
3
5
6
7
8
11
13
14
14
14
15
2 Linear Oscillators
2.1 Models for linear oscillators . . . . . . . . . . . . .
2.1.1 The motion of a mass attached to a spring .
2.1.2 Electrical Circuit . . . . . . . . . . . . . . .
2.2 General method to solve linear oscillator equations
2.2.1 Homogeneous equations . . . . . . . . . . .
2.2.2 Inhomogeneous equations . . . . . . . . . .
2.3 Linear oscillators . . . . . . . . . . . . . . . . . . .
2.3.1 harmonic oscillators . . . . . . . . . . . . . .
2.3.2 Damping . . . . . . . . . . . . . . . . . . . .
2.3.3 Resonance . . . . . . . . . . . . . . . . . . .
2.3.4 Damping plus forcing . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
17
17
17
18
18
19
19
20
20
21
22
23
.
.
.
.
27
27
27
28
29
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
CONTENTS
3.2
1
33
37
38
CONTENTS
Chapter 1
Introduction
In this course, we have two important classes to model via differential equations.
population dynamics in biology
dynamics in classical mechanics
The first one studies the growth behavior of population of species. It can also be
applied to physical mixing, chemical reactions, etc. The second one include many
important examples such as harmonic oscillators, pendulum, Kepler problems, electric
cricuit, etc. Basic physical laws such as growth rate, conservation laws, etc. for
modeling will be introduced.
The goal is to learn (i) how to do modeling, (ii) how to solve the corresponding
differential equations, (iii) how to interprete the solutions, and (iv) how to develop
general theory.
1.1
1.1.1
Let y(t) be the population (say European population in U.S.) at time t. The census
data are from 1790-2000 (every 10 years). We can build a model based on the following
hypothesis:
dy
= births deaths + migration.
(1.1)
dt
It is natural to assume that the births and the deaths are propotition to the population. Let us neglect the migration for the moment. In terms of mathematical
equations, this reads
dy
= ry
(1.2)
dt
1
CHAPTER 1. INTRODUCTION
where r is called the net growth rate, which is the natural growth rate minus the
death rate. We should have r > 0 if the population is growing. We can set the initial
value
y(0) = y0 ,
(1.3)
the population at year 1790. With (1.2) and (1.3), we can find its solution
y(t) = y0 ert .
We can find the growth rate r by fitting the data, say the census at year 1800. This
yields that r = 0.03067. We find it fits very well until 1820. From then on, the
discrepancy becomes larger and larger. It suggests that
the growth rate r is treated as a constant is only valid local in time;
environmental limitation is not taken into account.
1.1.2
One critism of the simple growth model is that it does not take the limit of environment into consideration. With this consideration, we should expect that there is a
environmental carrying capacity K such that
when y < K, the rate y 0 > 0,
when y > K, the rate y 0 < 0.
A simple model with these considerations is the following
y
0
.
y = ry 1
K
(1.4)
1.1.3
1
y 0 (t) dt = rt + C.
y(1 y/K)
1
1
+
y K y
dy = rt + C
1
1
+
dy = rt + C
y K y
y
= rt + C.
log
K y
y
1 rt
K y = C1 e .
Z
1.1.4
Another qualitative method to study the logistic equation is by using the phase portrait analysis. Let us try to plot all solutions on the t-y plane. Two important
observations are
For any point (t0 , y0 ), there is a solution y() passing through (t0 , y0 ). In other
words, y(t0 ) = y0 .
No more than one solution can pass through (t0 , y0 ).
CHAPTER 1. INTRODUCTION
They are the existence and uniqueness of the ODE. With these, we can use the
equilibria to classify our general solutions.
The first step is to find all equilibria of this system. Let us denote the right-hand
side of (1.4) by f (y), i.e.
y
f (y) = ry 1
.
K
They are the constant states, i.e. they satisfy
f (y) = 0.
In our case, the equilibria are y(t) 0 and y(t) K.
The second step is to classify all other solutions. On thet-y plane, we first draw the
above two constant solutions. Now, by the uniqueness, no solution can pass through
these two constant solution. Therefore, the y-space is naturally partitioned into three
regions
I1 = (, 0), I2 = (0, K), I3 = (K, ).
If y(0) I` , then the corresponding y(t) stays in I` for all t.
The third step is to characterize all solutions in each regions. For any solution in
I2 , we claim that y(t) K as t . From f (y) > 0 in I2 and f (y) < 0 in I1 I3 ,
we can conclude that
y() is increasing in I1 and decreasing in I1 or I3 .
We claim that y(t) K as t for any solution in region I2 . Indeed, y(t) is
increasing and has an upper bound K. By the monotone convergence property of R,
y(t) has a limit as t tends to infinity. Let us call this limit y. We claim that y = K.
Indeed, we have that y 0 (t) 0 as t (try to justify by yourself). If so, then
f (
y ) = lim f (y(t)) = lim y 0 (t) = 0.
t
Since f (y) = 0 has only two roots (0 and K), and since 0 < y K, we conclude that
y = K. Similarly, we can show that y(t) 0 as t . To summarize, we have
the following theorem.
Theorem 1.1 All solutions of (1.4) are classified into the follows.
1. equilibria: y(t) 0 and y(t) K;
2. If y(0) I1 , then limt y(t) = 0 and limt y(t) = ;
3. If y(0) I2 , then limt y(t) = 0 and limt y(t) = K;
4. If y(0) I3 , then limt y(t) = and limt y(t) = K;
The biological interpretation is the follows.
If y(0) < K, then y(t) will increase to a saturated population K as t .
If y(0) > K, , then y(t) will decrease to the saturated population K as t .
y(t) K is the stable equilibrium, whereas y(t) 0 is an unstable equilibrium.
1.1.5
Suppose migration is considered. Let e be the migration rate. We should modify the
model by
y
y 0 = ry 1
ey.
(1.5)
K
The migration rate e can be positive (migrate out) or negative (migrate in).
This model is often accepted in ecology for harvesting a renewable resources such
as shrimps, fishes, plants, etc. In this case, e > 0 is the harvesting rate which measures
the harvesting effort. The quantity ey is the amount from harvesting per unit time.
It is called the harvesting yield per unit time.
This harvesting model is still a logistic equation
ry
0
y = (r e)y 1
(1.6)
(r e)K
with new growth rate r e. The new equilibrium is
e
,
Kh = K 1
r
which is the sustained population. When e < r, we have 0 < Kh < K. This
means that the saturated population Kh decreases due to harvesting. When e > r,
then the species will be extinct due to overharvesting. Indeed, you can check that
y(t) 0 is the stable equilibrium and y(t) Kh is the unstable equilibrium now.
The quantity Y (e) = eKh is called the sustained harvesting yield. An ecological goal
is to maximize this sustained harvesting yield at minimal harvesting effort. We see
that the maximum occurs at e = r/2. The corresponding sustained harvesting yield
is
r r K
rK
=
=
.
Y
2
2 2
4
There is another way to model harvesting of natural resources. We may use
harvesting amount C instead of the harvesting rate e as our parameter. The model
now reads
y
0
y = ry 1
C := fC (y).
(1.7)
K
The equilibrium (i.e. fC (y) = 0) occurs at fC (y) = 0. On the C-y plane, fC (y) = 0 is
a parabola. For each C, when C rK/4, there are two solutions:
r
K
K 2 CK
y =
.
2
4
r
For C > rK/4, there is no real solution. On C-y plane, we can draw arrows to
indicate the sign of fC in that region. We conclude that y+ is a stable equilibrium.
We rename it Kh .
To have sustained resource, we need Kh > 0. This is equivalent to
C
rK
.
4
CHAPTER 1. INTRODUCTION
rK
.
4
K + tanh(
(t + C0 ))
r
2K
where
=
p
2K
r
rK(rK 4C), C0 =
arctanh( (2y0 K)).
In additional to the constraint C K/4, we should also require y(0) > 0. Otherwise,
there would be no harvesting at all. This would give another constraint on C. You
may find it by yourself.
Homeworks
B-D-Hs book, pp. 16, 8-12.
Projects
Do any one of the following two projects.
1. B-D-Hs book, pp. 140, Lab. 1.3: Logistic population models with harvesting
2. B-D-Hs book, pp. 143, Lab. 1.5: Modeling the extinction of the passenger
pigeon
1.1.6
(1.8)
y(0) = y0
(1.9)
1.2
First-order equations
(1.10)
y(0) = y0 .
(1.11)
CHAPTER 1. INTRODUCTION
That is, (f, 1) is also normal to the curve. In other words, = const is a solution
of the ODE y 0 = f (t, y) is equivalent to
(t , y ) k (f, 1).
We can consider more general equation called the Phaffian equation:
M (t, y)dt + N (t, y)dy = 0.
(1.12)
1.2.1
Separation of variables
Suppose the function M (t, y) and N (t, y) in (1.12) are separable, ththat is
M (t, y) = f1 (t)f2 (y),
N (t, y) = g1 (t)g2 (y),
Dividing (1.12) by f2 (y)g1 (t), then the Phaffian equation (1.12) becomes
f1 (t)
g2 (t)
dt +
dy = 0.
g1 (t)
f2 (y)
We can integrate it to obtain an integral :
Z
(t, y) :=
f1 (t)
dt +
g1 (t)
g2 (y)
dy.
f2 (y)
Then (t, y) = constant defines a solution implicitly. In this example, 1/f2 (y)g1 (t) is
an integration factor.
Examples
1. y 0 = t/y 2 . This implies y 3 /3 = t2 /2 + C, or
y(t) =
3t2
+k
2
1/3
.
ex
y =1+C 2 .
x
2
10
CHAPTER 1. INTRODUCTION
Examples
The Burgers equation is a prototype equation in continuum mechanics. It has the
form
2
u
ut +
= uxx ,
2 x
where > 0 is a constant and represent the viscosity. A important special solution
is called the traveling wave, which has the form u(x, t) = u(x ct), where c is the
speed of the wave. The wave has two constant end states
u() = u
and travels at constant speed c without changing its shape. These two end states and
the traveling speed c are related by so called Rankine-Hugoniot relation:
u2+ u2
= c(u+ u ).
2
2
(1.13)
To find the traveling wave solution, we plug u(x ct) into the Burgers equation and
get
2 0
u
0
= u00 .
cu +
2
This yields
u2
cu C := f (u),
(1.14)
2
where C is a constant which should is determined by the property of u at infinity.
We require u() = u and u() = u+ . These would imply u0 () = 0 (I dont
give a proof of this at this moment). From these requirements, we find
u0 =
u2
u2
cu = C + cu+ .
2
2
These is exactly the Rankine-Hugoniot relation. Hence,
u2 u2
f (u) =
c(u u ).
2
We notice that the Rankine-Hugoniot condition (1.13) gives
f (u+ ) = 0.
Thus, f (u) has only two zeros u and u+ . Indeed,
1
f (u) = (u u )(u u+ ).
2
Finally, we can use method od separation of variable to integrate (1.14) to find u().
The travel wave is a tanh function. Not all (u , u+ ) are physical admissible. Only
those with u u+ are admissible. This is called the entropy condition in the theory
of shock wave. We shall not discuss here.
11
Homogeneous equations
We consider the equation:
P (x, y) dx + Q(x, y) dy = 0.
Suppose P and Q are homogeneous of degree n. Let v = y/x. We have P (x, xv) =
xn P (1, v) and Q(x, xv) = xn Q(1, v). The equation becomes
(P (1, v) + vQ(1, v)) dx + xQ(1, v)dv = 0.
We can use method of separation of variables:
dv
dx
+
= 0,
R(v)
x
where
R(v) = v +
The solution is
1.2.2
P (1, v)
.
Q(1, v)
dv
= log |x| + C.
R(v)
(1.15)
12
CHAPTER 1. INTRODUCTION
b(s)eA(s) ds
By plugging the initial datum, we obtain C(0) = y(0). Hence, the general solution is
given by
Z t
A(t)
y(t) = y(0)e
+
b(s)eA(s)+A(t) ds.
0
2
y 0 + y = t 1.
t
Let
Z
A(t) =
2 dt
= ln t2
t
13
t4 t3
+ C.
4
3
y(t) =
t C
t2
+ 2.
4
3 t
Hence,
Homework
1. B-D-H: pp. 121, 4, 12, 13
2. Read B-D-H, Sec. 1.9 (pp. 123-129) and do the homework pp. 134, 8,9,10
1.2.3
Bernoulli equation
(1.16)
0
1
y 1n = a(t)y 1n + b(t)
1n
This suggests the following change of variable:
z = y 1n .
Then
z 0 = (1 n)a(t)z + (1 n)b(t)
which can be solved.
Homeworks
(Courant and John, Vol. II, pp. 690) Solve the following equations
1. xy 0 + y = y 2 log x
2. xy 2 (xy 0 + y) = a2
3. (1 x2 )y 0 xy = axy 2 .
(1.17)
14
1.2.4
CHAPTER 1. INTRODUCTION
* Riccati equation
(Courant and John, Vol. II, pp. 690) The Riccati equation reads
y 0 = a(t)y 2 + b(t)y + c(t)
(1.18)
1
.
y y1
* Homeworks
Courant and John, Vol. II, pp. 690, Exercises 48.
1.3
Existence, uniqueness
In this section, we shall state but without proof the existence and uniqueness theorems. We also show some examples and counter-examples regarding to the existence,
uniqueness.
1.3.1
Theorem 3.2 (Local existence theorem) Suppose f (t, y) is continuous in a neighborhood of (t0 , y0 ). Then the initial value problem
y 0 = f (t, y),
y(t0 ) = y0
has a solution y() existing on a small interval (t0 , t0 + ) for some small number
> 0.
This theorem states that there exists an interval (may be small) where a solution does
exist. The solution may not exist for all t. Let us see the following example.
Example
Consider the initial value problem
y0 = y2
y(0) = y0
By the method of separation of variable,
dy
= dt
y2
15
Z
y0
dy
=t
y2
+ y01 = t
y0
y(t) =
.
1 ty0
When y0 < 0, the solution does exist in [0, ). But when y0 > 0, the solution can
only exist in [0, 1/y0 ). The solution blows up when t 1/t0 :
lim y(t) = .
t1/y0
In the local existence theorem, it only states that the solution exists in a small
region. If the solution does have a limit at the end of this small interval, we can
solve the equation again to extend this solution. Eventually, we can find the maximal
interval of existence.
Homeworks
Find the maximal interval of existence for the problems below.
1. y 0 = 1 + y 2 , y(0) = y0
2. y 0 = y 3 , y(0) = y0
3. y 0 = ey , y(0) = y0
4. y 0 = y ln y, y(0) = y0 > 0.
1.3.2
Uniqueness theorem
The initial value problem may not have a unique solution. Let us see the following
problem:
y 0 = y 1/2 , y(0) = 0
By the method of separation of variable,
dy
= dt,
y
Z
dy
= t + C,
y
2 y =t+C
t2
4
16
CHAPTER 1. INTRODUCTION
Chapter 2
Linear Oscillators
In this chapter, we shall study two models for linear second-order equations of the
form:
ay 00 + by 0 + cy = f (t)
They are the linear springs and the electric circuits.
2.1
2.1.1
Consider a mass attached to a spring in one dimension. Let y be its location, and let
y = 0 be its location at rest. The motion of the mass is governed by Newtons force
law. There are three kinds of forces the mass may be exerted.
1. Restoration force. As the mass moves to y, it is exerted a restoration force by
the spring. According to Hooks law, this restoration force is linearly proportional to
y. That is,
Fr = ky
where k is the spring constant. The minus sign indicates that the force is opposite to
the direction of the mass motion.
2. Friction force. Suppose there is a friction. The friction force is proportional
to the velocity with opposite direction. That is
Ff = y 0 ,
where is the damping (or friction) coefficient.
3. External force. The mass may be exerted by the gravitational force, or some
other external force modeled by f (t).
The Newtons law then gives
my 00 = y 0 ky + f (t).
17
(2.1)
18
2.1.2
Electrical Circuit
dQ
1
d2 Q
+
R
+
Q = f (t)
dt2
dt
C
(2.2)
2.2
(2.3)
where a, b, c > 0 are constants. We need to prescribe the following initial data:
y(0) = y0 , y 0 (0) = y1
(2.4)
(2.5)
d
dt
y = 0,
where L(s) = as2 + bs + c. It is important to observe that the solution set of (2.5)
forms a linear space. That is, if y1 () and y2 () are two solutions of (2.5), so are their
2.2.1
Homogeneous equations
We try the solution of the form et . By plugging into the equation, we obtain
d
L
et = L()et = a2 + b + c et = 0.
dt
This leads to
a2 + b + c = 0.
Let 1 , 2 be its two roots. There are two cases.
Case 1. 1 6= 2 . Then we can check that the general solutions for the homogeneous equation have the form
y(t) = C1 e1 t + C2 e2 t .
The constants C1 and C2 are determined by the initial condition (2.4):
C1 + C2 = y0
1 C1 + 2 C2 = y1 .
From 1 6= 2 , we see that C1 and C2 can be solved uniquely.
Case 2. 1 = 2 . We can check that e1 t and te1 t are two independent solutions.
Indeed, from 1 being the double root of L() = 0, we have L(1 ) = 0, and L0 (1 ) = 0.
By plugging te1 t into the equation (2.5), we obtain
d
L
te1 t = L(1 ) te1 t + L0 (1 ) te1 t = 0.
dt
It is clear that e1 t and te1 t are independent.
So, general solution has the form C1 e1 t + C2 te1 t . The constants C1 and C2 are
determined by the initial data. This leads to
C1 = y0
1 C1 + C2 = y1 .
2.2.2
Inhomogeneous equations
20
This is equivalent to
C10 y1 + C20 y2 = 0.
(2.7)
(2.8)
Equations (2.7) and (2.8) give a first-order differential equation for C1 and C2 :
0
C1 (t)
0
1
= (t)
,
(2.9)
C20 (t)
f (t)/a
where
(t) :=
y1 (t) y2 (t)
y10 (t) y20 (t)
.
(2.10)
2.3
Linear oscillators
2.3.1
harmonic oscillators
To understanding the meaning of the solutions, let us first assume there is no damping
term(i.e. friction or resistance). That is
d2 y
d
y = a 2 + cy = 0.
(2.11)
L
dt
dt
We notice that both a, c > 0. Hence the roots
r
r
c
c
, 2 = i
,
1 = i
a
a
are pure imaginary. Let us denote
r
=
c
a
(2.12)
21
(2.13)
where A and 0 are constants. A is called the amplitude and 0 is the initial phase.
They are related to the initial data y0 and y1 by
y0 = A cos(0 ), y1 = A cos(0 ).
This motion is called harmonic oscillation. It is important to note that through a
transformation:
y = cos
the ODE (2.11) is converted to a linear motion with constant speed:
d
d2
= 0,
=
2
dt
dt
(2.14)
2.3.2
Damping
(2.15)
The coefficient b > 0. We recall that the homogeneous equation has two independent
solutions e1 t and e2 t , where
b + b2 4ac
b b2 4ac
1 =
, 2 =
,
2a
2a
are the two roots of L() = 0. We have the following cases.
Case 1. Over damping. When b2 4ac 0, i are real and negative. The two
independent solutions
yi (t) = ei t 0, as t , i = 1, 2.
We call this is damping. It means that there is no oscillation and all solutions decay to
0 exponentially
fast. As a concrete example, consider
y 00 +3y 0 +y = 0. One eigenvalue
is 1 = 3/2+ 5/2. The other is 2 = 3/2 5/2. We see the solution y1 (t) = et
decays slower than y2 (t) := e2 t .
Case 2. Oscillation with damping When b 4ac < 0, we rewrite
1 = + i, 2 = i,
where = b/2a > 0, = 4ac b2 /2a > 0. Then two independent solutions are
y1 (t) = et cos(t), y2 (t) = et sin(t).
22
So, the general solutions for the homogeneous equation oscillate, but their amplitudes
damp to zero exponentially fast with rate b/2a. A simple example is y 00 + y 0 + y = 0.
Case 3. Critical case When b 4ac = 0, the eigenvalue 1 = b/2a is a double
root. In additional to the solution y1 (t) = e1 t , we can check
y2 (t) = te1 t
is another solution. You may check that this solution still decays to zero as t .
Certainly it is slower than y1 (t). A concrete example is y 00 + 2y 0 + y = 0.
Homework
1. B-D-H, pp. 136, 23: Consider the differential equation
y 00 (2 1)y 0 + ( 1)y = 0.
(a) Determine the values of for which all solutions tend to zero as t .
(b) Determine the values of for which all solutions become unbounded as
t .
2. Read B-D-H, pp 202-205, A Swaying Skyscraper.
3. B-D-H, pp. 329, 13, 18, 21, 26.
4. B-D-H, pp. 331, 34, 35, 36.
2.3.3
Resonance
23
where
, l =
2
2
indicate low and high frequencies, respectively; and
h =
(C + A, B) = A(cos(
1 ), sin(1 )), (C A, B) = B(cos(2 ), sin(2 )).
Let us take the case when . In this case,
C=
F0
2)
a(2
is very large, and hence A is very large. We concentrate on the solution y(t) =
A cos(l t1 ) cos(h t). In this solution, we may view A cos(l t1 ) as the amplitude
of the high frequency wave cos(h t). This amplitude itself is a low frequency wave,
which is called the envelope of the solution y(t). We call it the modulation wave. This
phenomenon occurs in acoustics when two tuning forks of nearly equal frequency are
sound simultaneously.
As = , the large amplitude wave indeed generates another solution, see below.
Case 2. = . In this case, we try a special solution of this form:
yp = R t cos(t + 1 ).
By plugging into the equation, we find that
2aR sin(t + 1 ) = F0 cos(t)
Hence,
yp =
F0
t sin(t).
2a
(2.16)
The amplitude of this solution increases linearly in time. Such a phenomenon is called
resonance.
2.3.4
24
Case 1. 1 6= 2
Overdamping case: p2 4q > 0. In this case, 1 and 2 are real and negative. yi (t) = ei t , i = 1, 2 are two independent solutions for the homogeneous
equation.
Underdamping case: p2 4q < 0. In this case,
1 = p + i 2 = p i,
where
=
4q p2 .
The two independent solutions of the homogeneous equations are y1 (t) = ept cos(t)
and y2 (t) = ept sin(t).
To find a special solution for the inhomogeneous equation, we try
yp = C cos(t) + D sin(t).
By plugging into the equation, we find
2 (C cos(t)+D sin(t))+p(C sin(t)+D cos(t))+q(C cos(t)+D sin(t)) = F0 cos(t).
This yields
2 C + pD + qC = F0
2 D pC + qD = 0
This solves C and D:
C = (q 2 )F0 /, D = pF0 /,
where
= (q 2 )2 + p2 2 .
Notice that 6= 0 whenever there is a damping. Let
A :=
C2
D2
F0
=
, 0 = arctan
p
q 2
.
Then
yp = C cos(t) + D sin(t)
= A cos(0 ) cos(t) A sin(0 ) sin(t)
= A cos(t + 0 )
Thus, a special solution is again a cosine function with amplitude A and initial phase
0 . The general solution is
y(t) = A cos(t + 0 ) + C1 e1 t + C2 e2 t .
Notice that y(t) A cos(t + 0 ) as t .
25
Case. 2: 1 = 2
In this case, p2 = 4q and 1 = 2 = p/2. The two independent solutions for the
homogeneous equation are
y1 (t) = ept/2 , y2 (t) = tept/2 .
For the forcing of the form F0 cos(t), we still get the same special solution
yp (t) = A cos(t + 0 ).
The general solution has the form:
y(t) = A cos(t + 0 ) + C1 ept/2 + C2 tept/2 .
We notice that y(t) A cos(t + 0 ) as t .
We also notice that A only when p = 0 (no damping) and q = 2 . This is
the resonance case. Otherwise, there is no resonance. In other word, periodic forcing
with damping implies only synchronization, not resonance.
Homework.
Find a special solution for the following equations
1. B-D-H, pp. 407, 19, 20,21,22,23
2. B-D-H, pp. 408, 27
3. B-D-H, pp 418, 3, 4, 6, 12.
4. y 00 + 2 y = t sin(t)
5. y 00 + 2 y = t3 .
Project.
1. B-D-H: pp. 432, Lab. 4.2 A periodic forced RLC circuit.
26
Chapter 3
2 2 Linear Systems with Constant
Coefficients
3.1
3.1.1
(3.1)
y0 = v
av 0 = bv cy + f
y1 = y2
y20 = y3
..
n0
y
= f (t, y 1 , y 2 , , y n )
In this chapter, we shall study general 22 linear system with constant coefficients:
1 0
1 1
y
a b
y
f
=
+
2
2
y
c d
y
f2
27
(3.2)
We shall study the initial value problem (3.2) with initial condition
y(0) = y0 .
(3.3)
You may take the following biological interpretation to help you to understand the
structure of the equations and the solutions. We consider the evolution of the population of two species. We call them y 1 and y 2 . The rate change of y 1 and y 2 depends
on themselves linearly.
Homework.
1. Find a similarity transformation T such that
a b
0
1
1
T
T =
c d
ad bc a + d
3.1.2
(3.4)
Since the equation is linear in y, we can see the following linearity property of the
solutions. Namely, if y1 and y2 are solutions of (3.4), so does their linear combination:
1 y1 + 2 y2 , where 1 , 2 are any two scalar numbers. Therefore, if S denotes the
set of all solutions of (3.4), then S is a vector space.
In the case of inhomogeneous equation (3.2), suppose we have already known a
particular solution yp , then so is yp + y for any y S. On the other hand, suppose
z is a solution of the inhomogeneous equation:
z0 = Az + f
then z yp satisfies the homogeneous equation (3.4). Hence, z yp = y for some
y S. We conclude that the set of all solutions of the inhomogeneous equation (3.2)
is the affine space
yp + S.
To determine the dimension of the solution, we notice that all solutions are uniquely
determined by their initial data (the uniqueness theorem),
y(0) = y0 R2 .
Hence, S is two dimensional. We conclude this argument by the following theorem.
Theorem 3.1 The solution space S for equation (3.4) is a two-dimensional vector
space. The solution space for equation (3.2)is the affine space yp + S, where yp is a
particular solution of (3.2).
3.1.3
29
In the solution space S, if we can find two independent solutions y1 () and y2 (), then
a general solution has the form
y = C1 y1 + C2 y2 .
Two solutions y1 and y2 are called independent if C1 y1 + C2 y2 = 0 implies C1 =
C2 = 0. This definition is for all t, but we only need to check one point. We have the
following theorem.
Theorem 3.2 Suppose y1 and y2 are solutions of (3.4). If y1 (t0 ) and y2 (t0 ) are
independent in R2 , then y1 (t) and y2 (t) are independent in R2 for all t in the maximal
interval of existence for both y1 and y2 which contains t0 .
Proof. Suppose y1 (t1 ) and y2 (t1 ) are dependent, then there exist constants C1 and
C2 such that
C1 y1 (t1 ) + C2 y2 (t1 ) = 0.
Let y = C1 y1 + C2 y2 . The maximal interval of existence of y include t1 and t0 .
Notice that both y and the zero constant solution have the same value at t1 . By the
uniqueness theorem, y 0. This contradicts to y1 (t0 ) and y2 (t0 ) being independent.
A general solution for (3.4) can be expressed by
y(t) = C1 y1 (t) + C2 y2 (t),
where the coefficients C1 and C2 can be determined from the initial data:
C1 y1 (0) + C2 y2 (0) = y0 .
From the independence of y1 (0) and y2 (0), the coefficients C1 and C2 can be determined uniquely. The matrix (y1 (t), y2 (t)) is called a fundamental solution of (3.4) or
(3.2).
To find a basis in the solution space for (3.4), we try a solution of the form
y(t) = et v, where v R2 is a constant. Plugging into (3.4), we get
vet = Avet .
We find that y(t) = et v is a solution of (3.4) if and only if
Av = v.
(3.5)
That is, is the eigenvalue and v is the corresponding eigenvector. The eigenvalue
satisfies the following characteristic equation
det (I A) = 0.
T + T 2 4D
T T 2 4D
1 =
, 2 =
.
2
2
There are three possibilities for the eigenvalues:
T 2 4D > 0. Then 1 6= 2 and are real.
T 2 4D < 0. Then 1 , 2 and are complex conjugate.
T 2 4D = 0. Then 1 is a double root.
Example 1.
The spring-mass system
my 00 + y 0 + ky = 0,
can be re-expressed in a first-order linear system
y0 = Ay,
with
y=
y
y0
, A=
0
1
k
m
m
.
k
)+
= 0.
m
m
This is exactly the characteristic equation for such that et is a solution of the
second-order equation
y 00 + y 0 + ky = 0.
( +
2 4mk
+ 2 4mk
, 2 =
.
2m
2m
31
y1
y10
y2
y20
1 t
=e
2 t
=e
1
1
1
2
,
.
Example 2.
Consider the matrix
A=
1 1
4 1
.
1
2
.
A=
2
1
4 1
,
The characteristic
equation
is det(I A) = 2 2 = 0. The roots are 1 =
7/2.
r 1
0 r
A=
A=
1 1
1 3
.
1
1
.
This is the only eigenvector. The solution e2t v2 is a solution of the ODE. To find the
other independent solution, we expect that there is a resonant solution te2t in the
direction of v2 . Unfortunately, te2t v2 cannot be a solution unless v2 = 0. Therefore,
we try
y(t) = te2t v2 + e2t v1 ,
for some unknown vector v1 . We plug it into the equation y0 = Ay to find v1 :
y0 = (e2t + 2te2t )v2 + 2e2t v1 ,
33
we obtain
2v2 te2t + (v2 + 2v1 )e2t = A(v2 te2t + v1 e2t )
Using Av2 = 2v2 , we get
(A 2I)v1 = v2 .
That is
1 1
1
1
v1
v2
=
1
1
.
0
1
.
is a solution.
Now, we find two solutions
y1 = e2t v2
y2 = te2t v2 + e2t v1 .
3.1.4
General theory
Let us go back to the discussion for general equation (3.4). There are several possibilities for eigenvalues of A:
Case 1. i are real and there are two independent real eigenvectors v1 and v2 .
Case 2. i are complex conjugate.
Case 3. i are real, 1 = 2 and the corresponding eigenspace is only one
dimension.
Case 1.
i are real and there are two independent real eigenvectors v1 and v2 . The corresponding two independent solutions are
y1 = e1 t v1 , y2 = e2 t v2 .
A general solution has the form
y(t) = C1 y1 (t) + C2 y2 (t)
If the initial data is y0 , then
C1 v1 + C2 v2 = y0 .
Let T be the matrix (v1 , v2 ). Then
C1
C2
= T 1 y0 .
8 11
6 9
.
1
1
11
6
Similarly, we obtain
v2 =
35
Case 2.
i are complex conjugate.
1 = + i, 2 = i.
Since A is real-valued, the corresponding eigenvectors are also complex conjugate:
w1 = u + iv, w2 = u iv.
We have two independent complex-valued solutions: z1 = e1 t w1 and z2 = e2 t w2 .
Since our equation (3.4) has real coefficients, its real-valued solution can be obtained by taking the real part (or pure imaginary part ) of the complex solution. In
fact, suppose z(t) = x(t) + iy(t) is a complex solution of the real-value ODE (3.4).
Then
d
(x(t) + iy(t)) = A (x(t) + iy(t)) .
dt
By taking the real part and the imaginary part, using the fact that A is real, we
obtain
dx
dy
= Ax(t),
= Ay(t)
dt
dt
Hence, both the real part and the imaginary part of z(t) satisfy the equation.
Now, let us take the real part and the imaginary part of one of the above solution:
z1 (t) = et (cos t + i sin t) (u + iv)
Its real part and imaginary part are respectively
y1 (t) = et (cos tu sin tv)
y2 (t) = et (sin tu + cos tv)
The other solution z2 is the complex conjugate of z1 . We extract the same real
solutions from z2 .
You may wonder now whether u and v are independent. Indeed, if v = cu for
some c R, then
A(u + iv) = 1 (u + iv)
gives
A(1 + ic)u = 1 (1 + ic)u
Au = 1 u = ( + i)u
This yields
Au = u, and u = 0,
because A is real. This implies = 0 if u 6= 0. This contradicts to that the eigenvalue
1 has nontrivial imaginary part.
From the independence of u and v, we conclude that y1 and y2 are also independent, and constitute a real basis in the solution space S. A general solution is given
by
y(t) = C1 y1 (t) + C2 y2 (t),
C1
C2
= T 1 y0 .
37
A I
N2 1 N1 1 {0}
Hence, there exists a v1 N2 such that
(A 1 I)v1 = v2 .
The matrix A, as represented in the basis v1 and v2 , has the form
1 1
1
T AT =
0 1
where T = (v1 , v2 ). This is called the Jordan canonical form of A.
The corresponding fundamental solutions are
y2 (t) = e1 t v2 ,
y1 (t) = te1 t v2 + e1 t v1
The general solution has the form
y(t) = C1 y1 (t) + C2 y2 (t).
The stability of the equilibrium, the 0 state lies on the sign of 1 .
1 < 0: the 0 state is a stable equilibrium.
1 > 0: the 0 state is an unstable equilibrium.
1 = 0: the general solution reads
y(t) = C2 tv2 + C1 v1
The 0 state is unstable.
3.1.5
Stability diagram
We can plot a stability diagram on the plane of the two parameters T and D, the
trace and the determinant of A. The eigenvalues of A are
T T 2 4D
T + T 2 4D
.
, 2 =
1 =
2
2
Let := T 2 4D. On the T -D plane, the parabola = 0, the line D = 0 and the
line T = 0 partition the plane into the following regions. The status of the origin is
as the follows.
3.2
We look for three independent solutions of the form et v. By plugging this into the
equation, we find and v have to be the eigenvalue and eigenvector of A:
Av = v.
The eigenvalue satisfies the characteristic equation
det (I A) = 0.
This is a third order equation because we have a 3 3 system. One of its roots must
be real. The other two roots can be both real or complex conjugate. We label the
39
first one by 3 and the other two by 1 and 2 . The corresponding eigenvectors are
denoted by vi , i = 1, 2, 3. It is possible that 1 = 2 . In this case, v1 and v2 are the
vectors to make A in Jordan block. That is
Av1 = 1 v1 + v2
Av2 = 2 v2 .
The general solution is
y(t) = C1 y1 (t) + C2 y2 (t) + C3 y3 (t).
The solution y1 and y2 are found exactly the same way as that in two dimension.
The solution y3 (t) = e3 t v3 . If 3 < 0, then the general solution tends to the plane
spanned by v1 and v2 . Let us denote this plane by < v1 , v2 >. On the other hand,
if 3 > 0, the solution leaves the plane < v1 , v2 >. The origin 0 is asymptotically
stable only when the real part of i
Re i < 0, i = 1, 2, 3.
Example.
Consider
0 0.1 0
0 0.2 .
A= 0
0.4 0
0
The characteristic equation is
3 0.008 = 0.
The roots are
3 = 0.2, 1 = 0.2ei2/3 , 2 = 0.2ei2/3 .
The eigenvectors are
1 i 3
1/2
1 + i 3
v3 = 1 , v1 = 2 i2 3 , v2 = 2 + i2 3 .
1
4
4
d
dt
y(t) = 0,