Lecture 19

Linear ODE’s of higher order

Solutions can be obtained by solving linear system of first order ODE’s.
Idea: Replace one equation by many equations, thereby reducing the
order of the ODE to just first order.
n ≥ 1: A n-th order linear ODE is of the form

dn u dn−1 u dn−2 u du
n
+ a1 n−1
+ a 2 n−2
+ · · · + an−1 + an u = f (x) (*)
dx dx dx dx
Here the coefficients aj could depend on x. We say that (∗) has constant
coefficients iff each aj is independent of x.
(∗) is homogeneous iff (x) = 0. First, look at the homogeneous case:
y1
 y2 
 
Define a vector y =  ..  in n-space where
.
yn

du d2 u dn−1 u
y1 = u, y2 = , y3 = 2 , . . . , yn = n−1 .
dx dx dx
Then (∗) can be rewritten suggestively as

dy1
= y2
dx
dy2
= y3
dx2
..
.
dyn−1
= yn
dxn−1
dyn
= −an y1 − an−1 y2 . . . . . . − a1 y n
dxn
This can be viewed as a matrix of linear equations, and so gives rise to the

1
linear system:
 
0 1 0
 0 1 
 
d  · · 
y = Ay, where A=


 (**)
dx  · · 
 0 0 1 
−an · · · −a2 −a1

Whether or not A has constant coefficients, a solution matrix is given by

∫x
A(t)dt
Φ(x) = e 0 ,

as long as A is integrable as a function of x, which is the case, for example,

when A is continuous. (If A is not defined at 0, then we can take some other
lower limit where A is defined; the effect of changing the lower limit is to
multiply Φ by a constant.)
∫x
Of course, when A is a constant matrix, 0 A(t)dt is just Ax, and we get
the more familiar expression

Φ(x) = eAx .
( )
If Φ(x) = y(1) y(2) . . . y(n) , then the y(j) are the different solution vec-
tors:  (j) 
y1
y (j) 
 
y(j) =  1. 
 .. 
(j)
yn

What we seek is a set of independent solutions to the original ODE (∗).

(j)
(j) dy1 (j)
If A has constant coefficients, then for every j ≤ n, y2 = dx
, . . . , yn =
(j)
dyn−1
dx
.
Useful fact 1: If A is a constant matrix with n distinct eigenvalues
λ1 , λ2 , . . . λn with corresponding eigenvectors v1 , v2 , . . . vn , then a basis of
solution for (∗∗) is given by

y(1) = v1 eλt , y(2) = v2 eλ2 t , . . . , y(n) = vn eλt .

2
  (j)

v1
 
If v(j) =  ... , then a basis of solutions of the n-th order ODE (∗) is given
(j)
vn
by
(1) (2) (n)
v1 eλ1 t , v1 eλ2 t , . . . , v1 eλn t

Useful fact 2: Recall that The eigenvalues of A are determined by

solving the characteristic
( equation:
) |λ − A|
( = 0. )
0 1 λ −t
n = 2: A = , λ−A=
−a2 −a2 a2 λ + a1

⇒ |λ − A| = λ2 + a1 λ + a2

General formula:

|λ − A| = λ2 = λn + a1 λn−1 + · · · + an−1 λ + an

As we started with the nth order linear, homogeneous ODE:

d2 y dn−1 u du
+ a1 n−1 + . . . an−1 + an u = 0,
dx dx dx
we can read off the characteristic equation directly from the ODE.

2nd order linear ODE’s

Many ODE’s occurring in nature are of second order, and it is essential
to understand them thoroughly, at a minimum when they are linear with
constant coefficients.
A prime example (which contains all the subtleties) is the base case of
a “mass on a spring”, where u = displacement, t = time, and k = spring
constant:
d2 u du
m = mg − k(L + u) − γ + Fe (t)
dt2 dt
|{z} | {z }
external force
damping force

Dividing by the mass m > 0, and noting that mg = kL by Hooke’s Law, we

may rewrite this as
d2 u γ du k
2
+ + u = Fe (t),
dt m dt m
3
which is a linear ODE with constant coefficients. It is homogeneous iff there
is no external force, i.e., Fext (t) = 0.
This spring vibrates, and we want to find an expression for u as a function
of t. If Fe (t) = 0, we say that we have free vibrations.
One says the vibrations are undamped if γ = 0, i.e., there is no damping
force.
Undamped free vibrations:
√
u′′ (t) + k
m
u(t) = 0, ω0 = k
m
: frequency
2 k
Characteristic equation: λ + = 0 m
We have two distinct eigenvalues
√
k
λ± = ±iω0 , ω0 = ,
m
both of which are non-real. (They are complex conjugates of each other.)
The corresponding linear system is
( )
dy u
= Ay, y = ,
dt u′

with
( )
0 1
A=
−mk
0

Eigenvectors: We want to find v± ̸= 0 st

( )
± y1 ± ±
Av = ±iω0 v , v
y2
( ) ( )
y2 ±iω0 y1
So need = ⇔ y2 = ±iω0 y1 .
−m
k
y1 ±iω0 y2
We can take ( ) ( )
+ 1 − 1
v = , v = .
iω0 −iω0
Hence a basis of solutions of the linear system is given by
( ) ( )
1 1
iω0 t
e , e−iω0 t ,
iω0 −iω0

4
implying
( that
) a basis of complex solutions of the second order ODE is given
1
by e−iω0 t is
−iω0
z1 = eiω0 t , z2 = e−iω0 t .
The general complex solution is

C1 eiω0 t + C2 e−iω0 t ,

for arbitrary (complex) constants C1 , C2 .

What we want, however, are real solutions!
To get a basis of real solutions,we look at the real and imaginary parts of
z1 , z2 . Since z2 is the complex conjugate of z1 , we need only look at z1 . Thus
a basis of real solutions is given by

cos t, sin t.

Hence the general real solution is

u(t) = A cos(ω0 t) + B sin(ω0 t), A, B ∈ R.

It is often convenient to express this as

u(t) = R cos(ω0 t − δ),

with R denoting the amplitude and δ denoting the phase.

5
Lecture 20
Second Order Linear ODE’s with constant coefficients
Typical example: Equation of “mass on a spring”

u′′ (t) + au′ (t) +bu(t) = Fe (t) ,

| {z } | {z }
dampingforce “external force′′

where
γ k
a= , b= >0
m m
In the process of analyzing (∗), we will introduce some key basic terms which
are also used in many other contexts. such as vibrations (free or forced),
frequency, period, phase, quasi-frequency, quasi period, damping, amplitude,
and resonance.

I Undamped Free Oscillations

“Vibration”: trajectory of a solution to (∗)
“forced” means Fe ̸= 0
“free” means Fe = 0 ⇔ (∗) is homogeneous
γ = damping constant
“undamped” means γ = 0
So the ODE (∗) becomes in this case:

k
u′′ + u = 0,
m
with the associated linear system being
( ) ( )
′ u 0 −1
y = Ay, y = , A=
u′ −w 0
k
The characteristic equation is λ2 + m = 0, so the eigenvalues are λ± =
±iω0 .√This is a case of distinct eigenvalues. It’s common to write
k
ω0 = m
: “frequency of vibrations”.

6
We discussed this case thoroughly last time, and found that a basis of
complex solutions of the linear system (∗∗) is
( iω t ) ( )
(1) e 0 (2) e−iω0 t
z = , z =
iω0 eiω0 t −iω0 e−iω0 t

Let z1 , z2 be the first entries of z (1) , z (2) respectively. Since the first
coordinate of y is u, it follows that a basis of (complex) solutions of
(∗) is given by
−iω0 t
z1 = eiω0 t
|{z} , z2 = |e {z } = z1 .
cos(ω0 t)+i sin(ω0 t) cos(ω0 t)−i sin(ω0 t)

A basis of real solutions to (∗) is given by

x1 = cos(ω0 t), y1 = sin(ω0 t),

which are the real and imaginary parts of z1 .

Hence the general real solution to (∗) is

u(t) = b1 cos(ω0 t) + b2 sin(ω0 t).

A better (more geometric) way to write the general real solution can
be expressed as
u(t) = R cos(ω0 t − δ)
where R is the amplitude, and δ the “phase” of the vibration.
Recall from Trigonometry:

cos(θ1 − θ2 ) = cos θ1 cos θ2 + sin θ1 sin θ2 ,

so that

R cos(ω0 t − δ) = R cos(δ) cos(ω0 t) + R sin(δ) sin(ω0 t).

This allows us to go back and forth between the two expressions for
the general real solution u(t).

7
II Damped free oscillations

γ ′ k
u′′ + u + u=0 (*)
m m
γ k
Characteristic equation: λ2 + m λ+ m =0
√
γ2
Eigenvalues: λ± = − 2m
γ
± 12 m 2 − m
4k

There three cases, depending on whether the eigenvalues are real and
distinct, or non-real (and complex conjugates of each other), or real
and repeated.

Case (i): γ 2 > 4km

Here λ± ∈ R with λ+ ̸= λ− .
General solution:
u(t) = b1 eλ+ t + b2 eλ− t ,
where b1 , b2 are arbitrary real numbers.
Note: 1 − 4km
γ2
< 1, so λ+ and λ− are both negative, so eλ− t → 0 and
eλ− t → 0 as t → ∞.

Case (ii): γ 2 < 4km

Here λ± : not real, with λ− the complex conj of λ+ (since characteristic

equation has real coefficients).
General solution:

u(t) = e−γt/2m (b1 cos(νt) + b2 sin(νt)),

where √
4km − γ 2
ν = >0
2m
If γ = 0, as if there is no damping, this is like I considered earlier.
2π
We call ν the quasi-frequency of vibration, and u
the quasi-period.
We can rewrite the general solution in the better form

u(t) = Re−γt/2m cos(νt − δ)

8
 Even though R is supposed to be the amplitude, it becomes evident by
drawing the trajectory of u(t) that the quantity Re−γt/2m is the true
amplitude (as it includes the effect of damping); it is sometimes called
the “quasi-amplitude.”

Case (iii): γ 2 = 4km (critically damped)

There is a unique eigenvalue λ = − 2m γ

with multiplicity of 2. Recall
( the
from the earlier lectures that ) fundamental solutions of the associ-
u
ated linear system for y = are given by
u′
( λt ) ( λt )
e te
, .
0 eλt

Hence a basis of solutions of the second order ODE (∗) for u is given
by
e−γt/2m , te−γt/2m .
The general solution is
γt
u(t) = (b1 + tb2 )e− 2m .

III Forced Vibrations

Here the external force Fe (t) is non-zero, so the ODE (∗) is inhomoge-
neous:
γ k
u′′ + au′ + bu = Fe (t), a = , b = .
m m
Important special case: Fe (t) is periodic

Example: Fe (t) = F0 cos(wt), w ̸= ω0

√
k
(Recall that ω0 = m .)

Idea: First find the general solution uc (t) of the homogeneous equation
u′′ + au′ + bu = 0, and then add to it any (one) particular solution U (t)
of the inhomogeneous equation.
This procedure in fact gives us all the solutions of the inhomogeneous
equation, and this idea is useful even when Fe is not periodic.

9
General solution:

u(t) = uc (t) + U (t)

|{z} | {z } |{z}
“forced response” transient solution particular solution

When Fe (t) = F0 cos(wt), the choice of U (t) = A cos(wt) + B sin(wt)

works for suitable A and B. (Check this!)
When there’s damping, i.e., γ ̸= 0, Uc (t) → 0 as t → ∞.
Key new effect: Resonance when w ≈ ω0

10
Lecture 21
Basic results on linear, homogeneous ODE’s of higher order

dn u dn−1 u du
n
+ a 1 n−1
+ · · · + an−1 + an u = 0 (*)
dt dt dt
Eigenvalues: Solutions of the characteristic equation λn + a1 λn−1 + . . . an =
0. Let λ1 , λ2 , . . . λk be the distinct roots (in C). For each xj , we get a set of
basic solutions of (∗)

(i) λj real with multiplicity mj ≥ 1. The corresponding solutions are:

tmj −1 λj t
eλj t , teλj t , . . . e .
(mj − 1)!

(ii) λj : complex (not real), λj = αj + iβj , αj , βj ∈ R.

The associated basic real solutions (to (λj , λj ) are eλj = eαj (cos(βj ) +
1 sin(βj ), if mj = 1. For general mj , a basis of real solutions is given
by
tmj − 1 αj tmj − 1 αj
eαj cos(βj ), . . . , e cos(βj ), eαj sin(βj ), . . . , e sin(βj ).
(mj − 1)! (mj − 1)!

We then put together all these solutions, as j ranges from 1 to k, to get a

full basis of real solutions.

Inhomogeneous equation
Example 1:

d2 u du
+ − 2u = e5t ← non-zero external force (*)
dt2 dt
Look at the associated homogeneous ODE:

d2 u du
+ − 2u = 0 (**)
dt2 dt

11
Characteristic equation:

λ2 + λ − 2 = 0 ⇒ (λ + 2)(λ − 1) = 0.

Hence the eigenvalues are

λ1 = −2, λ2 = 1.

Note: The general solution of (∗) is of form

u(t) = u (t) + U (t)

| c{z } |{z}
gen solution of (∗∗) particular solution of (∗)

Indeed, uc (t)+U t) is evidently a solution of (∗), and conversely, if we have two

solutions U1 (t), U2 (t) of (∗), then their difference U1 (t) − U2 (t) is a solution
of (∗∗).
Since λ1 = −2, λ2 = 1, we have

uc (t) = b1 e−2t + b2 et ,

with b1 , b2 ∈ R.
Try
U (t) = ce5t .
Then U ′ = 5ce5t and U ′′ = 25ce5t , so that

U ′′ + U ′ − 2U = c(25 + 5 − 2)e5t = 28ce5t .

This must equal the right hand side of (∗), which is e5t . So U (t) will be a
1
solution of (∗) if we take c = 28 .
So the general solution of (∗) is
1 5t
u(t) = uc (t) + U (t) = b1 e−2t + b2 et + e .
28
Remark: What made this idea work is that e5t is not e−2t or et .
Example 2:
d2 u du
+ − 2u = t2 e5t ,
dt2 dt
where uc (t) is same as in Example 1. Try

U (t) = ϕ(t)e5t ,

12
where ϕ(t) is twice differentiable. Then

U ′ = (ϕ′ (t) + 5ϕ(t))e5t , U ′′ = (ϕ′′ (t) + 5ϕ′ (t) + 5(ϕ′ (t) + 5ϕ(t))) e5t ,

so that

U ′′ + U ′ − 2U = (ϕ′′ (t) + 11ϕ′ (t) + 28ϕ(t)) e5t

?
= t2 e5t

We can get a particular solution U of (∗) this way if we can solve the simpler
ODE
ϕ′′ (t) + 11(ϕ)′ (t) + 28ϕ(t) = t2
We use the Method of Unknown Coefficients:
Try ϕ(t) = a + bt + ct2 , and solve for a, b, c. (Do the calculation!)
Once we solve for ϕ(t), it can be plugged back into the expression for
U (t).

Resonance
Consider

(∗) u′′ + w02 u = A cos(αt),

where ω0 is the frequency of the homogeneous system. If α is close to ω0 ,

then the (true) amplitude of the solution rises tremendously.
Characteristic equation: λ2 + w02 = 0, ω0 > 0
Eigenvalues: λ = iω0 , λ = −iω0
The basic real solutions of homogeneous equation

(∗∗) u′′ + w02 u = 0

are cos(ω0 t), sin(ω0 t), so that the general solution of the homogeneous equa-
tion is
uc (t) = b1 cos(ω0 t) + b2 (ω0 t), b1 , b2 ∈ R,
| {z }
R0 cos(iω0 −δ)

where R is the amplitude and δ the phase (when there is no forcing).

Case (i): α ̸= ω0 (“bounded response”, no “resonance”)
General solution of (∗) is

u(t) = u0 (t) + U (t),

13
with U a particular solution.
Try
U (t) = C cos(αt) + D sin(αt).
In fact, we do not need the sine term here since the first derivative does not
occur in (∗), so we may take D = 0. Then
U ′ = −Cα sin(αt), U ′′ = −Cα2 cos(αt)
⇒ U ′′ + w02 U = c(w2 − α2 ) cos(αt) = A cos(αt)
?

A
⇒C= 2
w − α2
A
⇒ U (t) = R0 cos(ω0 t − δ) + 2 cos(αt)
| {z } w − α2
uc (t)

The second term gets larger as α gets closer to ω0 .

Case (ii): α = ω0 (“unbounded response”, ideal resonance)
u′′ + w02 u = A cos(ω0 t), ω0 > 0.
The basic real solutions of the homogeneous equation
(∗∗) u′′ + w02 u = 0
are cos(ω0 t), sin(ω0 t), so that the general solution of (∗∗) is
uc (t) = b1 cos(ω0 t) + b2 sin(ω0 t), with b1 , b2 ∈ R.
Try
U (t) = ct sin(wt).
Then
U ′ = c sin(ω0 t) + cω0 t cos(ω0 t)
U ′′ = 2cω0 cos(ω0 t) − cw02 t sin(ω0 t)
⇒ U ′′ + w02 U = 2cω0 cos(ω0 t) = A cos(ω0 t)
?

We need: C = 2ωA0 to get a particular solution of (∗). Thus the general

solution of (∗) is
A
⇒ u(t) = R0 cos(wt − δ) + t sin(ω0 t)
| {z } 2ω0
stable | {z }
unstable

14
Quasi-Resonance (under-damping)
For simplicity, let us assume m = 1, and consider
u′′ + γu′ + w02 u = A cos(αt),
with γ > 0 the damping constant. Assume
γ 2 < 4w02 ,
which is the interesting case.
Characteristic equation: λ2 + γλ + ω02 = 0
Eigenvalues:
√
γ γ2
λ1 = − ± − 4ω02
2 | 2 {z }
imaginary since
γ 2 <4w02
γ
⇒ λ± = − + iσ.
2
Now
uc (t) = e−γt/2 (b1 cos(σt) + b2 sin(σt)), b1 , b2 ∈ R.
σ is called the quasi-frequency of the system.
Since u(t) = uc (t) + U (t), we need to find one particular solution U (t).
Suppose we are in the case α ̸= ω0 . Try
U (t) = C cos(αt) + D sin(αt).
Check that we need
A(ω02 − α2 )
C=
(ω02 − α2 ) + (γα)2
Aγα
D= 2
(ω0 − α2 ) + (γα)2
Amplitude: √ A
(ω02 −α2 )+γ 2 α2

Since we are assuming that the damping constant γ is non-zero, we may

also look at when α = ω0 , in which case
A
C = 0, and D = ,
γα

15
so that
A
U (t) = sin(αt).
γα
If we plot this for different values of γ > 0, then we see a narrowing and
simultaneous spiking of the trajectory as γ gets closer and closer to 0, whence
the name “quasi-resonance”.

16