revised March 17, 2013

1
Concourse 18.03 Lecture #4
The central theme for today is the Input-Response formulation for linear ODEs.
A linear ODE of the form
( ) ( 1)
1 1 0
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
n n
n
x t p t x t p t x t p t x t q t

' + + + + = where
1 1 0
( ), , ( ), ( ), ( )
n
p t p t p t q t

are functions of the independent variable t can be expressed in the form

( ( )) ( ) T x t g t = where T is a linear operator of the form
1
1 1 1 0
( ) ( ) ( )
n n
n n n
d d d
dt dt dt
T p t p t p t

= + + + + . The last
term refers to multiplication by
0
( ) p t . A useful way of formulating such an ODE is to thing of the left-hand-
side as corresponding to the system and the inhomogeneity ( ) q t on the right-hand-side as corresponding to
the input signal or, more simply, the input. The general solution of the ODE is then referred to as the
output signal or response. Some motivating examples are in order.
Banking: If we let ( ) x x t = represent how much money (in dollars) we have in a bank account after t years with
a fixed interest rate I , the simple model for this is
dx
dt
Ix = and we have solved this to get ( )
It
x t Ae = . If we
have only the initial deposit
0
(0) x x = , then well have
0
( )
It
x t x e = . Note, however, that we can write the ODE
as 0
dx
dt
Ix = , a homogeneous 1st order linear ODE. This corresponds to the situation where you make your
deposit and then go home and let the system grow your money without further intervention.
Now lets suppose that you make deposits and withdrawals according to some function ( ) q t (in dollars/year). If
we add this rate into our model, we have ( )
dx
dt
Ix q t = + , or ( )
dx
dt
Ix q t = . Note how this intervention (or input)
corresponds to the inhomogeneity of this linear ODE. The system will carry on as before but will be subject
to the input associated with the deposits and withdrawals. The response to all this internal and external
activity will be the output ( ) x t , i.e. how much money youll have in the bank at any given time.
Newtons Law of Cooling (diffusion): Suppose we have an enclosed space such as a building or a cooler chest
and that the temperature at any given time t in the interior space is measured as some function ( ) x t and that the
initial temperature is
0
(0) x x = . If the outside temperature is given by some function ( ) y t (possibly constant or
possibly variable), then we might expect the interior temperature to change depending on the quality of the
insulation and on the difference between outside and inside temperatures. That is, the rate of change of
temperature might be modeled as ( )
dx
dt
F y x = for some function F. We would expect that when y x > the
temperature would increase, i.e. that 0
dx
dt
> ; when y x < the temperature would decrease, i.e. that 0
dx
dt
< ; and
that when the outside temperatures are the same there would be no change in temperature, i.e. 0
dx
dt
= . The
simplest model for this would be ( )
dx
dt
k y x = for some positive constant k (called the coupling constant) that
depends on the level of insulation. We can rewrite this as
dx
dt
kx ky + = . In this form, we can think of the
homogeneous equation 0
dx
dt
kx + = as representing how this system would be governed if the outside
temperature remained constant (at 0) and the interior temperature gradually rose or fell to that level. The
inhomogeneous equation ( ) ( )
dx
dt
k x t k y t + = would govern how the interior temperature would respond to the
input ( ) y t (or ( ) ky t ) in the case where the outside temperature varied according to some known pattern, e.g.
the sinusoidal temperature change that might be associated with either a day/night cycle or seasonal cycle, or
perhaps some other temperature variation.
Hookes Law: A simple model for a frictionless mass-spring system is given by Hookes Law F kx = where
F represents an applied force, x represents the displacement of the mass from the equilibrium position, and
revised March 17, 2013
2
where k is the spring constant that corresponds to the stiffness of the spring. If we combine this with Newtons
2nd Law that F ma = where m is the mass,
dx
dt
v = is the velocity, and
2
2
dv d x
dt dt
a = = is the acceleration of the
mass, we have ma kx = or
2
2
d x
dt
m kx = . We can rewrite this as
2
2
0
d x k
m
dt
x + = . If these is some friction in the
system, a simple model suggests that this friction would grow proportionally to the velocity, i.e. there would be
an additional force
f
F cv = opposing the motion. The revised equation becomes
2
2
d x
dt
m kx cv = or
2
2
0
d x c dx k
m m
dt dt
x + + = . Physicists often favor the dot notation for time derivatives with
dx
dt
x = and
2
2
d x
dt
x = , so
the equation may also be expressed as 0
c k
m m
x x x + + = .
Now imagine that you mess with this spring system by driving the system with an additional external force
( ) E t . The model might then look like
2
2
( )
d x
dt
m kx cv E t = + and if we write ( ) ( ) E t mq t = for simplicity, the
ODE becomes
2
2
( )
( )
E t
d x c dx k
m m m
dt dt
x q t + + = = or ( )
c k
m m
x x x q t + + = . Once again, the inhomogeneity
corresponds to the external input imposed on the system, and the homogeneous ODE would govern how the
system would evolve without this intervention.
As weve seen previously, a good approach to solving all of these linear ODEs is to use linear methods that
involve finding all homogeneous solutions (the system), finding one particular solution, and combining these to
determine the overall response ( ) x t .
Example (diffusion): Suppose a closed container has an initial interior temperature of 32F at 10:00am and that
the outside temperature (also in F) rises steadily according to ( ) 60 6 y t t = + where time t is measured in hours.
Further suppose that Newtons Law of Cooling applies where the coupling constant is
1
3
k = . (a) How will the
interior temperature vary in time, and (b) at what time will the interior temperature reach 60F?
Solution: The temperature will be governed by
1
3
( )
dx
dt
y x = or
1 1 1
3 3 3
( ) (60 6 ) 20 2
dx
dt
x y t t t + = = + = + , so the
inhomogeneous ODE is
1
3
20 2
dx
dt
x t + = + . This can be solved using an integrating factor, but lets use linearity.
(1) The homogeneous equation
1
3
0
dx
dt
x + = easily yields the solutions of the form
1
3
( )
t
h
x t ce

= . Its worth
noting that over time any such homogeneous solution will tend toward 0 and become negligible. For this reason
we often refer to this as a transient. In the short term it may be relevant, but in the long term it is not.
(2) We can use undetermined coefficients to find a particular solution. The nature of the inhomogeneity
( ) 20 2 q t t = + suggests that we seek a solution of the form ( )
p
x t A Bt = + . We have ( )
p
dx
dt
t B = , so we must
have
1 1 1
3 3 3
( ) ( ) 20 2 B A Bt B A Bt t + + = + + = + , so
1
3
20 B A + = and
1
3
2 B = . This gives 6 B = and 42 A = , so
( ) 42 6
p
x t t = + . Once the transients have become negligible, this is all that will remain. For this reason we
might refer to this as the steady state solution.
(3) The general solution is
1
3
( ) ( ) ( ) 42 6
t
h p
x t x t x t ce t

= + = + + . If we substitute the initial condition (0) 32 x = ,

we have (0) 42 32 x c = + = , so 10 c = and
1
3
( ) 42 6 10
t
x t t e

= + . Note that eventually the interior

temperature will be rising at the same rate as the outside temperature but always 18F cooler.
The interior temperature will reach 60F at a time T when
1
3
42 6 10 60
T
T e

+ = or
1
3
6 10 18
T
T e

= . This
cannot be solved algebraically, but its easy to get a numerical solution using a graphing calculator and the trace
function. It gives a time 3.33 3 hrs, 20 min T ~ ~ , i.e. about 1:20pm.
Example #2 (exponential input): Solve the initial value problem
2
2
3 2 , (0) 4, (0) 2
t
d x dx
dt dt
x e x x' + + = = = .
revised March 17, 2013
3
Solution: This ODE is of the type we might expect from a mass-spring system, though the external driving
force is not especially realistic (relentlessly exponential in a single direction). It is nonetheless good for
illustrating the methods, and the exponential input will be very relevant in the days and weeks to come. For
simplicity, lets write the ODE as 3 2
t
x x x e '' ' + + = .
(1) For the homogeneous solutions, look for exponential solutions
rt
x e = to the equation 3 2 0 x x x '' ' + + = . This
gives
2 2
3 2 ( 3 2) 0
rt rt rt rt
r e re e r r e + + = + + = , so
2
3 2 ( 1)( 2) 0 1, 2 r r r r r r + + = + + = = = . Individual
homogeneous solutions are
1
( )
t
x t e

= and
2
2
( )
t
x t e

= , and by linearity any solution of the form

2
1 2
( )
t t
h
x t c e c e

= + will satisfy the homogeneous ODE. Its not hard to prove that these give all homogeneous
solutions if we think of the 2nd order homogenous linear ODE as a composition of two 1st order linear ODEs
and use the fact that we can always solve such equations. [See if you can complete the argument.]
Note that, in this case, the homogeneous solutions are transient.
(2) Once again, undetermined coefficients provide the simplest way to find a particular solution in this case.
The obvious choice is to try a solution of the form
t
x Ae = . This gives ,
t t
x Ae x Ae ' '' = = , and we get that
1
6
3 2 6
t t t t t
Ae Ae Ae Ae e A + + = = = , so our particular solution is
1
6
( )
t
p
x t e = .
(3) Our general solution is then
2
1 2
1
6
( )
t t t
x t c e c e e

= + + . We compute
2
1 2
1
6
( ) 2
t t t
x t c e c e e

' = + , and the
initial conditions give
1 2 1 2
1 2
1 2 1 2
23 1
6 6 19 17
2 3 1 11
6 6
(0) 4
,
(0) 2 2 2
x c c c c
c c
x c c c c
= + + = + =

= =
` `
' = + = =

) )
. So the unique
solution to the initial value problem is

2
19 17 1
2 3 6
( )
t t t
steady state
transient
x t e e e

= +

.
Example #3 (sinusoidal input): Find the general solution to the ODE 2 cos3
dx
dt
x t + =
Solution: As with all 1st order linear equations, solving using an integrating factor is always an option, though
it could lead to some difficult integration. In this example, the integrating factor is
2t
e which gives
2 2 2 2
2 ( ) cos3
t t t t
dx d
dt dt
e e x e x e t + = = . Integration gives
2 2
( ) cos3
t t
e x t e t C = +
}
and
2 2
( ) cos3
t t
x t e e t C

(
= +

}
.
The integration can be done using integration by parts (twice) and some additional algebra.
If we solve this using linearity:
(1) 2 0
dx
dt
x + = gives the homogeneous solutions
2
( )
t
h
x t ce

=
(2) For a particular solution, try cos3 sin3 x a t b t = + . We calculate 3 cos3 3 sin3 x b t a t ' = , and substitution
gives 2 (2 3 )cos3 ( 3 2 )sin3 cos3 x x a b t a b t t ' + = + + + = , so
2
13 1
13 3
13
2 3 1 2 3 1 2 3 1
3 2 0 3 2 0 3 2 0
a b a a
a b b b
(
+ = ( ( ( ( ( (
= = =
( `
( ( ( ( ( (
+ =
)
(

, so
3 2
13 13
( ) cos3 sin3
p
x t t t = + or
1
13
( ) (2cos3 3sin3)
p
x t t t = + .
(3) The general solution is therefore

2
1
13
( ) (2cos3 3sin3)
t
transient
steady state
x t ce t t

= + +

.


A little trigonometry
revised March 17, 2013
4
Any expression of the form cos sin a t b t e e + actually represents a single sinusoidal curve with frequency e
and an appropriate translation (phase shift), i.e. a function of the form
0
cos( ) A t e | . We can see this quickly
using the sum of angle formula for cosine:
0 0 0
cos( ) cos cos sin sin cos sin A t A t A t a t b t e | e | e | e e = + = +
We must therefore have
0
0
cos
sin
A a
A b
|
|
=
`
=
)
.
This is most easily understood with a right triangle as shown.
From this we see that
2 2
A a b = + and
0
tan
b
a
| = .
a
b
0
|
A

In our example with
1
13
( ) (2cos3 3sin3)
p
x t t t = + we would get
2 2 13 1 1
13 13
13
2 3 A = + = = and
0
3
2
tan| = . This
gives
0
56.31 | ~ or
0
0.9828 | ~ radians. The period of the oscillation would be
2 2
3
t t
e
= .
Next time well look at variation of parameters as another tool for finding particular solutions, and well
introduce complex numbers and related facts to reformulate some of the methods involving exponential and
sinusoidal inputs.


Notes by Robert Winters