Download as pdf or txt
Download as pdf or txt
You are on page 1of 25

MATH202 – Laplace transforms (notes)

You will recall some of the “fancy formulae” for solving DEs (integrating factors, reduction of
order, variation of parameters, Green’s functions1 ). These are especially useful for DEs that
come from physical systems that are subject to some external forcing. A typical equation
may be written as
ϕ(t, y, y 0 , y 00 , . . .) = f (t), t>0
where the multivariable function ϕ encodes all the physical force balances in an equation, and
f is the external forces.

input (drive) physical system output (response)

f (t) modelled by DE for y y(t)

Many of the afforementioned fancy formulae ultimately appear in the form


Z
y(t) = G(t, τ ) f (τ ) dτ.
| {z }
kernel
The kernel G acts a bit like an “inverse matrix” for the DE, and controls the way that the
system responds to forcing: if you know G then the DE can be solved by integrating. In a
transformed domain (that we’ll study below), the function G becomes known as a transfer
function (to engineers) or a propagator (in quantum physics). We’ll learn how to find G.

21 The Laplace transform and DEs


In this section we will consider initial value problems, solving DEs only for t > 0. The technique
automatically includes initial conditions and can handle some wild forcing functions.

We consider a specific integral transform:


Z ∞
Y (s) = K(s, t) y(t) dt, K(s, t) = e−st .
0

It converts differential equations in t into algebraic equations in s (usually solved easily)!


1
With some careful algebra, the method of variation of parameters for y 00 + p(t) y 0 + q(t) y = f (t) for
t ∈ [0, 1] can be written as
Z 1
y(t) = g(t, τ ) f (τ ) dτ
0
where
y2 (t)y1 (τ )
(
y1 (τ )y20 (τ )−y2 (τ )y10 (τ ) if τ < t
g(t, τ ) := y1 (t)y2 (τ )
y1 (τ )y20 (τ )−y2 (τ )y10 (τ ) if τ ≥ t
and y1 , y2 are solutions to the homogeneous equation. The function g is called a Green’s function, and it
allows a direct calculation of the solution to the DE via an integral involving the input forcing function f .

1
transform algebraic equation
DE in t in variable s

solve (easy)
invert transform
solution y(t) Y (s)

Example 21.1 Transform and solve the differential equation


y 0 = y + 4, y(0) = 1. (1)

Solution. We’ll first transform the equation. Starting on the left with integration by parts,
Z ∞ Z ∞
−st 0 −st ∞
e y (t) dt = e y(t) 0 − [e−st ]0 y(t) dt
0 0
Z ∞
−st b
= lim e y(t) 0 − [−s e−st ] y(t) dt
b→∞ 0
Z ∞
 −sb −s 0
e−st y(t) dt

= lim e y(b) − e y(0) + s
b→∞
|0 {z }
Y (s)

= [0 − 1] + s Y (s)
(since e−sb → 0 as b → ∞ when s > 0). Looking carefully, the left-hand-side of our DE (1)
has transformed to
−1 + s Y (s). (2)
On the other hand, the right-hand-side has transform
Z ∞ Z ∞ Z ∞
−st −st
e (y(t) + 4) dt = e y(t) dt + 4 e−st dt
0 0 0

−1 −st
= Y (s) +4 e
s 0
4
= Y (s) − lim e − e−s0
 −sb 
s b→∞
4
= Y (s) − [0 − 1]
s
4
= Y (s) + .
s
The right-hand-side of our DE (1) has transformed to
4
Y (s) + . (3)
s
Putting together (2) and (3):
4 4
−1 + s Y = Y + ⇒ sY − Y = 1 +
s s
4
⇒ (s − 1) Y = 1 +
s
1 4
⇒ Y = + .
s − 1 s (s − 1)
This is great! We have solved for Y . How do we transform back to find y?

2
Definition and notation: the Laplace Transform (L.T.) of a
function f (t) defined for t > 0:
Z ∞
F (s) = L {f (t)} = f (t) e−st dt
0

and
f (t) = L−1 {F (s)} .

Notice that the L.T. of a function f is denoted by F (sometimes, LTs are denoted by putting
a “hat” symbol on: fˆ). The other part of the notation is to use a curly “L” (for “Laplace”).

Example 21.2 For s > 0,


∞ ∞
−1 −st −[0 − 1]
Z
1
L {1} = e−s t 1 dt = e = = .
0 s 0 s s

Note: if s ≤ 0 then the improper integral is divergent, and the LT is not defined.

Example 21.3 (Transform of an exponential) For s > a,


Z ∞ Z ∞ ∞
 at −s t a t 1 (a−s)t [0 − 1] 1
L e = e e dt = e(a−s) t dt = e = = .
0 0 a−s 0 a−s s−a

Example 21.4 (Linearity) For two functions, and constants α, β,


Z ∞
L {α f (t) + β g(t)} = e−s t (α f (t) + β g(t)) dt
0
Z ∞ Z ∞
−s t
= α e f (t) dt + β e−s t g(t) dt
|0 {z } |0 {z }
= α F (s) + β G(s).

Example 21.5 (Rules in combination)

L e−t + 2 e2 t = L e−t + 2 L e2 t
  

1 2
= +
s − (−1) s − 2
1 2
= +
s+1 s−2
(s − 2) 2 (s + 1)
= +
(s − 2)(s + 1) (s − 2)(s + 1)
3s
= 2 .
s − s−2

3
Let’s try to invert a Laplace tranform.
1
Example 21.6 Find f (t) such that L {f (t)} = s (s−1)
.

Solution. We might recognise


1 1
= L ea t

= L {1} and when a = 1.
s s−1
Put another way,    
−1 1 −1 1
L =1 and L = et . (4)
s s−1
We can use these facts by doing a partial fractions decomposition:
1 A B
= + [for unknown A, B]
s(s − 1) s s−1
A (s − 1) + B s
=
s (s − 1)

⇒ A (s − 1) + B s = 1.
Using s = 1 we see B = 1. Using s = 0 we see A(−1) = 1 so A = −1. Hence
1 −1 1
= + .
s(s − 1) s s−1
The linearity of the Laplace Transform extends to inverses, so using (4),
       
−1 1 −1 −1 1 −1 1 −1 1
L =L + = −L +L = −1 + et .
s(s − 1) s s−1 s s−1

Example 21.7 (The DE (1)) By taking the LT of the y 0 = y + 4, y(0) = 1 we were able
to solve
1 4
Y (s) = + .
s − 1 s (s − 1)
The solution of the DE is completed by taking the inverse Laplace Transform:
y(t) = L−1 {Y (s)}
 
−1 1 4
= L +
s − 1 s (s − 1)
   
−1 1 −1 1
= L + 4L
s−1 s (s − 1)
t t
= e + 4 (−1 + e )
= 5 et − 4.

Summary
R∞
The Laplace Transform L {f (t)} = 0 e−st f (t) dt converts a differential equation for y(t)
into an algebraic equation for L {y(t)} = Y (s). After solving with easy algebra, the solution
to the DE is found as y(t) = L−1 {Y (s)}. To use this method, it is essential to be good at
inverting L.

4
22 A library of Laplace transforms
We begin a systematic study by examining which functions have Laplace Transforms.

• A function f : [0, ∞) → R is called piecewise continuous if it is continuous except


at isolated points, where it may have jumps. All one-sided limits limt→a± f (t) exist,
and they may disagree at only finitely many points on any bounded interval. Piecewise
continuous functions can be integrated.

f (t) f (t)
M ec t
f in here

t −M ec t

Left: piecewise continuity. Right: exponential domination.

• A function f is exponentially dominated if there are constants c, M such that

|f (t)| ≤ M ec t .

Exponentially dominated functions do not grow “too quickly”, and allow the improper
integral defining the LT to converge.

Theorem 22.1. If f : [0, ∞) → R is piecewise continuous and exponentially dominated then


the improper integral defining L {f (t)} converges, and the Laplace Transform F (s) exists for
all large enough s.

Proof. Because f is piecewise continuous, it is integrable on every finite interval. Moreover,


Z ∞ Z ∞ Z ∞
−st −st M e−(s−c)b
e f (t) dt ≤ ct
e M e dt = M e−(s−c)t dt = .
b b b s−c

When s > c, the right-hand-side → 0 as b → ∞. Hence the integral converges.

Example 22.1
2
• f (t) = et has no Laplace Transform because it is not exponentially dominated.
1
• f (t) = 1−t
has no Laplace transform because of the bad discontinuity at t = 1.

5
Heaviside, or unit step, function
We will see that it is convenient to define a function that is 0 or “off” up to a certain time and
then 1 or “on” after that time. This will enable us to build up all sorts of weird and wonderful
functions. The function that does this job is called the Heaviside function2 , and is defined by
(
0 t<0
H(t) =
1 t>0
H(t)

To “turn on” at a later time t = a (where a > 0) we use a shifted version of the Heaviside
function
(
0 t<a
H(t − a) =
1 t>a
which looks like
H(t − a)

t
a

We can find the L.T. of this function as follows:


Z ∞
L{H(t − a)} = H(t − a)e−st dt
Zt=0

= e−st dt
t=a

1 −s t
= e
−s t=a
1 h
−st −sa
i 1  e−as
= − lim e − e = − 0 − e−as = .
s t→∞ s s
2
Sometimes called the unit step function, U (t) or u(t).

6
More Laplace transforms
Example 22.2 Calculate F (s) = L {t H(t − 3)}.

Solution. Note that H(t − 3) “turns on” at t = 3, so



0 t<3
t H(t − 3) =
t t ≥ 3.
Hence, using integration by parts,
Z 3 Z ∞
−st
L {t H(t − 3)} = e e−st t dt
0 dt +
0 3
∞ Z ∞
−1 −st −1 −st 0
= 0+ e t − e [t] dt
s 0 3 s
−1   1 ∞ −st
Z
−s 3
= 0−e 3 + e dt
s s 3
3 e−3 s 1
= − 2 [0 − e−3s ]
s s
3 e−3 s e−3 s e−3 s (3 s + 1)
= + 2 = .
s s s2

Example 22.3 Calculate L {t2 }.

Solution. Using integration by parts (twice)


Z ∞
e−st t2 dt
2
L t =
0
−1 −st 2 ∞ −1 ∞ −st 2 0
Z
= e t |0 − e [t ] dt
s s 0
2 ∞ −st
Z
= 0+ e t dt
s 0
−1 −st ∞ 2 ∞ −st
Z
= e t|0 − 2 e dt
s2 s 0
2
= .
s3

n!
In general, L {tn } = sn+1
.

Example 22.4
1 
L{eikt } − L{e−ikt }

L {sin k t} =
2i  
1 1 1
= −
2i s − ik s + ik
1 (s + ik) − (s − ik) 1 2ik k
= = 2 2
= 2 .
2i (s − ik)(s + ik) 2i s + k s + k2

7
Laplace Transform of a derivative
To transform an ODE into an algebraic equation we also need to know how to transform the
derivatives in the ODE.

Theorem 22.2. Let f be (piecewise) differentiable and exponentially dominated. Then

L {f 0 (t)} = sL {f (t)} − f (0).

Proof. Using the definition of the L.T., and integration by parts, we have:
Z ∞
0
L{f (t)} = e−st f 0 (t) dt
0 | {z }
v×u0

Z ∞
−st
−se−st f (t) dt

= e f (t) −
| {z } 0 0 | {z }
v×u v 0 ×u
Z ∞
−st
= lim e f (t) −f (0) + s e−st f (t)dt .
t→∞
| {z } |0 {z }
vanishes? familiar!
Now, as long as f (t) does not grow too quickly, the limit is zero for large enough s. In
particular, when f is exponentially dominated, |e−st f (t)| ≤ M e−(s−c)t → 0 for s > c. So
Z ∞
0
L{f (t)} = 0 − f (0) + s f (t)e−st dt
0
= −f (0) + s L {f (t)}

Notice that the L.T. of the derivative of a function f (t) is s times the L.T. of f (t). In other
words, differentiation is being converted to multiplication by s, reinforcing the general principle
that differential equations can be converted into algebraic equations. Notice too that when
we transform the derivative of a function we have to supply an initial value, f (0), for that
function. So initial conditions are incorporated into the algebraic equations automatically.
We can also iterate this rule,

L {f 00 (t)} = L {(f 0 )0 (t)}


= s L {f 0 (t)} − f 0 (0)
= s (s L {f (t)} − f (0)) − f 0 (0)
= s2 L {f (t)} − s f (0) − f 0 (0)
L {f 000 (t)} = L {(f 00 )0 (t)}
= s L {f 00 (y)} − f 00 (0)
= s (s2 L {f (t)} − s f 0 (0) − f (0)) − f 00 (0)
= s3 L {f (t)} − s2 f (0) − s f 0 (0) − f 00 (0)

(and so on).

8
Example 22.5 Find the Laplace transform of cos kt.
1 d
Solution. We know that cos kt = k dt
sin kt. Hence
1 1 s k s
L {cos kt} = L {(sin kt)0 } = (sL {sin kt} − sin kt|s=0 ) = 2
−0= 2 .
k k k s +k s + k2

Example 22.6 Solve the IVP


y 00 + 5 y = sin(2 t), y(0) = 1, y 0 (0) = 0.

Solution. Let L {y(t)} = Y (s) and begin by taking LTs of both sides of the equation:
L {y 00 (t) + 5 y(t)} = s2 Y (s) − s y(0) − y 0 (0) +5 Y (s) = s2 Y (s)−s+5 Y (s) = (s2 +5)Y (s)−s.
| {z }
On the right-hand-side
2 2
L {sin 2t} = = 2 .
s2 +2 2 s +4
The DE thus transforms to
2
(s2 + 5)Y (s) − s = .
s2 +4
Solving:
2 s
Y (s) = + .
(s2 + 4)(s2 + 5) s2 + 5
s
The second term is easy to invert: given that L {cos kt} = s2 +k 2 we always have


 
−1 s
L 2
= cos( a t);
s +a
in this case,

 
s −1
L = cos( 5 t).
s2 + 5
For the other term, we apply partial fractions to seek
2 As + B Cs + D
= 2 + 2 .
(s2 2
+ 4)(s + 5) s +4 s +5
After some algebra, we find A = C = 0, B = 2, D = −2. Hence
2 2 2
= 2 − 2 .
(s2 2
+ 4)(s + 5)) s +4 s +5
Now,
( √ )

   
−1 2 2 2 5 2
L = sin 2t and L−1 = √ L−1 = √ sin( 5 t).
s2 + 4 s2 + 5 5 s2 + 5 5
Putting all these terms together,
2 √ √
y(t) = L−1 {Y (s)} = sin 2t − √ sin( 5 t) + cos( 5 t).
5

9
R∞
Table of Laplace Transforms L {f (t)} = 0 e−st f (t) dt

f (t) = L−1 {F (s)} F (s) = L {f (t)}

1
1 s

n!
tn sn+1

1
eat (s−a)

s
cos(ωt) (s2 +ω 2 )

ω
sin(ωt) (s2 +ω 2 )

s
cosh(at) (s2 −a2 )

a
sinh(at) (s2 −a2 )

s−a
eat cos(ωt) (s−a)2 +ω 2

ω
eat sin(ωt) (s−a)2 +ω 2

δ(t − t0 ) e−st0
Rt G(s)
0
g(τ ) dτ s

Laplace transform rules


1. Linearity L{αf (t) + βg(t)} = αF (s) + βG(s)
2. Derivatives L{f 0 (t)} = sF (s) − f (0)
L{f 00 (t)} = s2 F (s) − sf (0) − f 0 (0)
L{f 000 (t)} = s3 F (s) − s2 f (0) − sf 0 (0) − f 00 (0)
3. First shift L{eat f (t)} = F (s − a)
4. Second shift L{f (t − a)H(t − a)} = e−as F (s), a > 0
Rt
5. Convolution L{f ∗ g} = L{ 0 f (τ )g(t − τ ) dτ } = F (s)G(s)

Note: H(t) is the Heaviside step function

10
23 The shift theorems
Recall the general procedure for using LTs to solve a DE for y(t):

1. Take L {·} of both sides of the DE to get an algebraic relation in Y (s) (refer to table
as needed)
2. Rearrange to solve for Y (s)
3. Express Y (s) as a sum of “known” Laplace Transforms (refer to a table as needed)
Hence, invert to find y(t) = L−1 {Y (s)} = L−1 {known sum}

In broadening the class of inverse transforms that we recognise there are two key effects that
we need to understand:

1. Multiply y(t) by eat


2. Multiply Y (s) by eas

Theorem 23.1 (First shift theorem). If L {f (t)} = F (s) then


L eat f (t) = F (s − a).


Note: This result is called the first shift theorem because it says that the LT of eat f (t) is the
LT of f (t) shifted by a to the right. Put another way, L−1 {F (s − a)} = eat L−1 {F (s)}.

Proof.
Z ∞
at
e−st eat f (t)dt

L e f (t) = by definition
Z0 ∞
= e−(s−a)t f (t)dt
Z0 ∞
= e−σt f (t) dt having let σ = s − a
0
= F (σ)
= F (s − a) .

Example 23.1 We have seen that


k
L {sin (kt)} = .
s2 + k 2
The first shift theorem tells us that
k
L eat sin (kt) =

(5)
(s − a)2 + k 2
We have also seen that
s
L{cos (kt)} =
+ k2 s2
and so in a similar manner the first shift theorem tells us that
s−a
L{eat cos (kt)} = . (6)
(s − a)2 + k 2

11
n o
Example 23.2 (Using the first shift theorem) Find L−1 1
(s+3)4
.

1 1
Solution. First, write (s+3)4 = F (s + 3) where F (s) = s4 . The first shift theorem can be used

with a = −3 once we have calculated L−1 s14 . To do that part, notice from the table that


n!
L {tn } = .
sn+1
1
To attack s4
, take n + 1 = 4. Then

3!
L t3 = 4 ,

s
so
t3
 
1 3 1 1
L t = 4 or = L−1 .
3! s 3! s4
By the shift theorem,

e−3t t3
   
−1 1 −3t −1 1
L =e L = .
(s + 3)4 s4 6

Example 23.3 Find L−1 s+3



s2 +2s+5
.

Solution. Attempting to use partial fractions reveals that the denominator is an irreducible
quadratic. By completing the square,
s+3 s+3 s+1 2
= = + . (7)
s2 + 2s + 5 2
(s + 1) + 4 (s + 1) + 4 (s + 1)2 + 4
2

s 2
This whole expression is F (s + 1) where F (s) = s2 +4
+ s2 +4
. But
     
−1 −1 s 2 −1 s −1 2
L {F (s)} = L 2
+ 2 =L 2
+L 2
= cos 2t + sin 2t.
s +4 s +4 s +4 s +4
(8)
Hence, by the first shift theorem, together with equations (7) and (8),
   
−1 s+3 −1 s+1 2
L = L +
s2 + 2s + 5 (s + 1)2 + 4 (s + 1)2 + 4
= L−1 {F (s + 1)}
= e−t L−1 {F (s)} = e−t (cos 2t + sin 2t) .

12
Second shift (delay) theorem
Suppose that we shift a graph to the right by a time units in the time domain.
f (t) f (t)

t t
a

The new function is written as



0 t<a
= f (t − a) H(t − a).
f (t − a) t ≥ a
Theorem 23.2 (Second shift theorem). If L {f (t)} = F (s) and a > 0 then
L {f (t − a) H(t − a)} = e−as F (s) L−1 e−as F (s) = f (t − a) H(t − a).

or

Proof.
Z ∞
L {(t − a)H(t − a)} = e−st H(t − a)f (t − a) dt [by the definition]
Zt=0

= e−st f (t − a) dt [because H(t − a) = 0 when t < a]
Zt=a

= e−s(x+a) f (x) dx [changing variables x = t − a]
Zx=0

= e−sx e−sa f (x)dx
x=0
Z ∞
−sa
= e e−sx f (x)dx = e−as F (s).
| x=0 {z }


0 20 t < 2
Example 23.4 Solve y + 2y = , y(0) = 0.
10 t ≥ 2

Solution. The forcing function on the right-hand-side is the constant 20, with a step-down of
10 at a = 2. The DE can thus be written as y 0 + 2y = 20 − 10 H(t − 2). Taking LTs
20 − 10e−2s
sY (s) − 0 +2 Y (s) = 20L {1} − 10L {H(t − 2)} = .
| {z } s
Then,  
−2s 10 5 5
Y (s) = (2 − e ) = (2 − e−2s ) − .
s(s + 2) s s+2
But L−1 = 5 − 5e−2t so
5 5
s
− s+2

y(t) = L−1 {y(t)} = 10 − 10e−2t − (5 − 5e−2(t−2) ) H(t − 2).

13
24 Impulsive forcing
Consider forces that act on a system with high intensity for a very short period of time.

f (t)

t
tiny time interval

For example, impact from a hammer, sudden braking, point load on a beam (there, instead of
t, the independent variable would be a spatial variable x).
We can model this type of situation with a pulse that switches on and then off.

δa (t − t0 )

1/a

t
t0 t0 + a

This is switched on by a Heaviside function at t0 and then switched off by a second Heaviside
function at t0 + a a small time later. This basic model is

H(t − t0 ) − H(t − (t0 + a)).

In order to arrange a fixed amount of input, irrespective of a, the function is scaled to ensure
that the area underneath is always 1. This makes
1
δa (t − t0 ) = [H(t − t0 ) − H(t − (t0 + a))] .
a
The idea is that an impulse is modelled by letting a → 0. This seems to be problematic
because this “limit” would appear to assign
1
lim+ δa (t0 ) = lim+ = ∞,
a→0 a→0 a

14
while 0 is obtained everywhere else. As a function, this is a bad idea. Amazingly, in the
Laplace domain, everything works fine.
 
1
L {δa (t − t0 )} = L [H(t − t0 ) − H(t − (t0 + a))]
a
1
= [L {H(t − t0 )} − L {H(t − (t0 + a))}]
a
1 e−st0 e−s(t0 +a)

= −
a s s
−st0 −sa
 
e 1−e
= .
s a

Now
e−st0 1 − e−sa
 
lim L {δa (t − t0 )} = lim+
a→0+ a→0 s a
e−st0 d
[1 − e−sa ]
= lim+ da d
s a→0
da
a
e−st0 se−sa
= lim+
s a→0 1
e−st0
= s = e−st0 .
s

Definition. The Dirac delta δ(t − t0 ) models a unit impulse at t = t0 and satisfies

L {δ(t − t0 )} = e−s t0 .

δa (t − t0 )

t
t0

RNote: δ(t − t0 ) is not really a function, dueR its infinite behaviour at t0 . Moreover, since
∞ ∞
0
δa − t0 ) dt = 1 for every a, we also have 0 δ(t − t0 ) dt = 1 (despite δ(t − t0 ) = 0 when
(t
t 6= t0 ). This strange behaviour means that δ is actually a generalised function, or distribution
or measure. The details of the theory of these types ofR object are beyond Math202. It will,

however, be convenient to know the sifting property: 0 δ(t − t0 ) f (t) dt = f (t0 ).

15
Example 24.1 Solve y 0 + y = δ(t − 1), y(0) = 2.

Solution. Take LT of all parts of the equation:


L {y(t)} = Y (s)
L {y 0 (s)} = s Y (s) − y(0) = s Y (s) − 2
L {δ(t − 1)} = e−s·1 .
Hence,
(s Y (s) − 2) + Y = e−s .
Rearranging,
2 e−s
Y (s) = + .
s+1 s+1
n −t o
To invert, note that L−1 1
= e−t = f (t) so that L−1 s+1
e

s+1
= f (t − 1)H(t − 1). Putting
these together
e−s
     −t 
−1 −1 2 −1 1 −1 e
y(t) = L {Y (s)} = L + = 2L +L = 2 e−t +e−(t−1) H(t−1).
s+1 s+1 s+1 s+1
Drawing a graph,

y(t)

t
t=1

we see exponential decay, with a boost from the impulse at t = 1.

Example 24.2 Solve y 0 = δ(t − 1), y(0) = 0.

Solution. Take LT of all parts of the equation:


L {y(t)} = Y (s)
L {y 0 (s)} = s Y (s) − y(0) = s Y (s) − 0
L {δ(t − 1)} = e−s·1
e−s
so that sY (s) = e−s and hence Y (s) = Inverting
s
.
 −s 
−1 −1 e
y(t) = L {Y (s)} = L = H(t − 1).
s

This last example shows that the δ-function is a kind of derivative for the Heaviside function.
More generally, a δ-function forcing of a first order equation gives a jump discontinuity in the
solution.

16
Example 24.3 Solve y 00 = δ(t − a), y(0) = y 0 (0) = 0.

Solution. Taking Laplace transforms of both sides of the equation gives

s2 Y (s) − s 0 − 0 = e−s a .
e−sa
Since L−1
1
Hence, Y (s) = s2
. = t (applying the powers rule with n = 1),
s2
 −sa  
−1 −1 e 0 t<a
y(t) = L {Y (s)} = L = (t − a) H(t − a) =
s 2 t − a t ≥a

y(t)

t
t=a

The impulse puts a “kink” in this solution of a second order equation.

Example 24.4 Solve y 00 + y = 4 δ(t − 2π), y(0) = y 0 (0) = 0.

Solution. Taking Laplace transforms of both sides of the equation gives

2 −s 2π 4 e−s 2π
s Y (s) + Y (s) = 4 e ⇒ Y (s) = 2 .
s +1
Then,

4 e−s 2π
  
−1 0 t < 2π
y(t) = L = 4 sin(t − 2π)H(t − 2 π) =
s2 + 1 4 sin t t ≥ 2π

y(t)

t
t = 2π

The forcing provides a kick of energy which switches on the natural oscillation of the solution.

Homework: Solve y 00 + y = δ(t − π), y(0) = 0, y 0 (0) = 1.

17
Application: point load on a beam
Thin beams obey the equation

E I y 0000 = applied load

where

• E is the Young’s modulus (essentially a measure of stiffness)

• I is the moment of inertia of a cross-section of the beam

• the independent variable is usually taken as x, referring to length

The end supports may be of several types

a a b
simple, y(a) = y 00 (a) = 0 clamped, y(a) = y 0 (a) = 0 free, y 00 (b) = y 000 (b) = 0

Example 24.5 Find the deflection of a beam of length l if it has simple supports at either
end, and a point load of strength P is applied midway along the beam.

P
y

x
0 l

Solution. This equation can be set up

E I y 0000 = −P δ(x − l/2), y(0) = y 00 (0) = y(l) = y 00 (l) = 0.

As a first move, we can rewrite the equation as


−P
y 0000 = δ(x − l/2) = −k δ(x − l/2)
EI
where k = P/EI. Next, take the Laplace transform in x:
Z ∞
Y (s) = e−sx y(x) dx.
0

This gives
s4 Y (s) − y 000 (0) − sy 00 (0) − s2 y 0 (0) − s3 y(0) = −k e−sl/2 .

18
Notice that this requires initial values (derivative information at 0). However, the problem
that we have been given has some boundary values at 0, and some at l. The best way around
this is to assign variables to the unknown initial values, and then try to solve for them later.
To this end, put y 0 (0) = α and y 000 (0) = β. Since y(0) = y 00 (0) = 0, the transformed DE is:

s4 Y (s) − s2 α − β = −k e−sl/2 .

Solving,
α β k e−sl/2
Y (s) = + − .
s2 s4 s4
Now using the inverse Laplace transform
β

β k αx+ x3 0 ≤ x < l/2
y(x) = α x + x3 − (x − l/2)3 H(x − l/2) = 6
β 3 k 3
6 6 αx+ 6
x − 6 (x − l/2) l/2 < x ≤ l.

Finally, we choose the values of α, β to satisfy the boundary conditions at x = l. We have


β 3 k
y(l) = α l + l − (l/2)3 , y 00 (l) = β l − k(l/2).
6 6
−k l2
Since both conditions are supposed to be 0, we have β = k/2 and α = 16
.

19
25 Convolution and drive–response
Earlier on, it was promised that when the solution of a DE forced by f (t) has a formula of the
type Z
y(t) = G(t, τ ) f (τ ) dτ,

Laplace Transforms would reveal the source of G. We can now make good on this promise.
We have seen numerous operational properties of Laplace Transforms. We can now understand
how the equations respond to forcing. Suppose that the DE is

an y (n) + · · · + a2 y 00 + a1 y + a0 = f (t), y(0) = y 0 (0) = · · · = y (n−1) = 0.

Then the Laplace Transform of the DE becomes

(an sn + · · · a2 s2 + a1 s + a0 ) Y (s) = F (s),


| {z }
polynomial
or
1
Y (s) = F (s).
a sn+ · · · a2 s 2 + a1 s + a0
|n {z }
use partial fractions
We know that the first term can be decomposed with partial fractions, and inverted in an
elementary way. Suppose that
1
G(s) = and g(t) = L−1 {G(s)} .
an sn + · · · a2 s 2 + a1 s + a0
In some sense, g “inverts” the differential equation (certainly G(s) solves in the Laplace
domain). Let’s see how multiplying together in the Laplace domain plays out.
Definition: The convolution of two functions f, g is
Z t
f ∗g = f (τ ) g(t − τ ) dτ.
0

Integrate across here


to calculate f ∗ g

Theorem 25.1 (Convolution). L {f ∗ g} = L {f } L {g}.

20
Proof. First interchange the order of integration:
Z ∞ Z t Z ∞ Z ∞
−st
L {f ∗ g} = e f (τ ) g(t − τ ) dτ dt = e−st f (τ ) g(t − τ ) dt dτ
t=0 τ =0 τ =0 t=τ
R∞ R∞ −st
R∞ −sτ
R∞ 0
But τ =0 t=τ e f (τ ) g(t − τ ) dt dτ = τ =0
e f (τ ) t0 =0
e−st g(t0 ) dτ dt0 = L {f } L {g},
using t0 = t − τ .

Example 25.1 Let f (t) = cos t and g(t) = 1.


Calculate f ∗ g, and verify the convolution formula.

Solution. From the formula,


Z t
f ∗ g(t) = f (τ ) g(t − τ ) dτ
0
Z t
= cos τ 1 dτ
0
= sin τ |tτ =0
= [sin t − sin 0] = sin t.
1
Now, L {f ∗ g} = L {sin t} = s2 +1
. On the other hand,
s 1 1
L {f } L {g} = L {cos t} L {1} = = 2 .
s2 +1 s s +1

The real power of the convolution comes with calculating inverse Laplace Transforms.

Example 25.2 Calculate L−1 s21+1 s−1 1



.

1 1
Solution. Suppose that F (s) = s2 +1
and G(s) = s−1
. We know that
f (t) = L−1 {F (s)} = sin t and g(t) = L−1 {G(s)} = et .
Now,
F (s) G(s) = L {f } L {g} = L {f ∗ g} ,
by the Theorem. Hence,
L−1 {F (s) G(s)} = L−1 {L {f ∗ g}} = f ∗ g.
We thus compute
  Z t
−1 1 1
L 2
= sin τ et−τ dτ
s +1 s−1 0
Z t
= et
sin τ e−τ dτ
0
 t
t −1 −τ
= e e (sin τ + cos τ )
2 τ =0
t 
−e −t 
= e (cos t + sin t) − 1
2
1 t 
= e − cos t − sin t .
2

21
Example 25.3 Solve y 00 + 4y = cos 2t, y(0) = y 0 (0) = 0.

Solution. By taking LTs,


s
(s2 + 4) Y (s) = .
+ 22 s2
Hence,  
1 s 1
Y (s) = 2 =L sin 2t L {cos 2t} .
s + 4 s2 + 4 2
By the convolution theorem
1
y(t) = sin 2t ∗ cos 2t
2Z
1 t
= sin 2 τ cos 2(t − τ ) dτ
2 0
1 t1
Z
= (sin(2τ + 2(t − τ )) + sin(2τ − 2(t − τ ))) dτ [using the product-to-sum formula]
2 0 2
1 t 1 t
Z Z
= sin(2t) dτ + sin(4τ − 2t) dτ
4 0 4 0
 t !
1 1 −1
= t sin 2t + cos(4τ − 2t)
4 4 4 τ =0
 
1 1
= t sin 2t + [cos(2t) − cos(−2t)]
4 4
1
= t sin 2t.
4

Example 25.4 (Drive-response) Consider the more generally driven DE


y 00 + 4 y = f (t), y(0) = y 0 (0) = 0.
If L {f (t)} = F (s) then the Laplace Transform method implies that
 
1 1
Y (s) = F (s) 2 = F (s) L sin 2t .
s +4 2
The solution is therefore
t
sin 2 (t − τ )
Z
1
y(t) = f (t) ∗ sin 2t = f (τ ) dτ.
2 0 2
sin 2 (t−τ )
This is precisely the form of an integral transform against a kernel G(t, τ ) = 2
, allowing
arbitrary forcing to be translated into a solution.

Example
Rt 25.5 (Laplace transform of an integral) Here is a cool trick. Let f (t) =
0
g(τ ) dτ . Notice that we can write
Z t Z t
g(τ ) dτ = g(τ ) × 1 dτ = g ∗ 1.
0 0
nR o
t
Hence, L 0 g(τ ) dτ = L {g ∗ 1} = L {g} L {1} = G(s) 1s . So, just as differentiating
causes an Laplace Transform to be multiplied by s, dividing causes division by s!

22
26 Further applications

Integral equations
An equation of the form Z t
y(t) = f (t) + y(τ ) g(t − τ ) dτ
0
is called a Volterra integral equation. Note that the unknown function y appears in an
integral, rather than differential form.

Example 26.1 Solve the integral equation


Z t
1
y(t) = t2 + e−t − y(τ ) et−τ dτ.
2 0

Solution. A bit of inspection shows that this equation is of the form

y =f −y∗g

where f (t) = 21 t2 + e−t and g(t) = et . Taking Laplace transforms,

Y (s) = F (s) − Y (s) G(s)

where Y (s) = L {y(t)},


 
1 2 −t 1 1 1
G(s) = L et =

F (s) = L t +e = 3+ and .
2 s s+1 s−1
Putting all of this together:
1 1 1
Y = + − Y .
s3 s + 1 s−1
Rearranging,
   
1 1 1 s−1 1 1 1 1 1 1
1+ Y = 3+ ⇒ Y = 3
+ = 3
− 4+ − .
s−1 s s+1 s s s+1 s s s + 1 s(s + 1)
The last term can be decomposed by partial fractions as
1 1 1
− = −
s(s + 1) s+1 s
so that
1 1 2 1
Y (s) = − + − .
s3 s4 s + 1 s
Taking inverse Laplace transforms,

t2 t3
 
−1 1 1 2 1
y(t) = L 3
− 4
+ − = − + 2 e−t − 1.
s s s+1 s 2 6

Rt
Exercise: Solve y 0 (t) = 1 − sin t − 0
y(τ ) dτ , y(0) = 0.

23
Laplace Transforms for systems
Laplace transforms can also be used for systems of equations. We simply need to transform
all the variables.

Example 26.2 Consider the general problem of two mixing tanks that receive an input to
Tank 1, an outflow from Tank 2, and the two tanks are connected by a pipe. All the pipes
have flow rates F , and the tanks have volumes V1 and V2 respectively and are well stirred.
Water flows in at a temperature yin and the temperatures in the tanks are y1 and y2 . The
outflow temperature yout is the same as in Tank 2.

Temp yin
F Volume V1

Temp y1 F

Volume V2

Temp y2 F
yout = y2

The DEs modelling this situation are


dy1 F
= (yin − y1 )
dt V1
dy2 F
= (y1 − y2 )
dt V2
F F
To simplify the notation a bit let a1 = V1
and a2 = V2
. Transforming:

sY1 (s) − y1 (0) = a1 (Yin (s) − Y1 (s))


sY2 (s) − y2 (0) = a2 (Y1 (s) − Y2 (s))

After some reorganisation, this can be written as


    
s + a1 0 Y1 y1 (0) + a1 Yin (s)
= .
−a2 s + a2 Y2 y2 (0)
Suppose that a1 = 0.2, a2 = 0.5, both tanks have initial temperatures at 20◦ and the inflow
temperature is initially at 20◦ but steps up to 50◦ at time t = 1. Then

y1 (0) = 20 = y2 (0) and yin (t) = 20 + 30 H(t − 1).

The Laplace transform is therefore


20 e−s
Yin (s) = + 30 .
s s
When all of this is fed back into the system,
    −s 
s + 0.2 0 Y1 20 + 4+6e
= s .
−0.5 s + 0.5 Y2 20

24
This can now be solved:
4 + 6e−s 4 + 6e−s
 
1 20
Y1 (s) = 20 + = +
s + 0.2 s s + 0.2 s(s + 0.2)

and
1 20 10 2 + 3e−s
Y2 (s) = (20 + 0.5 Y1 (s)) = + + .
s + 0.5 s + 0.5 (s + 0.2)(s + 0.5) s(s + 0.2)(s + 0.5)

Using the partial fractions decompositions

1 10/3 10/3 1 10 50/3 20/3


= − and = − + ,
(s + 0.2)(s + 0.5) s + 0.2 s + 0.5 s(s + 0.2)(s + 0.5) s s + 0.2 s + 0.5

Y2 can be rewritten as
 
20 30 50 20
Y2 (s) = + − + e−s .
s s s + 0.2 s + 0.5

The outflow temperature yout = y2 , and this is obtained as

yout (t) = L−1 {Y2 } (s) = 20 + (30 − 50 e−0.2 (t−1) + 20e−0.5(t−1) ) H(t − 1).

Summing up
The method for solving DEs by Laplace Transforms is always the same:

1. Transform both sides of the equation, incorporating initial data and looking up Tables
as required.

2. Rearrange the algebraic equation to find Y (s).

3. Invert the Laplace Transform to find y(t) = L−1 {Y (s)}.

The method can handle quite wild forcing functions (steps, and impulses), and works for
systems. Step 3 may be tricky, requiring

• partial fractions decompositions

• recognising certain functions from tables

• shift theorems

• convolutions

25

You might also like