440 Ordinary Differential Equations Chapter 8

9. SOLUTION OF DIFFERENTIAL EQUATIONS BY LAPLACE TRANSFORMS

We are going to discuss the solution of linear differential equations with constant
coefficients (see Sections 5 and 6). Laplace transforms can reduce such an equation
to an algebraic equation and so simplify solving it. Also, since Laplace transforms
automatically use given values of initial conditions, we find immediately a desired
particular solution without the extra step of determining constants to satisfy the
initial conditions. Discontinuous forcing functions are messy to deal with by Section
6 methods; the Laplace transform method handles them easily.
We are going to take Laplace transforms of the terms in differential equations;
to do this we need to know the transforms of derivatives y_ = dy/dt, y__ = d2y/dt2,
etc. To find L(y_), we use the definition (8.1) and integrate by parts, as follows
L(y
_) =
_∞
0
y
_(t)e
−ptdt = e
−pty(t)
____
∞
0
− (−p)
_∞
0
y(t)e
(9.1) −ptdt
= −y(0) + pL(y) = pY − y0
where for simplicity we have written L(y) = Y and y(0) = y0. To find L(y__), we
think of y__ as (y_)_, and substitute y_ for y in (9.1) to get
L(y
__) = pL(y
_) − y
_(0).
Using (9.1) again to eliminate L(y_), we finally have
(9.2) L(y
__) = p2L(y) − py(0) − y
_(0) = p2Y − py0 − y
_
0.
Continuing this process, we obtain the transforms of the higher-order derivatives
(Problem 1 and L35).
We are now ready to solve differential equations. We illustrate the method by
some examples.
Example 1. Solve y__ + 4y_ + 4y = t2e−2t with initial conditions y0 = 0, y_
0 = 0.
We take the Laplace transform of each term in the equation, using L35 and L6
in the table of Laplace transforms. We get
p2Y − py0 − y
_
0 + 4pY − 4y0 + 4Y = L(t2e
−2t) =
2
(p + 2)3 .
But the initial conditions are y0 = y_
0 = 0. Thus we have
(p2 + 4p + 4)Y =
2
(p + 2)3 or Y =
2
(p + 2)5 .
Now we want y, which is the inverse Laplace transform of Y. We look in the table
for the inverse transform of 2/(p+ 2)5. By L6, we get
y=
2t4e−2t
4!
= t4e−2t
12 .
This is much simpler than the general solution; we have obtained just the solution
satisfying the given initial conditions.Section 9 Solution of Differential Equations by Laplace Transforms
441
Example 2. Solve y__ + 4y = sin2t, subject to the initial conditions y0 = 10, y_
0 = 0.
Using the table, take the Laplace transform of each term of the equation to get
p2Y − py0 − y
_
0 + 4Y = L(sin 2t) =
2
p2 + 4.
Then we substitute the initial conditions and solve for Y as follows:
(p2 + 4)Y − 10p =
2
p2 + 4,
Y=
10p
p2 + 4
+
2
(p2 + 4)2 .
Finally, taking the inverse transform using L4 and L17, we have the desired solution:
y = 10cos2t + 1
8 (sin 2t − 2t cos 2t) = 10cos2t + 1
8 sin 2t − 1
4 t cos 2t.
Example 3. Solve y__ + 4y_ + 13y = 20e−t, y0 = 1, y_
0 = 3.
We take the transform of each term and solve for Y as follows:
p2Y − p − 3 + 4pY − 4+ 13Y =
20
p + 1,
Y=
1
p2 + 4p + 13

20
p+1
+p+7
_
= p2 + 8p + 27
(p + 1)(p2 + 4p + 13).
Since this Y is not in our table, we can either use a larger table, or use partial
fractions to split Y into fractions which are in our table (which you can do by
computer) or find the inverse transform by computer. We find:
Y=
2
p+1
+
−p + 1
p2 + 4p + 13
=
2
p+1
+
3
(p + 2)2 + 9
−p+2
(p + 2)2 + 9
and by L2, L13, and L14,
y = 2e
−t + e
−2t sin 3t − e
−2t cos 3t.
Sets of simultaneous differential equations can also be solved by using Laplace
transforms. Here is an example.
Example 4. Solve the set of equations
y
_ − 2y + z = 0,
z
_ − y − 2z = 0,
subject to the initial conditions y0 = 1, z0 = 0.
We shall call L(z) = Z and L(y) = Y as before. We take the Laplace transform
of each of the equations to get
pY − y0 − 2Y + Z = 0,
pZ − z0 − Y − 2Z = 0.
442 Ordinary Differential Equations Chapter 8
After substituting the initial conditions and collecting terms, we have
(p − 2)Y + Z = 1,
Y − (p − 2)Z = 0.
We solve this set of algebraic equations simultaneously for Y and Z (by any of the
methods usually used for a pair of simultaneous equations—elimination, determinants,
etc.). For example, we may multiply the first equation by (p − 2) and add
the second to get
[(p − 2)2 + 1]Y = p −2 or Y = p − 2
(p − 2)2 + 1.
We find y by looking up the inverse transform of Y using L14. We get
y = e2t cos t.
Similarly, solving for Z and looking up the inverse transform, we find
Z=
1
(p − 2)2 + 1, z= e2t sin t
. Alternatively, we could find z from the first differential equation by substituting
the y solution:
z = 2y − y
_ = 2e2t cos t + e2t sin t − 2e2t cos t = e2t sin t.
Solving linear differential equations with constant coefficients is not the only use
of Laplace transforms. As you will see in Chapter 13, Section 10, we may solve some
kinds of partial differential equations by Laplace transforms. Also a table of Laplace
transforms can be used to evaluate definite integrals of the type
_∞
0 e−ptf(t)dt.
Example 5. By L15 with a = 3 and p = 2, we have
_∞
0
e
−2t(1 − cos 3t) dt =
32
2(22 + 32)
=
9
26.
Actually, there is more to the subject than this. Although we are discussing
in this chapter the use of Laplace transforms as a tool, they also can play a more
theoretical role in applied problems. It is often possible to find desired information
about a problem directly from the Laplace transform of the solution without ever
finding the solution. Thus the use of Laplace transforms may lead to a better
understanding of a problem or a simpler method of solution. (Compare the use of
matrices, for example, or the use of Fourier transforms.)
10. CONVOLUTION
In solving differential equations by Laplace transforms in Section 9, we found Y
and then found the inverse transform y either in a table or by computer. We had
no way of writing a formula for y. We now want to consider another way of finding
inverse transforms. (Also see Bromwich integral, Chapter 14, page 696.)
Let us first see why the method we are going to discuss in this section is useful.
Consider differential equations of the kind discussed in Sections 5 and 6, namely
linear second-order equations with constant coefficients. Recall that such equations
describe the vibrations or oscillations of either a mechanical or an electrical system.
If the right-hand side of the equation is a function of t, called the forcing function,
then the differential equation describes forced vibrations.
Example 1. Let us solve the following representative equation by Laplace transforms,
assuming that the system is initially at rest and that the force f(t) starts being
applied at t = 0.
(10.1) Ay
__ + By
_ + Cy = f(t), y0 = y
_
0 = 0.
We take the Laplace transform of each term, substitute the initial conditions, and
solve for Y as follows:
(10.2) Ap2Y + BpY + CY = L(f) = F(p), Y=
1
Ap2 + Bp + C
F(p).
Note that Y is a product of two functions of p. We know the inverse transform
of F(p), namely f(t). The factor T (p) = (Ap2 + Bp + C)−1 (called the transfer
function) can always be written as
T (p) =
1
A(p + a)(p + b)
by factoring the quadratic expression in the denominator. Hence by L7 (or L6
if a = b) we can find the inverse transform of T (p) for any problem. Then y
444
Section 10 Convolution 445
[the inverse transform of Y in (10.2)] is the inverse transform of a product of two
functions whose inverse transforms we know. We are going to show how to write y
as an integral (that is, we are going to verify L34 in the table).
Let G(p) and H(p) be the transforms of g(t) and h(t). We want the inverse
transform of the product G(p)H(p). By the definition (8.1)
(10.3) G(p)H(p) =
_∞
0
e
−ptg(t) dt ·
_∞
0
e
−pth(t) dt.
Let us rewrite (10.3) replacing t by different dummy variables of integration so that
we can write the product of the two integrals as a double integral. We then have
G(p)H(p) =
_∞
0
e
−pσg(σ) dσ ·
_∞
0
e
(10.4) −pτh(τ) dτ
=
_∞
0
_∞
0
e
−p(σ+τ)g(σ)h(τ) dσ dτ.
Now we make a change of variables; in the σ integral (that is, with τ fixed), let
σ+τ = t. Then σ = t−τ, dσ = dt, and the range of integration with respect to t is
from t = τ (corresponding to σ = 0) to t = ∞ (corresponding to σ = ∞). Making
these substitutions into (10.4), we get
(10.5) G(p)H(p) =
_∞
τ=0
_∞
t=τ
e
−ptg(t − τ)h(τ) dt dτ.
Figure 10.1
Next we want to change the order of
integration. From Figure 10.1, we see
that the double integral in (10.5) is over
the triangle in the first quadrant below
the line t = τ. The t integral ranges
from the line t = τ to t = ∞ (indicated
by a horizontal strip of width dτ
from t = τ to ∞) and then the τ integral
sums over the horizontal strips from
τ = 0 to τ = ∞ covering the whole
infinite triangle. Let us integrate with
respect to τ first; τ then ranges from 0
to the line τ = t [indicated by a vertical
strip in Figure 10.1] and then the t integral sums over the vertical strips from t = 0
to ∞. Making this change in (10.5), we get
G(p)H(p) =
_∞
t=0
_t
τ=0
e
(10.6) −ptg(t − τ)h(τ) dτ dt
=
_∞
0
e
−pt
__ t
0
g(t − τ)h(τ) dτ
_
dt
=L
__ t
0
g(t − τ)h(τ) dτ
_
. (See L34.)
The last step follows from the definition (8.1) of a Laplace transform.
446 Ordinary Differential Equations Chapter 8
Definition of Convolution The integral
(10.7)
_t
0
g(t − τ)h(τ) dτ = g ∗ h
is called the convolution of g and h (or the resultant or the Faltung). Note the
abbreviation g ∗ h for the convolution integral, and do not confuse the symbol ∗,
written on the line, with a star used as a superscript meaning complex conjugate.
It is easy to show (Problem 1) that g ∗ h = h ∗ g; this result and (10.6) and (10.7)
give L34 in the table.
Now let’s see how to use (10.6) or L34 to solve the kind of problem indicated in
(10.1) and (10.2).
Example 2. Solve y__ + 3y_ + 2y = e−t, y0 = y_
0 = 0.
Taking the Laplace transform of each term, substituting the initial conditions,
and solving for Y , we get
p2Y + 3pY + 2Y = L(e
−t),
Y=
1
p2 + 3p + 2L(e
−t).
Since we are intending to use the convolution integral, we do not bother to look up
the transform of e−t. We dowant, however, the inverse transform of 1/(p2+3p+2);
by L7, this is e−t − e−2t, so we have
Y = L(e
−t − e
−2t)L(e
−t) = G(p)H(p),
with g(t) = e−t − e−2t and h(t) = e−t. We now use L34 to find y. Observe from
L34 that we may use either g(t − τ)h(τ) or g(τ)h(t − τ) in the integral. It is well
to choose whichever form is easier to integrate; usually it is best to put (t − τ) in
the simpler function [here h(t)]. Then we have
y=
_t
0
g(τ)h(t − τ) dτ =
_t
0
(e
−τ − e
−2τ )(e
−(t−τ ) dτ
=e
−t
_t
0
(1 − e
−τ ) dτ = e
−t(τ + e
−τ )
____
t
0
=e
−t(t + e
−t − 1) = te
−t + e
−2t − e
−t.
It is not always as easy to evaluate the convolution integral as it was in this
example. However, let us observe that, at the very worst, we can always write the
solution to a forced vibrations problem [equation (10.1)] as an integral (which can,
if necessary, be evaluated numerically). This is true because, as we showed just
after (10.2), we can always find the inverse transform of the transfer function T (p),
and so have Y as a product of two functions whose inverse transforms we know.
Then y is given by the convolution (10.7) of the forcing function f(t) and the inverse
transform of the the transfer function. Also note (Problem 16) that a combination
of L6, L7, L8 and L18 will handle any terms arising in a problem with nonzero
initial conditions.
Section 10 Convolution 447
Fourier Transform of a Convolution We have shown that the Laplace transform
of the convolution of two functions is the product of their Laplace transforms.
There is a similar theorem for Fourier transforms; let us see what it says. Let g1(α)
and g2(α) be the Fourier transforms of f1(x) and f2(x). By analogy with equations
(10.3), (10.4), (10.5), and (10.6), we might expect the product g1(α) · g2(α)
to be the Fourier transform of something; l _ et’s investigate this idea. Assuming that ∞
−∞ |f1(x)f2(x)|dx is finite, then by the definition of a Fourier transform [Chapter 7,
equation (12.2)], we have
g1(α) · g2(α) =
1
2π
_∞
−∞
f1(v)e
−iαv dv · 1
2π
_∞
−∞
f2(u)e
(10.8) −iαu du
=

1
2π
_2 _ ∞
−∞
_∞
−∞
e
−iα(v+u)f1(v)f2(u) dv du.
[We have used different dummy integration variables as in (10.4).] Next we make
the change of variables x = v + u, dx = dv, in the v integral, to get
g1(α)g2(α) =

1
2π
_2 _ ∞
−∞
_∞
−∞
e
(10.9) −iαxf1(x − u)f2(u) dx du
=

1
2π
_2 _ ∞
−∞
e
−iαx
__ ∞
−∞
f1(x − u)f2(u) du
_
dx.
If we define the convolution of f1(x) and f2(x) by
(10.10) f1 ∗ f2 =
_∞
−∞
f1(x − u)f2(u) du, †
then (10.9) becomes
(10.11) g1 · g2 =
1
2π
_
1
2π
_∞
−∞
f1 ∗ f2 e
−iαx dx
_
=
1
2π
· Fourier transform of f1 ∗ f2.
In other words,
(10.12) g1 · g2 and
1
2π
f1 ∗ f2 are a pair of Fourier transforms.
Because of the symmetry of the f(x) and g(α) integrals, there is a similar result
relating f1 · f2 and the convolution of g1 and g2. We find that (Problem 19)
(10.13) g1 ∗ g2 and f1 · f2 are a pair of Fourier transforms.
As discussed in Chapter 7, after (12.2) and after (12.10), various references differ
†Note that (10.10) is really the same as (10.7) if we agree that, for Laplace transforms, f(t) = 0
when t < 0 (see the first paragraph of Section 8, page 437). For then in (10.7), h(τ) = 0 for τ < 0
and g(t − τ) = 0 for τ > t, so the integral would not really be different if written with infinite
limits (in fact, it is sometimes written that way).
in the position of the factor 1/(2π). Some authors include factors of 1/(2π) or
1/
√
2π in the convolution definition (10.10); this definition as well as Chapter 7,
equation (12.2), affects (10.12) and (10.13). Check the notation in any reference
you are using.
11. THE DIRAC DELTA FUNCTION
In mechanics we consider the idea of an impulsive force such as a hammer blow
which lasts for a very short time. We usually do not know the exact shape of the
force function f(t), and so we proceed as follows. Let the impulsive force f(t) lasting
from t = t0 till t = t1 be applied to a mass m; then by Newton’s second law we
have
(11.1)
_ t1
t0
f(t) dt =
_ t1
t0
m
dv
dt
dt =
_ v1
v0
mdv = m(v1 − v0).
Figure 11.1
This says that the integral of f(t) [called the impulse of f(t)]
is equal to the change in the momentum of m, and we note
that the result is independent of the shape of f(t) but depends
only on the area under the f(t) curve. If this area is 1,
we call the impulse a unit impulse. If t1−t0 is very small, we
may simply ignore the motion of m during this small time,
and say only that the momentum jumped from mv0 to mv1
during the time t1−t0. If v0 = 0, the graph of the momentum as a function of time
would be as in Figure 11.1, where we have simply omitted the (unknown) part of
the graph between t0 and t1. We note that if t1 − t0 is very small, the graph in
Figure 11.1 is almost the unit step function (L24). Let us imagine making t1 − t0
smaller and smaller while keeping the jump in mv always 1.
In Figures 11.2, 11.3 and 11.4 we have sketched some possible sequences of functions
fn(t) which would do this. We could draw many other similar sets of graphs;
the essential requirement is that f(t) should become taller and narrower (that is,
that the force should become more intense but act over a shorter time) in such a
way that the impulse [area under the f(t) curve] remains 1. We might then consider
the limiting case in which Figure 11.1 has a jump of 1 at t0; the force f(t) required
to produce this result would have to be infinite and act instantaneously. Also from
equation (11.1), we see that the function f(t) is the slope of the mv graph; thus
we are asking for f(t) to be the derivative of a step function at the jump. We see
immediately that no ordinary function has these properties. However, we also note
449