TMA 682 Lecture Notes

M. Asadzadeh and F. Bengzon

15th September 2004

Contents

1 Laplace Transformation 5
1.1 The Laplace Transform . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.2 Existence . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.3 General Properties of the Laplace Transform . . . . . . 8
1.1.4 Table of Laplace Transforms . . . . . . . . . . . . . . . 13
1.2 The Inverse Laplace Transform . . . . . . . . . . . . . . . . . 15
1.2.1 Method of Partial Fractions . . . . . . . . . . . . . . . 16
1.3 Applications of Laplace Transforms . . . . . . . . . . . . . . . 19
1.3.1 Initial Value Problems . . . . . . . . . . . . . . . . . . 19
1.3.2 Integral Equations . . . . . . . . . . . . . . . . . . . . 22

2 Fourier Analysis 25
2.1 Periodic Functions . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2 Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2.2 Complex Representation of Fourier Series . . . . . . . . 30
2.2.3 Derivation of the Euler Formulas . . . . . . . . . . . . 30
2.2.4 Even and Odd Functions . . . . . . . . . . . . . . . . . 32
2.2.5 Bessel Inequality and Riemann-Lebesgue Lemma . . . 35
2.2.6 Convergence of Fourier Series . . . . . . . . . . . . . . 37
2.2.7 Functions of Arbitrary Period . . . . . . . . . . . . . . 39
2.2.8 Sine and Cosine Series . . . . . . . . . . . . . . . . . . 41

3 Partial Differential Equations 45

3.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.1.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.1.2 Linearity and Superposition . . . . . . . . . . . . . . . 46

3
4 CONTENTS

3.1.3 Classification . . . . . . . . . . . . . . . . . . . . . . . 47
3.1.4 Physical Derivation of the Heat Equation . . . . . . . . 49
3.1.5 Boundary and Initial Conditions . . . . . . . . . . . . . 50
3.2 Separation of Variables . . . . . . . . . . . . . . . . . . . . . . 52
3.2.1 Overview of Method . . . . . . . . . . . . . . . . . . . 52
3.2.2 The Heat Equation . . . . . . . . . . . . . . . . . . . . 52
3.2.3 The Wave Equation . . . . . . . . . . . . . . . . . . . . 58
3.2.4 Inhomogeneous Equations . . . . . . . . . . . . . . . . 60

A Answers to Exercises 63
Chapter 1

Laplace Transformation

Laplace transformation is a powerful technique for solving differential equations

with constant coefficients. Areas of application are widespread but traditional
fields include mechanics, electronics, and automatic control engineering.
Before the advent of computers it was a tedious task to multiply numbers
such as 1.4142 and 3.1416. Therefore logarithms were used to transform the
complicted operation of multiplication into the simpler operation of addition via
the formula
log(1.4142 · 3.1416) = log(1.4141) + log(3.1416).
By consulting tables of precomputed logarithms and exponentials one obtained
the result 4.4429. Roughly speaking, Laplace transformation works analoguosly
and reduces problems of calculus into simple algebraic problems via tables and
general properties of the transform.

1.1 The Laplace Transform

1.1.1 Definition
Let f (t) be a function defined for all t ≥ 0. If the improper integral
Z ∞
F (s) = f (t)e−st dt, (1.1.1)
0

converges for any s, then F (s) is said to be the Laplace transform 1 of f (t).
1
Pierre Simon de Laplace (1749-1827) French mathematician.

5
6 CHAPTER 1. LAPLACE TRANSFORMATION

Example 1. Find the Laplace transform of the Heaviside step function,

(
1, t > 0,
θ(t) = (1.1.2)
0, t < 0.

Solution. By the definition (1.1.1) we get, for s > 0

∞ ∞  ∞
1 1
Z Z
−st −st
F (s) = θ(t)e dt = e dt = − e−st = . (1.1.3)
0 0 s 0 s

A common shorthand notation for the operation of taking the Laplace

transform of a function f (t) is L[f (t)]. For example, L[θ(t)] = 1s .
Since the integral (1.1.1) has the limits 0 and ∞, it follows that F (s) is
not influenced by f (t) when t < 0. As a result, if f1 and f2 are two functions
such that f1 = f2 for t ≥ 0, then these functions have the same Laplace
transform, even if they differ for t < 0. Because of this ambiguity, we shall
henceforth always assume that f (t) is causal, which is to say, f (t) = 0 for all
t < 0.
If f (t) is not causal to begin with, we can always force it to become so
by multiplying it with the Heaviside step function θ(t) (1.1.2). We illustrate
such a case below.
1
f (t)

−4 −2 2 4
−1

1
θ(t)f (t)

−4 −2 2 4
−1

Figure 1.1: A causal restriction of the function f (t).

1.1. THE LAPLACE TRANSFORM 7

Example 2. Find the Laplace transform of f (t) = ect , where c is a constant.

Solution. Again, by (1.1.1) we get

∞  ∞
1 1
Z
ct −st −(s−c)t
F (s) = e e dt = − e = . (1.1.4)
0 s−c 0 s−c

Note that, for the above integral to converge, we must assume s > c.

1.1.2 Existence
Not any function f (t) have a Laplace transform L[f (t)]. For example, it is
2
easy to see that L[et ] does not exist, since its associated integral diverges
as t → ∞. As a rule, f (t) must be of exponential order to have a Laplace
transform. By this we mean that there must exist a constant, say a, such
that
lim |f (t)e−at | = 0. (1.1.5)
t→∞

If this indeed is the case, then by choosing s > a, we see that the integrand
f (t)e−st of (1.1.1) goes to zero as t tend to infinity and, hence, the integral
for L[f (t)] converges. Let us formalize this result by stating it as a theorem.

Theorem 1. If f (t) is a piecewise continuous2 function for all t ≥ 0, and if

|f (t)| ≤ Meat , (1.1.6)

for some constants a and C, then the Laplace transform F (s) of f (t) exists.

Proof. If |f (t)| ≤ Meat and s > a, then

∞ ∞
M
Z Z
−st
|F (s)| ≤ |f (t)|e dt ≤ Me−(s−a)t dt = . (1.1.7)
0 0 s−a

Hence, if s > a the integral (1.1.1) converges.

2
A function is said to be piecewise continuous if it is discontinuous only at isolated
points, and its left and right limits are defined at each discontinuity point.
8 CHAPTER 1. LAPLACE TRANSFORMATION

1.1.3 General Properties of the Laplace Transform

Theorem 2. Laplace transformation is a linear operation, that is, for any
functions f (t) and g(t) whose Laplace transform exist and any constants a
and b, we have

L[af (t) + bg(t)] = aL[f (t)] + bL[g(t)]. (1.1.8)

Proof. By definition, it holds that

Z ∞
L[f (t) + g(t)] = (af (t) + bg(t))e−st dt
0
Z ∞ Z ∞
−st
=a f (t)e dt + b g(t)e−st dt
0 0
= aL[f (t)] + bL[g(t)]. (1.1.9)

Example 3. Find the Laplace transforms of sinh t = 21 (et − e−t ).

Solution. Using the above linearity and (1.1.4) with c = ±1, we have

L[sinh t] = L[ 21 (et − e−t )] = 21 L[et ] − 21 L[e−t ] = 21 ( s−1

1 1
− s+1 )= 1
s2 −1
. (1.1.10)

Example 4. Find the Laplace transforms of sin ωt and cos ωt.

Solution. If we set c = iω in (1.1.4) then we have

1 s + iω
L[eiωt ] = =
s − iω (s − iω)(s + iω)
s + iω s ω
= 2 2
= 2 2
+i 2 . (1.1.11)
s +ω s +ω s + ω2
On the other hand we also have

L[eiωt ] = L[cos ωt + i sin ωt] = L[cos ωt] + iL[sin ωt]. (1.1.12)

Equating the real and imaginary parts of these two equations, we get
s
L[cos ωt] = , (1.1.13)
s2
+ ω2
ω
L[sin ωt] = 2 . (1.1.14)
s + ω2
1.1. THE LAPLACE TRANSFORM 9

As the last examples show, the definition (1.1.1) is rarely the starting
point for deriving Laplace transforms. Instead, one usually first consults a
table of standard transforms, and then tries to adapt any of these to the
problem at hand using a set of general properties, such as the linearity, of
the Laplace transform. Below, we derive a number of other such properties
and illustrate their use.
Theorem 3 (1st Shifting Rule). If f (t) has the Laplace transform F (s)
then for any constant c, we have

L[ect f (t)] = F (s − c). (1.1.15)

Proof. Inserting ect f (t) directly into the definition (1.1.1) gives, with
s > c,
Z ∞ Z ∞
ct ct −st
L[e f (t)] = e f (t)e dt = f (t)e−(s−c)t dt = F (s − c). (1.1.16)
0 0

Example 5. Find the Laplace transform of 3e−2t cos 5t.

Solution. By the previous example, we have

s
L[cos 5t] = . (1.1.17)
s2 + 25
Applying now the linearity, and the 1st Shifting Rule, we get
3(s + 2) 3s + 6
L[3e−2t cos 5t] = 2
= 2 . (1.1.18)
(s + 2) + 25 s + 4t + 29

Theorem 4 (2nd Shifting Rule). Suppose f (t − T ) is a function that is

zero for t ≤ T , then
L[f (t − T )] = e−T s F (s). (1.1.19)
Proof. Let τ = t − T , then
Z ∞ Z ∞
−st
L[f (t − T )] = f (t − T )e dt = f (t − T )e−st dt
Z0 ∞ −∞
Z ∞
−s(τ +T ) −T s
= f (τ )e dτ = e f (τ )e−sτ dτ
−∞ 0
−T s
=e F (s). (1.1.20)
10 CHAPTER 1. LAPLACE TRANSFORMATION

Introducing a generalized form of the Heaviside step function,

(
1, t > T,
θ(t − T ) = (1.1.21)
0, t < T,

we can state the 2nd Shifting Rule (1.1.19) formally as

L[θ(t − T )f (t − T )] = e−T s F (s). (1.1.22)

Theorem 5. If f (t) satisfies (1.1.6) for some constants M and a, then

L[tf (t)] = −F ′ (s). (1.1.23)

Proof. Changing the order of differentiation and integration, we have

d ∞
Z ∞
∂e−st
Z
′ −st
F (s) = f (t)e dt = f (t) dt
ds ∂s
Z ∞0 0

= −tf (t)e−st dt = −L[tf (t)]. (1.1.24)

0

Example 6. Find the Laplace transform of t sinh t.

Solution. Recall that

1
L[sinh t] = . (1.1.25)
s2 −1
By the last theorem, we get
d 1 2s
L[t sinh t] = − 2
= 2 . (1.1.26)
ds s − 1 (s − 1)2
Theorem 6. If f (t) satisfies (1.1.6) for some constants M and a, and if
limt→0 1t f (t) exists, then
Z ∞
1
L[ t f (t)] = F (ω) dω. (1.1.27)
s

Proof. Let g(t) = 1t f (t), i.e., f (t) = tg(t). The previous theorem then
gives F (s) = −G′ (s). By the fundamental theorem of calculus, and the fact
that G(s) → 0 as s → ∞, we have
Z ∞
G(s) = F (ω) dω. (1.1.28)
s
1.1. THE LAPLACE TRANSFORM 11

sin t
Example 7. Find the Laplace transform of .
t
Solution. Recall that L[sin t] = (s2 + 1)−1 . Since
sin t
lim = 1, (1.1.29)
t→0 t
the assumptions of the last theorem are satisfied and thus we have
Z ∞
dω π
1
L[ t sin t] = 2
= [arctan ω]∞
s = − arctan s. (1.1.30)
s ω +1 2
A fundamental property of the Laplace transform is the fact that, roughly
speaking, taking the derivative of the original function f (t) corresponds to
multiplying its transform F (s) by s.

Theorem 7. Suppose f (t) and f ′ (t) are two continuous, piecewise smooth
functions satisfying the inequality (1.1.6) for the same values of M and a.
Then, it holds
L[f ′ (t)] = sF (s) − f (0). (1.1.31)
Proof. Integrating by parts, we have
Z ∞
′
L[f (t)] = f ′ (t)e−st dt
0
Z ∞
−st ∞
f (t)e−st dt = −f (0) + sF (s).
 
= f (t)e 0
+s (1.1.32)
0

Applying this result to f ′′ (t) yields

L[f ′′ (t)] = sL[f ′(t)] − f ′ (0) = s2 F (s) − sf (0) − f ′ (0). (1.1.33)

Similarly,

L[f ′′′ (t)] = s3 F (s) − s2 f (0) − sf ′ (0) − f ′′ (0). (1.1.34)

By induction, we obtain the transform of the n-th derivative, viz.,

L[f (n) (t)] = sn F (s) − sn−1 f (0) − sn−2 f ′ (0) − . . . − f (n−1) (0). (1.1.35)
12 CHAPTER 1. LAPLACE TRANSFORMATION

Theorem 8. The Laplace transform of

Z t
f (τ ) dτ, (1.1.36)
0

is given by 1s F (s).
Rt
Proof. Let h(t) = 0 f (τ ) dτ . By construction we then have h′ (t) = f (t)
and h(0) = 0. Applying then the result (1.1.31) to h(t) we immediately get
F (s) = sH(s) − h(0). Hence, H(s) = 1s F (s).

Problem 1. Find the Laplace transform of the following functions.

a. t b. t2 c. t3 d. tn
1
e. t + 1 f. (t − 1)2 g. (1 + t)4 h. t
t2
i. e−t j. e3t+4 k. tet l. e
m. cosh t n. cos t o. sin 2t p. sinh2 t

Problem 2. Find the Laplace transform of the following functions.

a. eat cos bt b. θ(t − 1) c. e−t θ(t − 1)

d. t2 sinh t e. t3 et f. te−t cos t

g. sin(ωt + α) h. t sin 2t i. ln t
1
j. t
(1 − cos t) k. cosh t cos t l. cos2 t
1.1. THE LAPLACE TRANSFORM 13

1.1.4 Table of Laplace Transforms

f (t) F (s)

af (t) + bg(t) aF (s) + bG(s)

tf (t) −F ′ (s)

tn f (t) (−1)n F (n) (s)

e−at f (t) F (s + a)

f (t − T )θ(t − T ) e−T s F (s)

f ′ (t) sF (s) − f (0)

f ′′ (t) s2 F (s) − sf (0) − f ′ (0)

n
X
n
f (n)
(t) s F (s) − sn−k f (k−1) (0)
k=1
t
F (s)
Z
f (τ ) dτ
0 s

Table 1.1: Operational properties of the Laplace transform.

14 CHAPTER 1. LAPLACE TRANSFORMATION

1
θ(t)
s
tn 1
n! sn+1
1
e−at
s+a
s
cosh at
s2 − a2
a
sinh at
s2 − a2
s
cos bt
s2 + b2
b
sin bt
s2 + b2
t s
sin bt
2b (s2 + b2 )2
1 1
(sin bt − bt cos bt)
2b3 (s2 + b2 )2
a 2 /4t √
√ e−a e−a s
4πt3

Table 1.2: Standard transform pairs.

1.2. THE INVERSE LAPLACE TRANSFORM 15

1.2 The Inverse Laplace Transform

Finding the inverse Laplace transform of a function f (t) is the operation of
recovering f (t) from its Laplace transform F (s). One usually denotes this
by
f (t) = L−1 [F (s)]. (1.2.1)
Although there exist a so-called inversion formula, which gives a closed
form expression for L−1 [F (s)], we shall be content with the simple minded
approach of finding inverse Laplace transforms by using a table of standard
Laplace transforms. Indeed, it turns out that with the aid of a table and a
little algebra, we are able to find L−1 [F (s)] for a large class of functions f (t).
Due to the fact that the Laplace transform is linear it follows that also
the inverse Laplace transform is linear. Hence, if a and b are constants, then
we have
L−1 [aF (s) + bG(s)] = aL−1 [F (s)] + bL−1 [G(s)], (1.2.2)

Example 8. Find the inverse Laplace transform f (t) = L−1 [F (s)] of

e−s e−2t
F (s) = − 4 . (1.2.3)
s2 s
Solution. From a table of Laplace transforms, we have

L−1 [ s12 ] = t, L−1 [ s14 ] = 1 3

6
t, (1.2.4)

so by the linearity of L−1 , we obtain

L−1 [ es2 − e−2s

= L−1 [ es2 ] − L−1 [ e s4 ].
−s −s −2s
s4
] (1.2.5)

Using now the 2nd Shifting Rule, we find

L−1 [ es2 ] = θ(t − 1)(t − 1), L−1 [ e s4 ] = 16 θ(t − 2)(t − 2)3 .

−s −2s
(1.2.6)

Hence, the inverse transform of F (s) is

f (t) = L−1 [F (s)] = θ(t − 1)(t − 1) − 61 θ(t − 2)(t − 2)3 . (1.2.7)

16 CHAPTER 1. LAPLACE TRANSFORMATION

1.2.1 Method of Partial Fractions

A common situation is when F (s) has the form
Q(s)
F (s) = , (1.2.8)
P (s)
where Q(s) and P (s) are real polynomials and the degree of Q is less than
the degree of P . It is then necessary to decompose F (s) into partial fractions
to obtain L−1 [F (s)].
We demonstrate this technique for three cases of denominators P (s).

1. P (s) is a Quadratic with real Roots. Consider, for instance,

2s − 8
F (s) = . (1.2.9)
s2 − 5s + 6
Obviously, F (s) cannot be inverted by inspection and neither do we have it
tabulated. However, since the denominator s2 − 5s + 6 has two real roots,
s = 2 and s = 3, it is possible to decompose F (s) into partial fractions, viz.,
A B
F (s) = − . (1.2.10)
s−2 s−3
where A and B are numbers. Our goal is to determine these, because then it
is easy to obtain the inverse transform of F (s). By elementary manipulations,
we get
A B A(s − 3) + B(s − 2) (A + B)s + (−3A − 2B)
+ = = , (1.2.11)
s−2 s−3 (s − 2)(s − 3) s2 − 5s + 6
which implies
2s − 8 (A + B)s + (−3A − 2B)
= . (1.2.12)
s2 − 5s + 6 s2 − 5s + 6
Comparing the right and left hand side, it is obvious that
2 = A + B, −8 = −3A − 2B, (1.2.13)
which is a system of equations for A and B, i.e.,

A+B = 2,
(1.2.14)
3A + 2B = 8.
1.2. THE INVERSE LAPLACE TRANSFORM 17

Solving, we obtain A = 4 and B = −2. Hence,

4 2
F (s) = − . (1.2.15)
s−2 s−3
1 1
By recognizing s−2
as the transform of e2t and s−2
as that of e3t , we obtain
f (t) = L−1 [F (s)] = 4e2t − 2e3t . (1.2.16)

2. P (s) is a Quadratic with a Double Root. Let

s+1
F (s) = , (1.2.17)
(s + 2)2
The denominator has a double root −2 and the partial fractions are therefore
s+1 A B As + (2A + B)
2
= + 2
= . (1.2.18)
(s + 2) s + 2 (s + 2) (s + 2)2
Comparing the left and right hand side of the above expression, we find A = 1
1
and B = −1. Recalling that L[ s+2 ] = e−2t , we can use L[tf (t)] = −F ′ (s) to
deduce that the inverse transform of (s + 2)−2 is te−2t . Hence, the inverse of
F (s) is
f (t) = e−2t − te−2t = e−2t (1 − t). (1.2.19)

3. P (s) is a Quadratic and has Complex Conjugated Roots. If

s+1
F (s) = , (1.2.20)
s2
+ 4s + 5
then the denominator has the roots −2 ± i. Completing the square, we get
s2 + 4s + 5 = s2 + 4s + 4 + 1 = (s + 2)2 + 1, (1.2.21)
i.e.,
s+1
F (s) = . (1.2.22)
(s + 2)2 + 1
By rewriting
s+1 s+2 −1
2
= + , (1.2.23)
(s + 2) + 1 (s + 2) + 1 (s + 2)2 + 1
2

and recalling the transforms of sin t and cos t it is clear that

f (t) = L−1 [F (s)] = e−2t cos t − e−2t sin t. (1.2.24)
18 CHAPTER 1. LAPLACE TRANSFORMATION

Example 9. Find the inverse transform of

s+2
F (s) = . (1.2.25)
s3 − s2 + s − 1
Solution. Note that s3 − s2 + s − 1 = s2 (s − 1) + (s − 1) = (s − 1)(s2 + 1).
Thus,
s+2
F (s) = . (1.2.26)
(s − 1)(s2 + 1)
Here, the appropriate decomposition into partial fractions is given by
A Bs + C s2 (A + B) + s(C − B) + (A − C)
F (s) = + 2 = . (1.2.27)
s−1 s +1 s3 − s2 + s − 1
Identifying coefficients it is clear that

A + B = 0, C − B = 1, A − C = 2, (1.2.28)

which implies, A = 23 , B = − 32 , and C = − 21 . Hence,

3 3 1
2 2
s 2
F (s) = − − . (1.2.29)
s−1 s2 + 1 s2 + 1
Consulting a table of transforms, we recognize F (s) as the transform of

f (t) = 32 et − 23 cos t − 12 sin t. (1.2.30)

Problem 3. Find the inverse Laplace transform of the following functions.

1 1 s+1 1
a. b. c. d.
s+1 s2 +4 s2 + 1 s2 −1
s + 12 s s+1 e−s
e. f. g. h.
s2 + 4s (s + 2)2 (s − 3)4 s
Problem 4. Find the inverse Laplace transform of the following functions.

s s+2 1
a. b. c.
s2 − 2s − 3 s2 + 4s + 5 (s − 2)2 + 9
s+1 3s 1
d. e. f.
s3 + s2 − 6s s2 + 2s − 8 s(s + 1)(s + 2)
1.3. APPLICATIONS OF LAPLACE TRANSFORMS 19

1.3 Applications of Laplace Transforms

1.3.1 Initial Value Problems
Enough with theory, let us find the solution y(t) of the initial value problem

y ′ (t) + 2y(t) = 12e3t , y(0) = 3. (1.3.1)

By taking the Laplace transform of every term in the given differential

equation, we get
L[y ′ (t)] + L[2y(t)] = L[12e3t ]. (1.3.2)
Put Y (s) = L[y(t)]. Now,

L[y ′(t)] = sY (s) − y(0) = sY (s) − 3, (1.3.3)

L[2y(t)] = 2Y (s), (1.3.4)
12
L[12e3t ] = . (1.3.5)
s−3
Inserting these formulas into (1.3.2) above, we get the subsidiary equation
12
sY (s) − 3 + 2Y (s) = . (1.3.6)
s−3
Rearranging, we obtain
12 3 + 3s
(s + 2)Y (s) = +3= , (1.3.7)
s−3 s−3
or
3s + 3
Y (s) = . (1.3.8)
(s + 2)(s − 3)
At this point, we decompose Y (s) into partial fractions, viz.,

3s + 3 A B (A + B)s − 3A + 2B
= + = , (1.3.9)
(s + 2)(s − 3) s+2 s−3 (s + 2)(s − 3)

which gives rise to a system of equations for A and B, namely,


A+B = 3,
(1.3.10)
−3A + 2B = 3.
20 CHAPTER 1. LAPLACE TRANSFORMATION

3 12
Solving this, we find A = 5
and B = 5
. Hence,
3 12
5 5
Y (s) = + . (1.3.11)
s+2 s−3
Consulting a table of standard Laplace transforms, we finally have

y(t) = L−1 [Y (s)]

= 53 L−1 [ s+2
1
] + 12
5
L−1 [ s−3
1
]
3 12 3t
= e−2t + e . (1.3.12)
5 5

Summary of Solution Process. Note the three steps of the solution

process:

1. Take the Laplace transform of both sides of the given hard problem
for y(t). As a result a simple algebraic equation for Y (s) = L[y(t)] is
obtained.

2. Solve this so-called subsidiary equation for Y (s).

3. Use partial fractions and a table of elementary Laplace transforms to

invert Y (s) and so produce the required solution y(t) = L−1 [Y (s)].

Example 10. Solve the following initial value problem for t > 0

y ′′ (t) + 4y ′(t) + 3y(t) = 0, (1.3.13)

y(0) = 3, y ′ (0) = 1. (1.3.14)

Solution. We have

L[y ′ (t)] = sY (s) − 3, L[y ′′ (t)] = s2 Y (s) − 3s − 1, (1.3.15)

Laplace transformation of (1.3.13) yields the subsidiary equation

s2 Y (s) + 4sY (s) + 3Y (s) = 3s + 13, (1.3.16)

or
(s + 3)(s + 1)Y (s) = 3s + 13. (1.3.17)
1.3. APPLICATIONS OF LAPLACE TRANSFORMS 21

Solving for Y (s) and using a decomposition into partial fractions, we get
3s + 13 A B
Y (s) = = +
(s + 3)(s + 1) s+3 s+1
A(s + 1) + B(s + 3) (A + B)s + A + 3B
= = , (1.3.18)
(s + 3)(s + 1) (s + 3)(s + 1)
from which we obtain A = −2, and B = 5. Thus,
2 5
Y (s) = − + . (1.3.19)
s+3 s+1
Recalling (1.1.4) it is obvious that

L−1 s+3 = e−3t , L−1 = e−t .

 1   1

s+1
(1.3.20)

Hence, the solution is given by

y(t) = −2e−3t + 5e−t . (1.3.21)

A simple way to check whether the correct solution has been obtained
is to see if the initial condition is satisfied by the found function y(t). Here
we have y(0) = −2 + 5 = 3 and, since y ′(t) = 6e−3t − 5e−t , we also have
y ′ (0) = 6 − 5 = 1. Hence, the given initial conditions are indeed satisfied by
(1.3.21).
Problem 5. Solve the following differential equations for t > 0
a. y ′ + 2y = e−3t , y(0) = 4.

b. y ′ − y = e2t , y(0) = −1.

c. y ′′ + 2y ′ + y = e−t , y(0) = 0, y ′ (0) = 1.

d. y ′′ + 4y ′ + 13y = 2e−t , y(0) = 0, y ′(0) = −1.

e. y ′′ + 4y = 8e2t , y(0) = 0, y ′(0) = 3.

f. y ′′ − 2y ′ + 2y = cos t, y(0) = 1, y ′ (0) = 0.

g. y ′′ + 4y ′ = 3et , y(0) = 2, y ′(0) = 1.

h. y ′′ + 2y ′ + 2y = 2, y(0) = 0, y ′(0) = 0.
22 CHAPTER 1. LAPLACE TRANSFORMATION

1.3.2 Integral Equations

Apart from solving differential equations, the Laplace transform technique
may also be used to solve integral equations. For instance, consider the flow
of electric current around a circuit consisting of a resistor, a capacitance, and
a battery.
R

v(t) C
i(t)

Figure 1.2: Electric RC-circuit.

It follows3 that the current i(t) satisfies the integral equation

1 t
Z
Ri(t) + i(τ ) dτ = v(t), (1.3.22)
C 0
where R and C are respectively the resistance and capacity of the circuit, and
v(t) is the electromotive force of the battery. For simplicity, let us assume
that C = R = 1, and that v(t) has the form of a square puls of amplitude 1,
applied between t = 1 and t = 2, i.e.,

0, t < 1,

v(t) = θ(t − 1) − θ(t − 2) = 1, 1 < t < 2, (1.3.23)

0, t > 2.


The Laplace transform V (s) of v(t) is given by

Z ∞ Z 2  −st 2
−st −st e e−2s e−s
V (s) = v(t)e dt = e dt = − =− + . (1.3.24)
0 1 s 1 s s
Assuming that i(0) = 0, we may transform (1.3.22) to obtain
I(s)
RI(s) + = V (s), (1.3.25)
Cs
3
From the Kirchoff voltage law.
1.3. APPLICATIONS OF LAPLACE TRANSFORMS 23

or, since R = C = 1,
I(s) 1
I(s) + = (e−s − e−2s ). (1.3.26)
s s
Solving for I(s) we obtain, after elementary manipulations,
1
I(s) = (e−s + e−2s ). (1.3.27)
s+1
1
Noting that L−1 [ s+1 ] = e−t , we then use the 2nd Shifting Rule, to obtain

i(t) = e−(t−1) θ(t − 1) − e−(t−2) θ(t − 2). (1.3.28)

Hence,

0,
 t < 1,
1 −t
i(t) = e e , (1.3.29)
1 < t < 2,

 1
(e − e2 )e−t , t > 2.

i(t)

0.5

0 t
1 2 3
−0.5

−1.0

Figure 1.3: Graph of i(t) for 0 < t < 4.

Problem 6. Solve the integral equation for t > 0

Z t
2y(t) + y(τ ) dτ = 4.
0
24 CHAPTER 1. LAPLACE TRANSFORMATION

Problem 7. Solve the following integral equations for t > 0

Z t
′
a. y (t) + 2y(t) + y(τ ) dτ = cos t, y(0) = 1.
0
Z t
′
b. y (t) + 2y(t) + 2 y(τ ) dτ = 1 + e−t , y(0) = 1.
0
Z t
′′
c. y (t) − 7y(t) + 6 y(τ ) dτ = 1, y(0) = 7, y ′(0) = −12.
0
Chapter 2

Fourier Analysis

Periodic phenomenon occur frequently throughout nature and their study is of

the utmost importance for our understanding of many real-world systems. For
example, the signals from radio pulsars allow astronomers to study space, the
seasonal periodicity of the weather governs the crop of corn, and the regular
beats of a heart is necessary for the survival of every mammal. Periodicity can
be found everywhere and concern any absolute variable, i.e., time, space, velocity,
etc. In this chapter we shall begin to study periodic functions, and especially,
their representation as sums of sine and cosine functions, Fourier series.
Fourier series has long provided one of the principal tools of analysis for
mathematical physics, engineering, and signal processing. It has spurred many
generalizations, and applications that continue to develop right up to the present.
While the original theory of Fourier series applies to periodic functions describing
wave motion, such as with light and sound, its generalizations often relate to
wider settings, for example, the time-frequency analysis underlying the recent
theories of wavelet analysis and local trigonometric analysis. We shall, however,
be content with presenting the basic theory and its application to the solution of
partial differential equations.

25
26 CHAPTER 2. FOURIER ANALYSIS

2.1 Periodic Functions

A function f (x) is said to be periodic if there is a constant p > 0, such that

f (x + p) = f (x), (2.1.1)

for all x. Any positive number p with this property is called a period of f (x).
For example, f (x) = cos x has periods 2π, 4π, etc. However, the smallest
number p > 0 with the property (2.1.1) is called the prime period, and it
is generally this value that is meant when a function is referred to as being
p-periodic, or, of period p.

0 p 2p

Figure 2.1: Illustration of a p-periodic function.

Lemma 1. Suppose f (x) is periodic with period p, then the integral

Z a+p
f (x) dx, (2.1.2)
a

is independent of starting point a.

Proof: Let
Z a+p Z a+p Z a
g(a) = f (x) dx = f (x) − f (x) dx. (2.1.3)
a 0 0

By the fundamental theorem of calculus, we have g ′(a) = f (a + p) − f (a) but

since f (p + a) = f (a) we also have g ′ (a) = 0. Hence, g(a) is constant and
thus independent of a.
2.2. FOURIER SERIES 27

2.2 Fourier Series

2.2.1 Definition
From about 1800 onwards, the French scientist Joseph Fourier1 was lead
by problems of heat conduction to consider the possibility of representing a
more or less arbitrary 2π-periodic function f (x) as a linear combination2 of
the functions

1, cos x, sin x, cos 2x, sin 2x, cos 3x, sin 3x, . . . (2.2.2)

Fourier conjectured that any integrable periodic function f (x) of period

2π can be written, at almost every point x, as the sum of a trigonometric
series of the form
∞
X
1
f (x) = a0 +
2
(an cos nx + bn sin nx), (2.2.3)
n=1

where an and bn , n = 0, 1, 2, . . ., are real numbers, defined by

1 π
Z
an = f (x) cos nx dx, (2.2.4)
π −π
1 π
Z
bn = f (x) sin nx dx. (2.2.5)
π −π

Here the term 12 a0 is due to the constant function cos 0 = 1, the factor 12
being included for reasons of later convenience. Further, b0 does not exist,
since sin 0 = 0.
Fourier managed to solve several problems of heat flow using such series
representations, and, as a result, (2.2.3) is today called the Fourier series of
f (x). Similarly, the corresponding numbers an and bn are called the Fourier
coefficients of f (x).
1
Jean Baptiste Fourier (1768-1830) French physicist and mathematician.
2
Recall that if {φ1 , φ2 , φ3 , . . .} is a collections of things (e.g., numbers, vectors, or
functions) that can be multiplied by scalars and added together, then a linear combination
is any expression of the form

c1 φ1 + c2 φ2 + c3 φ3 + . . . , (2.2.1)

where cj , j = 1, 2, 3, . . . are constants.

28 CHAPTER 2. FOURIER ANALYSIS

Example 11. Find the Fourier series of the 2π-periodic ramp function

f (x) = x, 0 ≤ x ≤ 2π. (2.2.6)

6.28

3.14

0
−12.56 −6.28 0.00 6.28 12.56
Figure 2.2: Graph of the ramp function.

Solution. By definition we have for n = 0,

π 2π  2π
1 1 1 x2
Z Z
a0 = f (x) dx = x dx = = 2π, (2.2.7)
π −π π 0 π 2 0

and for n = 1, 2, 3, . . .

1 π
Z
an = f (x) cos nx dx
π −π
1 2π
Z
= x cos nx dx
π 0
 2π Z 2π
x sin nx 1
= − sin nx dx
π n 0 nπ 0
 2π
1
= cos nx = 0. (2.2.8)
πn2 0

Similarly,

1 π
Z
bn = f (x) sin nx dx
π −π
1 2π
Z
= x sin nx dx
π 0
h x cos nx i2π Z 2π
1 2
= − + cos nx dx = − . (2.2.9)
π n 0 nπ 0 n
2.2. FOURIER SERIES 29

Hence, the Fourier series of f (x) is given by

∞
X 2
f (x) = π − sin nx
n=1
n
 
1 1 1
= π − 2 sin x + sin 2x + sin 3x + sin 4x + . . . . (2.2.10)
2 3 4
To get a feeling for Fourier series, it is helpful to add up and graph the
first few terms of this series. Below, we show the sum (2.2.10) terminated
after 5 terms.
6.28

3.14

0
−12.56 −6.28 0.00 6.28 12.56
6.28

3.14

0
−12.56 −6.28 0.00 6.28 12.56
Figure 2.3: Partial sums of the Fourier series for the ramp function.

As seen, the sum is indeed attempting to reproduce the ramp function

rather successfully. Adding more terms will further reduce the remaining
wiggles.
Problem 8. Find the Fourier series of the 2π-periodic function
(
1, 0 < x < π,
f (x) =
0, −π < x < 0.
Problem 9. Find the Fourier series of the 2π-periodic function
(
x, 0 ≤ x < π,
f (x) =
0, −π < x ≤ 0.
30 CHAPTER 2. FOURIER ANALYSIS

2.2.2 Complex Representation of Fourier Series

We can greatly simplify the expression for the Fourier series (2.2.3) by using
the relations
einx + e−inx einx − e−inx
cos nx = , sin nx = , (2.2.11)
2 2i
√
where i = −1 is the imaginary unit. Hence,
∞
X
f (x) = 21 a0 + 1
a (einx
2 n
+ e−inx ) − 2i bn (einx − e−inx )
n=1
∞
X ∞
X
inx
= 1
a
2 0
+ 1
(a
2 n
− ibn )e + 1
(a
2 n
+ ibn )e−inx . (2.2.12)
n=1 n=1

Defining c0 = 21 a0 , cn = 21 (an − ibn ) and c−n = 12 (an + ibn ) for n = 1, 2, 3, . . .,

we have
∞
X X ∞ X∞
inx −inx
f (x) = c0 + cn e + c−n e = cn einx , (2.2.13)
n=1 n=1 −∞

which is called the complex Fourier series of f (x).

2.2.3 Derivation of the Euler Formulas

Although the complex Fourier series and the sine-cosine series are identical,
some manipulations and calculations become more streamlined when using
the complex representation. For instance, it is easy to justify the definition
of the numbers an and bn , the so-called Euler formulas (2.2.4) and (2.2.5).
To do so, we need the next lemma.
Lemma 2. {einx }∞
−∞ forms a so-called orthogonal set on −π ≤ x ≤ π, in
the sense that
Z π (
i(n−k)x 0, n 6= k,
e dx = (2.2.14)
−π 2π, n = k.
Proof.
Z π  i(n−k)x π
i(n−k)x e (−1)n−k − (−1)n−k
e dx = = = 0, n 6= k,
−π i(n − k) −π i(n − k)
(2.2.15)
2.2. FOURIER SERIES 31

whilst Z π Z π
i(n−k)x
e dx = dx = 2π, for n = k. (2.2.16)
−π −π

Now, multiplying (2.2.13) by e−ikx , and integrating term-by-term from

−π to π, we obtain
Z π ∞
X Z π
−ikx
f (x)e dx = cn ei(n−k)x dx. (2.2.17)
−π −∞ −π

By the orthogonality property above, this leads to

Z π
f (x)e−ikx dx = 2πck for n = k. (2.2.18)
−π

Relabeling the integer k by n we thus have a formula for cn , namely,

Z π
1
cn = f (x)e−inx dx. (2.2.19)
2π −π

Finally, using the fact that an = cn + c−n and that bn = i(cn − c−n ) we find
for n = 0,
1 π
Z
a0 = 2c0 = f (x) dx, (2.2.20)
π 0
and for n = 1, 2, 3 . . .,

an = cn + c−n
Z π π
1 1
Z
= f (x)(einx + e−inx ) dx = f (x) cos nx dx, (2.2.21)
2π −π π −π
bn = i(cn − c−n )
Z π π
i 1
Z
= f (x)(e−inx − einx ) dx = f (x) sin nx dx. (2.2.22)
2π −π π −π
32 CHAPTER 2. FOURIER ANALYSIS

2.2.4 Even and Odd Functions

Much unnecessary work and corresponding sources of error can be avoided
when calculating an and bn , if f (x) is either even or odd.
Recall that a function g(x) is said to be even if g(x) = g(−x) for all x.
Similarly, a function h(x) is said to be odd if h(x) = −h(−x) for all x.
From the graphs of g(x) and h(x) it is obvious that the integrals of these
functions over any symmetric interval, e.g., −a ≤ x ≤ a, may be simplified,
viz.,
Z a Z a Z a
g(x) dx = 2 g(x) dx, h(x) dx = 0. (2.2.23)
−a 0 −a

2 g(x) h(x)
1

−6 −4 −2 2 4 6
−1

−2

Figure 2.4: Illustration of even and odd functions.

Moreover, the product q(x) = g(x)h(x) is odd, since
q(−x) = g(−x)h(−x) = g(−x)[−h(x)] = −g(x)h(x) = −q(x). (2.2.24)
Thus, if f (x) is even, then f (x) sin nx is odd, since sin nx is odd. Similarly,
if f (x) is odd, then f (x) cos nx is odd, since cos nx is even. Hence, from
the definitions (2.2.4) and (2.2.5) of an and bn we get the results of the next
lemma.
Lemma 3. If f (x) is even, then
2 π
Z
an = f (x) cos nx dx, bn = 0. (2.2.25)
π 0
whereas if f (x) is odd, then
π
2
Z
an = 0, bn = f (x) sin nx dx. (2.2.26)
π 0
2.2. FOURIER SERIES 33

Example 12. Find the Fourier series of the square wave, defined by
(
+1, 0 < x < π,
f (x) = (2.2.27)
−1, −π < x < 0,
f (x) = f (x + 2π). (2.2.28)

2
1
0
−1
−2
−9.42 −6.28 −3.14 0.00 3.14 6.28 9.42
Figure 2.5: Graph of the square wave.

Solution. Since f (x) is odd we have an = 0 for all n, and

2 π 2 h cos nx iπ 2 (−1)n − 1
Z
bn = sin nx dx = − =− , (2.2.29)
π 0 π n 0 π n

since cos nπ = (−1)n . Hence, the Fourier series of f (x) is given by

∞
2 X 1 − (−1)n
f (x) = sin nx
π n=1 n
 
4 1 1 1
= sin x + sin 3x + sin 5x + sin 7x + . . . . (2.2.30)
π 3 5 7

Below, we graph a few partial sums of this Fourier series. As is evident,

they provide increasingly better approximations to the square wave as the
number of terms increase, except possibly near points of discontinuity, i.e.,
x = 0, ±π, ±2π, etc. Intuitively, this is what is to be expected, since it is
hard to express a jump discontinuity using the perfectly smooth functions
sin nx and cos nx.
34 CHAPTER 2. FOURIER ANALYSIS
2 4
(sin x + 13 sin 3x)
π
1

−6.28 −3.14 −1 3.14 6.28

−2

2 4
(sin x + 13 sin 3x + 51 sin 5x)
π
1

−6.28 −3.14 −1 3.14 6.28

−2

2 4
(sin x + 31 sin 3x + . . . + 1
sin 17x)
π 17
1

−6.28 −3.14 −1 3.14 6.28

−2

Figure 2.6: Partial sums of the Fourier series for the square wave.

Notice that if we let x = π2 , then the Fourier series (2.2.30) yields

1 = π4 sin π2 + 31 sin 3π 1 5π 1 7π


2
+ 5
sin 2
+ 7
sin 2
+ . . . , (2.2.31)
since f ( π2 ) = 1. But this implies that we can find the sum of a non-trivial,
alternating series, namely,
1 1 1 π
1− + − + ... = . (2.2.32)
3 5 7 4
Problem 10. Calculate the sum
∞
X (−1)n+1 1 1 1
=1− + − + ...
n=1
n2 4 9 16

by using the Fourier series expansion of the 2π-periodic function

f (x) = x2 , −π ≤ x ≤ π.
2.2. FOURIER SERIES 35

2.2.5 Bessel Inequality and Riemann-Lebesgue Lemma

Rπ
Theorem 9 (Bessel Inequality). If −π
|f (x)|2 dx is finite, then

∞ π
1
X Z
2
|cn | ≤ |f (x)|2 dx. (2.2.33)
−∞
2π −π

Proof. Let
N
X
f
SN (x) = cn e−inx , (2.2.34)
n=−N

denote the N-th partial sum of the Fourier series for f (x). We then have

N
X
f
|f (x) − SN (x)|2 = |f (x) − cn einx |2
n=−N
N
! N
!
f¯(x) −
X X
= f (x) − cn einx c̄n e−inx
n=−N n=−N
N
cn f¯(x)einx + c̄n f (x)e−inx
X
= |f (x)|2 −


n=−N
N
X N
X
+ cm c̄n ei(m−n)x , (2.2.35)
n=−N m=−N

where bars denote complex conjugates. Dividing by 2π and integrating from

−π to π, we get

π N
1
Z X
0≤ |f (x) − cn einx |2 dx
2π −π n=−N
π N  Z π Z π 
1 1 X
Z
= |f (x)| dx − 2
cn f¯(x)e dx + c̄n
inx ¯
f(x)einx
dx
2π −π 2π n=−N −π −π
N N Z π
1 X X
+ cm c̄n ei(m−n)x dx. (2.2.36)
2π n=−N m=−N −π
36 CHAPTER 2. FOURIER ANALYSIS

Now, using the orthogonality (2.2.14) and the formula (2.2.19) for cn , this
simplifies, viz.,

π N N
1
Z X X
2
0≤ |f (x)| dx − (cn c̄n + c̄n cn ) + |cn |2
2π −π n=−N n=−N
π N
1
Z X
= |f (x)|2 dx − |cn |2 . (2.2.37)
2π −π n=−N

Finally, let N → ∞ and the proof is complete.

Because of the Bessel inequality, it follows that

lim cn = 0, (2.2.38)
|n|→∞

which is known as the Riemann-Lebesgue Lemma.

Rπ
Lemma 4 (Riemann-Lebesgue Lemma). If −π |f (x)|2 dx is finite, then
it holds
lim an = 0, lim bn = 0, lim cn = 0. (2.2.39)
n→∞ n→∞ |n|→∞

Proof. From (2.2.33) we know that the sum

∞
X
|cn |2 , (2.2.40)
−∞

converges, so it follows that cn → 0 as |n| → ∞. But since an = cn + c−n

and bn = i(cn − c−n ) it also follows that an → 0 and that bn → 0 as n → ∞.

Hence, according to the Riemann-Lebesgue lemma the terms of a Fourier

series decrease as n tend to infinity, which is necessary if the series is to
converge.
Rπ
Problem 11. Is there a function f (x) with a finite −π |f (x)|2 dx, such that
∞
X
f (x) = sin nx. (2.2.41)
n=1
2.2. FOURIER SERIES 37

2.2.6 Convergence of Fourier Series

All convergence theorems are concerned with how the partial sums
N
X
f
SN (x) = cn einx , (2.2.42)
n=−N

converge to f (x).
The question of pointwise convergence, for example, concerns whether
f
lim SN (x0 ) = f (x0 ), (2.2.43)
N →∞

holds for each fixed value of x0 .

Below, we state a fundamental convergence theorem for Fourier series,
f
which says that SN (x0 ) converges pointwise to the average value of the left
and right hand limits of f (x) at x0 , as N tend to infinity. Consequently,
if f (x) is continuous at x0 , then its Fourier series will eventually converge
to f (x0 ). On the other hand, if f (x) is discontinuous at x0 , then we must
redefine f (x0 ) as 21 (f (x0− ) + f (x0+ )) to be able to claim the convergence of
its Fourier series at this point.

Theorem 10 (Dirichlet Theorem). If f (x) is a 2π-periodic function,

f
which is piecewise smooth for all x, and if SN (x) is the N-th partial sum
of its Fourier series, then

f f (x0− ) + f (x0+ )
lim SN (x0 ) = , (2.2.44)
N →∞ 2
f
for all points x0 . In particular, limN →∞ SN (x) = f (x0 ) for every point x0 at
which f (x) is continuous.

A trivial consequence of the last theorem is that

Z π
f
lim |f (x) − SN (x)|2 dx = 0, (2.2.45)
N →∞ −π

i.e., the average error between f (x) and its Fourier series tends to zero as N
tend to infinity. However, this means that the Bessel inequality (2.2.33) is
actually a genuine equality. Usually, it is then called the Parseval Indentity.
38 CHAPTER 2. FOURIER ANALYSIS

Theorem 11 (Parseval Identity). With reference to the last theorem, it

holds that Z π ∞
1 2
X
|f (x)| dx = |cn |2 , (2.2.46)
2π −π −∞

or, equivalently,
π ∞
1
Z X
2
|f (x)| dx = 1
|a |2
2 0
+ |an |2 + |bn |2 . (2.2.47)
π −π n=1

Example 13. Compute the sum

1 1 1 4
4
+ 4 + 4 + 4 + ..., (2.2.48)
1 3 5 7
using the Fourier series expansion of f (x) = |x|, −π ≤ x ≤ π.

Solution. We notice that f (x) is even, so bn = 0 and

2 π 2 π
Z Z
an = |x| cos nx dx = x cos nx dx. (2.2.49)
π 0 π 0
Thus, for n = 0,  π
π
2 2 x2
Z
a0 = x dx = = π, (2.2.50)
π 0 π 2 0
and for n > 0,
π
2 π 2 π sin nx

2 x sin nx
Z Z
an = x cos nx dx = − dx
π 0 π n 0 π 0 n
2 h cos nx iπ 2 cos nπ − 1 2 (−1)n − 1
= = = . (2.2.51)
π n2 0 π n2 π n2
since sin nπ = 0 and cos nπ = (−1)n . Consequently,
∞
π 2 X (−1)n − 1
f (x) = + cos nx. (2.2.52)
2 π n=1 n2

Furthermore,
π π
2 3
Z Z
2
|f (x)| dx = 2 x2 dx = π . (2.2.53)
−π 0 3
2.2. FOURIER SERIES 39

Thus, using the Parseval identity (2.2.47) we obtain

∞
12 3 1 2 4 X ((−1)n − 1)2
π = π + 2 , (2.2.54)
π3 2 π n=1 n4

i.e.,

2 2 π 2 16
 
1 1 1 1
π = + 2 + + + + ... . (2.2.55)
3 2 π 14 34 54 74

Hence,
1 1 1 4 π4
+ + + + . . . = . (2.2.56)
14 34 54 74 96
Problem 12. Compute the sum
∞
X 1
2
, (2.2.57)
n=1
n

using the Fourier series expansion of f (x) = x, −π ≤ x ≤ π.

2.2.7 Functions of Arbitrary Period

Fourier series expansions are not limited to periodic functions of period 2π
only. For a periodic function f (x) of arbitrary period, 2L say, a simple
change of variables can be used to transform the period from −L ≤ x ≤ L to
−π ≤ x ≤ π. Therefore, let g(t) be periodic of period 2π and let t = πx/L.
We have f (x+2L) = f (x) and g(t) = f ( Lπ t) = f (x). Thus, by the periodicity,
we get
g(t + 2π) = f ( Lπ t + 2L) = f ( Lπ t) = g(t). (2.2.58)
It is now clear that the Fourier series for g,
∞
X
1
a
2 0
+ an cos nt + bn sin nt, (2.2.59)
n=1

can be written as
∞ 
1
X nπx nπx 
a
2 0
+ an cos + bn sin . (2.2.60)
n=1
L L
40 CHAPTER 2. FOURIER ANALYSIS

Its Fourier coefficients are given by

1 π
Z
an = g(t) cos nt dt
π −π
1 L nπx π
Z
= f (x) cos dx
π −L L L
1 L nπx
Z
= f (x) cos dx,
L −L L
1 L nπx
Z
bn = f (x) sin dx. (2.2.61)
L −L L

Example 14. Find the Fourier series of the rectified sine wave,

f (x) = | sin πx|, 0 ≤ x ≤ 1, f (x + 1) = f (x). (2.2.62)

Solution. Since f (x) is even bn = 0, and

1/2 1/2
4
Z Z
1/2
a0 = 2 f (x) dx = 4 sin πx dx = − π4 [cos πx]0 = . (2.2.63)
−1/2 0 π

Further, using that sin α cos β = 12 sin(α + β) + 12 sin(α − β), we get

Z 1/2
an = 4 sin πx cos 2nπx dx
0
Z 1/2 Z 1/2
=2 sin(1 + 2n)πx dx + 2 sin(1 − 2n)πx dx
0 0
2 1/2 2 1/2
=− [cos(2n + 1)πx]0 − [cos(1 − 2n)πx]0
(1 + 2n)π (1 − 2n)π
2 2 4
=− (−1) − (−1) = . (2.2.64)
(1 + 2n)π (1 − 2n)π (1 − 4n2 )π

Hence, we have
∞
2 4X 1
f (x) = − 2
cos 2nπx. (2.2.65)
π π n=1 (4n − 1)
2.2. FOURIER SERIES 41

Problem 13. Let the function f (x) be periodic of period 4 and let
f (x) = |x|, |x| < 2.
a. Graph f (x) for |x| < 6.
b. Find the Fourier series of f (x).
c. Use the series obtained to compute the sum
1 1 1
1+ 2
+ 2 + 2 + ...
3 5 7
d. Show, using the Parseval identity, that
1 1 1 π4
1+ + + + . . . = .
34 54 74 96
2.2.8 Sine and Cosine Series
In various physical and engineering problems there is a practical need to
apply trigonometric series expansions to functions f (x) only defined for a
finite interval, say 0 ≤ x ≤ L. This can be accomplished by first defining
f (x) for −L ≤ x ≤ 0 and then extending it to a periodic function of period
2L. In doing so, we can use the freedom of definition to make the extension
either even or odd as is convenient.

Fourier Sine Series. We define the odd extension F (x) of f (x) by


f (x)
 if 0 ≤ x ≤ L,
F (x) = 0 if x = 0, (2.2.66)

−f (−x) if − L ≤ x ≤ 0,


F (x + 2L) = F (x). (2.2.67)

By construction, F (x) is odd and periodic, i.e., its series expansion is
∞
X nπx
F (x) = bn sin , (2.2.68)
n=1
L
where bn are given by
1 L nπx 2 L nπx
Z Z
bn = F (x) sin dx = f (x) sin dx. (2.2.69)
L −L L L 0 L
42 CHAPTER 2. FOURIER ANALYSIS

Fourier Cosine Series. Similarly, we define the even extension F (x) of

f (x) by
(
f (x) if 0 ≤ x ≤ L,
F (x) = (2.2.70)
f (−x) if − L ≤ x ≤ 0,
F (x + 2L) = F (x). (2.2.71)

In this case, we thus have

∞ L
nπx 2 nπx
X Z
1
F (x) = a
2 0
+ an cos , an = f (x) cos dx. (2.2.72)
n=1
L L 0 L

Example 15. Develop a sine series for the function f (x) = cos x, 0 ≤ x ≤ π.
1

−9.42 −6.28 −3.14 3.14 6.28 9.42

−1

Figure 2.7: Odd extension of cos x.

Solution. We have b1 = 0 and

2 π
Z
bn = cos x sin nx dx
π 0
1 π 1 π
Z Z
= sin(n − 1)x dx + sin(n + 1)x dx
π 0 π 0
 π  π
1 cos(n − 1)x 1 cos(n + 1)x
= − + −
π n−1 0 π n−1 0
n n
1 (−1) − 1 1 (−1) − 1 2n (−1)n − 1
= + = . (2.2.73)
π n−1 π n+1 π n2 − 1
Hence, the sine series of cos x is given by
∞
X n (−1)n − 1
2 sin nx. (2.2.74)
n=1
π n2 − 1
2.2. FOURIER SERIES 43

Problem 14. Classify the following functions as even, odd, or neither even
nor odd.
a. ex b. x sin x c. ln x
2
d. sin2 x e. e−x f. 1 + x + x5 + x7
Problem 15. Develop a cosine series for the function

f (x) = x, 0 ≤ x ≤ 1.

Problem 16. Develop a cosine series for the function

f (x) = x2 , 0 ≤ x ≤ π.

Problem 17. Let the function f (x) be odd and periodic of period 2π, where

f (x) = x − π, 0 < x < π.

a. Graph f (x) for −3π ≤ x ≤ 3π.

b. Find the Fourier series of f (x).

c. Use the series obtained to compute the sum 1 − 13 + 51 − 71 + . . .

44 CHAPTER 2. FOURIER ANALYSIS
Chapter 3

Partial Differential Equations

Partial differential equations are of vast importance in science and engineering,

since they provide mathematical models to a wide variety of real-world systems.
Indeed almost all natural laws of physics can be formulated in terms of partial
differential equations. For example, the Ampère law of electromagnetics, which
relates the magnetic field H around a conducting electric wire of current density
J , can be stated simply as ∇ × H = J , which is a partial differential equation
for H.
Although mathematical physics is the traditional area of application, partial
differential equations are today found in such diverse areas as biology, chemistry,
economics, and medicine.
Our goal here is to develop a few of the most basic ideas of the theory of
partial differential equations, and apply them to the simplest models arising from
physics.

3.1 Basic Concepts

3.1.1 Definition
Definition 1. A partial differential equation 1 for a function u(x, y, . . .) with
partial derivatives ux , uy , uxx , uxy , . . ., is a mathematical relation of the form

F (x, y, . . . , u, ux, uy , uxy , . . .) = 0, (3.1.1)

where F is a given function of the variables x, y, . . . , u, ux, uy , uxx , . . .

1
Sometimes abbreviated PDE.

45
46 CHAPTER 3. PARTIAL DIFFERENTIAL EQUATIONS

Generally, the variable u which we differentiate is called the dependent

variable, whereas those which we differentiate with respect to (e.g., x and y)
are called the independent variables.
By itself, the equation (3.1.1) is too general to allow for any systematic
study. Instead, each of the major types of equations that commonly arise
is studied individually. However, there are a few basic concepts which are
essential for studying all types of equations. One such concept is the order
of a PDE.

Definition 2. The order of a partial differential equation is the order of the

highest partial derivative present within the equation.

So, for example,

ut + uux = 0, is of first order,

ut − 6uux + uxxx = 0, is of third order, etc.

Because the dynamics of most real-world systems usually involve at most

two derivatives2 , second-order equations are the most frequently occurring
and thus the most important.

3.1.2 Linearity and Superposition

One should recall from elementary calculus that the operation of taking a
derivative is a linear operation, that is, for any functions u(x) and v(x) it
holds that (u + v)x = ux + vx . Therefore, (3.1.1) is said to be linear, if it can
be rewritten as
L(u) = f, (3.1.2)
where the operator L(u) involves only sums and compositions of derivatives
without higher power terms (e.g., u2 or uux). More specific, L has to satisfy

L(av + bw) = aL(v) + bL(w), (3.1.3)

for all functions v and w, and all numbers a and b. Thus,

esin y uxx + log(1 + x3 )uyy = 0, (3.1.4)

2
Recall that, e.g., the equation of motion F = ma involves two time derivatives.
3.1. BASIC CONCEPTS 47

is linear in u, even though its so-called coefficients esin y and log(1 + x3 ) are
highly non-linear functions of x and y. However, the much simpler-looking
minimal surface equation

(1 + u2y )uxx − 2ux uy uxy + (1 + u2x )uyy = 0, (3.1.5)

is non-linear since the derivatives of u are multiplied together and squared.

Finally, if the right hand side f is identically zero, then (3.1.2) is said to
be homogeneous; otherwise it is inhomogeneous.

Theorem 12 (Superposition Principle). If u1 and u2 are two individual

solutions to the linear homogeneous equation L(u) = 0, then so is any linear
combination of them, i.e.,

u = c1 u 1 + c2 u 2 , (3.1.6)

where c1 and c2 are any constants.

Proof. Using the linearity of L(u) we have

L(u) = L(c1 u1 + c2 u2 ) = c1 L(u1 ) + c2 L(u2 ). (3.1.7)

But as L(u1 ) = 0 and L(u2 ) = 0, it immediately follows that also L(u) = 0.

For example, since u1 = x2 − y 2 and u2 = ex cos y, both satisfy Laplace’

equation L(u) = uxx + uyy = 0, so does any linear combination of them (i.e.,
u = c1 u1 + c2 u2 = c1 (x2 − y 2 ) + c2 ex cos y). As we shall see, this property is
extremely useful for constructing solutions that must obey certain auxiliary
constraints.

3.1.3 Classification
Let it suffice to classify linear second-order equations of the type

Auxx + Buxy + Cuyy + Dux + Euy + F = G, (3.1.8)

where A, . . . , G can be constants, or given functions of x and y.

By analogy with the conic sections, ellipse, hyperbola, and parabola,
all linear partial differential equations can be classified as of either elliptic,
hyperbolic, or parabolic type.
48 CHAPTER 3. PARTIAL DIFFERENTIAL EQUATIONS

Definition 3. Depending on if the quantity 4AC − B 2 , is positive, negative

or zero, the general linear second-order equation (3.1.8) is said to be either
elliptic, hyperbolic, or parabolic, respectively.

Example 16. Classify

uxx + uuxy + (1 + u2 )uyy = 0, (3.1.9)

as either hyperbolic, parabolic, or elliptic.

Here, A = 1, B = u, and C = 1 + u2 , so

4AC − B 2 = 4(1 + u2 ) − u2 = 3u2 + 4 > 0. (3.1.10)

Hence, the equation is always elliptic.

Problem 18. Classify the following equations as linear or non-linear.

a. uxx + yux + xuy = x3 y.

b. uxyx + 2x2 uxx yyy = sin xy.

c. uxy + x1 uy = 0.

Problem 19. Classify the following into elliptic, hyperbolic, or parabolic

types

a. uxx + 2uxy + (1 − x2 − y 2 )uyy = x2 + y 2 .

b. uxx + x3 uyy = 0.
3.1. BASIC CONCEPTS 49

3.1.4 Physical Derivation of the Heat Equation

One of the classical3 partial differential equations of mathematical physics
is the heat equation, describing the evolution of temperature within a heat
conducting solid.
Consider a long thin rod of length L, oriented along the x-axis. We wish to
develop a model of heat flow through the rod. For simplicity, we assume that
the rod is laterally insulated so that any flow of heat is essentially directed
lengthwise.
According to the first law of thermodynamics, which states conservation
of energy, the rate of change of internal energy equals at all times the net
inflow of heat. Thus, if j(x, t) and e(x, t) are the heat flux and the thermal
energy per unit length of the rod at a point x and time t, respectively, we
have Z L
et dx = j(0, t) − j(L, t). (3.1.11)
0

By the fundamental theorem of calculus, it follows that

Z L
j(L, t) − j(0, t) = jx dx, (3.1.12)
0

so (3.1.11) can also be written as

Z L
(et + jx ) dx = 0. (3.1.13)
0

However, since the length of the rod L is arbitrary, we must conclude that

et + jx = 0. (3.1.14)

Experimental studies show that the relation between the total thermal
energy H of a body of mass m and its temperature u is roughly linear and
given by
H = cmu, (3.1.15)
where c is known as the specific heat capacity of the material. Because our
rod is long and thin, it is reasonable to believe that its temperature u only
varies with longitudinal position x and time t, i.e., u = u(x, t). Moreover,
3
It originates back to Fourier and his thesis Theorie Analytique de la Chaleur (1822).
50 CHAPTER 3. PARTIAL DIFFERENTIAL EQUATIONS
RL
since we have defined the total energy of the rod by H = 0
e dx, it follows
that
e = ̺cu, (3.1.16)
where ̺ is the mass density of the rod.
Heat always flow from warm regions to cooler ones, but at rate which is
dependent upon the conducting medium. For example, if a copper rod and
an iron rod are joined together end to end, and the ends are subsequently
heated, the heat will conduct through the copper end more quickly than the
iron end because copper has a higher thermal conductivity.
Fourier considered the above properties and summarized them elegantly,
viz.,
j = −κux , (3.1.17)
a relation which determines the heat flux j for a given temperature profile u
and thermal conductivity κ.
Substituting (3.1.16) and (3.1.17) into (3.1.14) and assuming κ constant,
we finally arrive at
̺cut = κuxx , (3.1.18)
which is generally know as the heat equation.

3.1.5 Boundary and Initial Conditions

Motivation for Boundary Conditions. Consider the simple equation

y ′(t) = 4t. (3.1.19)

By integration we can easily obtain its general solution y(t) = 2t2 + C,

where C is a constant of integration. However, because C is arbitrary, we
cannot determine a unique solution. To do so, we must have another piece
of information about y(t).
For example, if we were asked for the solution of (3.1.19) that satisfies
y(0) = 0, then there would be only one possible answer, namely y(t) = 2t2 .
If we try any of the other solutions to the equation, e.g., y(t) = 2t2 + 1 or
y(t) = 2t2 + 2, say, we find that y(0) is not zero, but 1 or 2, respectively.
A constraint, such as y(0) = 0, which is given together with a differential
equation to fix a constant of integration, and so give a unique solution, is
called a boundary condition.
3.1. BASIC CONCEPTS 51

With second-order differential equations, we need to integrate twice to

obtain its general solution. As a result, we end up with two constants of
integration and need two boundary conditions to specify which solution we
want.

Boundary and Initial Conditions for the Heat Equation. Since the
heat equation contains a total of three derivatives, two spatial ones and
one with respect to time, we expect its solution to contain three degrees of
freedom, that is, constants of integration. Consequently, it is necessary to
supply three auxiliary constraints along with this equation to get a unique
solution.
Regarding our heat conducting rod, we must thus know how heat flows
through the ends of the rod, and we must know the temperature distribution
within the rod at some initial time.
If the ends of the rod are perfectly insulated, so that the heat flux across
these at x = 0 and x = L are zero, then by (3.1.17) we have the boundary
conditions
ux (0, t) = 0, ux (L, t) = 0, t > 0. (3.1.20)
On the other hand, if the ends of the rod are held at the fixed temperature
zero, e.g., by submerging them into ice water, then we have the boundary
conditions
u(0, t) = 0, u(L, t) = 0, t > 0. (3.1.21)
Either of these boundary conditions can be inhomogeneous, i.e., have a non-
zero right hand side, and we could of course also have mixed conditions,
where one end of the rod is insulated and the other end is held at a constant
temperature.
A boundary condition that specifies the value of the solution is called a
Dirichlet condition, while a condition that specifies the value of a derivative
is called a Neumann condition. If the boundary condition is of mixed type,
then it is called a Robin condition.
To completely determine the temperature as a function of time t and
space x, we must also specify the temperature profile at some initial time t0
via a so-called initial condition
u(x, t0 ) = f (x). (3.1.22)
52 CHAPTER 3. PARTIAL DIFFERENTIAL EQUATIONS

3.2 Separation of Variables

3.2.1 Overview of Method
Separation of variables is one of the oldest4 analytical methods for solving
PDE-problems. In its traditional form, however, it can only be applied to
linear homogeneous equations subject to homogeneous boundary conditions.
The basic assumption of separation of variables is that the solution u
of a linear homogeneous equation involving the variables x and t, say, can
be written as a product of two functions X and T , each of which is only
dependent upon a single independent variable, i.e.,

u(x, t) = X(x)T (t). (3.2.1)

Separation of variables is really best grasped by example, so instead of

discussing the general ideas of this technique any further, let us apply it to
a specific problem.

3.2.2 The Heat Equation

Recall that the temperature u within a laterally insulated rod of length L,
which has its ends kept a zero temperature, is given by the initial-boundary
value problem

ut = kuxx , 0 ≤ x ≤ L, t ≥ 0, (3.2.2)
u(0, t) = u(L, 0) = 0, (3.2.3)
u(x, 0) = f (x). (3.2.4)

Substituting the ansatz u(x, t) = X(x)T (t) into (3.2.2) we get

X(x)T ′ (t) = kX ′′ (x)T (t). (3.2.5)

Dividing both sides by kX(x)T (x) we then obtain

1 T ′ (t) X ′′ (x)
= . (3.2.6)
k T (t) X(x)
Here, the left hand side is a function of t alone, whereas the right hand side
is a function of x alone. But since x and t are independent variables, this is
4
Usually credited to J. Bernoulli who developed it around 1755.
3.2. SEPARATION OF VARIABLES 53

impossible, unless both sides are equal to a constant, λ say. Hence, we arrive
at
1 T ′ (t) X ′′ (x)
= = λ, (3.2.7)
k T (t) X(x)
where λ is called the separation constant. Rearranging, we have

T ′ = kλT, X ′′ = λX, (3.2.8)

which is two simple ordinary differential equations for X and T that can be
solved by any elementary method. Solving the ODE for T by inspection, we
have
T (t) = Cekλt , (3.2.9)
where C is arbitrary. Contemplating this result, we see that T (t) increases
exponentially with time, if λ ≥ 0. However, without any additional heating
mechanism, such a behavior is obviously unphysical for the temperature u
of the rod. Therefore, we choose the separation constant negative. To force
this we let
λ = −µ2 . (3.2.10)
As a result, the ODE for X is then

X ′′ (x) + µ2 X(x) = 0, (3.2.11)

Thus,
X(x) = A cos µx + B sin µx, (3.2.12)
where A and B are constants of integration. Multiplying X and T , we get
2
u(x, t) = X(x)T (t) = e−µ t (A cos µx + B sin µx). (3.2.13)

Although these functions do satisfy the heat equation ut = kuxx , they do

not satisfy the prescribed boundary conditions u(0, t) = u(L, t) = 0. So, the
next step is to enforce these by specifying A, B, and µ suitably. In doing, so
we find
2 2
u(0, t) = e−µ t (A cos 0 + B sin 0) = e−µ t A = 0, (3.2.14)

i.e., A = 0, and
2
u(L, t) = Be−µ t sin µL = 0, (3.2.15)
54 CHAPTER 3. PARTIAL DIFFERENTIAL EQUATIONS

which implies sin µL = 0. Consequently, we pick

µL = ±π, ±2π, ±3π, . . . , (3.2.16)

or, equivalently
n2 π 2
λ=− , n = 1, 2, 3, . . . (3.2.17)
L2

We have thus obtained a sequence of solutions of the form

2 n2 t/L2 nπx
un (x, t) = Bn e−kπ sin , n = 1, 2, 3 . . . , (3.2.18)
L

but, since (3.2.2) obeys the superposition principle, its general solution is
given by a linear combination of these individual solutions un , that is to say,

∞
X 2 n2 t/L2 nπx
u(x, t) = Bn e−kπ sin . (3.2.19)
n=1
L

Finally, the initial condition implies

∞
X nπx
u(x, 0) = Bn sin = f (x), (3.2.20)
n=1
L

and the problem is to choose the numbers Bn , so that the series for u(x, 0)
equals f (x). However, recognizing (3.2.20) as a Fourier sine series, we know
that this can be done by choosing Bn as

L
2 nπx
Z
Bn = f (x) sin dx. (3.2.21)
L 0 L

1
As a particular example, let us consider the case L = 1, k = 10
, and
3.2. SEPARATION OF VARIABLES 55

f (x) = x(1 − x). We get

Z 1
Bn = 2 x(1 − x) sin nπx dx
0
Z 1
2  1 2
= − x(1 − x) cos nπx 0 + (1 − 2x) cos nπx dx
nπ nπ 0
Z 1
2
= (1 − 2x) cos nπx dx
nπ 0
Z 1
2  1 2
= (1 − 2x) sin nπx 0 + (−2) sin nπx dx
(nπ)2 (nπ)2 0
Z 1
4
=− sin nπx dx
(nπ)2 0
4  1 (−1)n − 1
=− − cos nπx 0
= 4 . (3.2.22)
(nπ)3 (nπ)3
Hence, the final solution is given by
∞
4 X (−1)n − 1 −n2 π2 t/10
u(x, t) = 3 e sin nπx. (3.2.23)
π n=1 n3

Problem 20. Solve formally by separation of variables the heat equation

ut = uxx , 0 < x < 1, t > 0,

u(0, t) = u(1, t) = 0,
u(x, 0) = 1.

Problem 21. Solve the homogeneous heat equation

ut = 7uxx , 0 < x < π, t > 0,

u(0, t) = u(π, t) = 0,
u(x, 0) = 3 sin 2x − 6 sin 5x.

Problem 22. Solve the homogeneous heat equation

uxx = 4ut , 0 < x < 1, t > 0,

u(0, t) = u(1, t) = 0,
u(x, 0) = min{x, 1 − x}.
56 CHAPTER 3. PARTIAL DIFFERENTIAL EQUATIONS

Problem 23. Solve the homogeneous heat equation

uxx = ut + u, 0 < x < 1, t > 0,

u(0, t) = u(1, t) = 0,
u(x, 0) = sin πx.

Problem 24. Solve the heat equation

uxx = ut + u, 0 < x < 1, t > 0,

u(0, t) = u(1, t) = 0,
u(x, 0) = 1.

Example 17. Solve the heat equation

ut = uxx , 0 < x < 1, t ≥ 0, (3.2.24)

ux (0, t) = ux (1, 0) = 0, (3.2.25)
(
0, 0 < x < 21
u(x, 0) = f (x) = (3.2.26)
1, 21 < x < 1.

Solution. Put u(x, t) = X(x)T (t) and plug it into (3.2.24). We get

T ′ (t) X ′′ (x)
= = λ, (3.2.27)
T (t) X(x)

where we choose the separation constant λ = −µ2 . Solving the differential

equations T ′ (t) = −µ2 T (t) and X ′′ (x)+µ2 X(x) = 0, for X(x) and T (t) yields
2
T (t) = Ce−µ t and X(x) = A cos µx + B sin µx. Hence, we have solutions of
the form
2
u(x, t) = X(x)T (t) = e−tµ (A cos µx + B sin µx). (3.2.28)

Further, the boundary conditions gives us

2
ux (0, t) = µBe−tµ = 0, (3.2.29)

i.e., B = 0, and
2
ux (1, t) = −Ae−tµ sin µ = 0, (3.2.30)
3.2. SEPARATION OF VARIABLES 57

which implies sin µ = 0. Thus, we must choose µ = ±nπ, n = 0, 1, 2, . . . to

get a non-trivial solution. Moreover, by the superposition principle we have
∞
X 2 n2 t
u(x, t) = An e−π cos nπx.
n=1

This last result suggests that we should expand the initial condition f (x)
using a cosine series. In doing so, it is better to write the solution u(x, t),
viz.,
∞
X 2 n2 t
u(x, t) = 12 A0 + An e−π cos nπx.
n=1

that is, to redefine A0 . We then have, for n = 0

Z 1 Z 1
A0 = 2 f (x) dx = 2 dx = 1, (3.2.31)
0 1/2

and for n = 1, 2, 3, . . .,
Z 1 Z 1
An = 2 f (x) cos nπx dx = 2 cos nπx dx
0 1/2

 1 −1, if n = 1, 5, 9, . . .
sin nπx 2 
=2 = 0, if n = 2, 4, 6, . . . (3.2.32)
nπ 1/2 nπ 
1, if n = 3, 7, 11, . . .


Hence, the final solution is given by

1 2 −π2 t 2 −9π2 t
u(x, t) = − e cos πx + e cos 3πx
2 π 3π
2 −25π2 t 2 −49π2 t
− e cos 5πx + e cos 7πx − . . . . (3.2.33)
5π 7π
Problem 25. Solve the heat equation

uxx = ut , 0 < x < 1, t > 0

ux (0, t) = ux (1, t) = 0,
u(x, 0) = x.
58 CHAPTER 3. PARTIAL DIFFERENTIAL EQUATIONS

3.2.3 The Wave Equation

Let us illustrate separation of variables for something other than the heat
equation.
The one-dimensional wave equation is a model of the small transverse
oscillations of a string stretched between two points, e.g., a vibrating violin
string. We want to find a function u(x, t) that gives the deflection of the
string at any point x, 0 ≤ x ≤ L, where L is the string length, and at any
time t > 0.
For simplicity, we assume that the string is perfectly flexible, made of a
single homogeneous material, and of constant tension, so that the effects of
gravity can be ignored. Taking the above assumptions into consideration,
the motion of the string can be modeled by the following initial-boundary
value problem.
utt = c2 uxx , 0 < x < L, t > 0, (3.2.34)
u(0, t) = u(L, t) = 0, (3.2.35)
u(x, 0) = f (x), (3.2.36)
ut (x, 0) = g(x). (3.2.37)
Here, the constant c > 0 is the wave speed of the string and depends upon
the linear density and tension of the string. The fact that the string is fixed
at the endpoints x = 0 and x = L is reflected by the boundary conditions
(3.2.35). Furthermore, the initial displacement u(x, 0) of the string and the
initial velocity ut (x, 0) at each point along it are specified by the functions
f (x) and g(x), respectively.
Now applying separation of variables u(x, t) = X(x)T (t) to (3.2.34) we
get
T ′′ (t)X(x) = c2 T (t)X ′′ (x). (3.2.38)
Dividing through by c2 X(x)T (t) we have
1 T ′′ (t) X ′′ (x)
= , (3.2.39)
c2 T (t) X(x)
but, since a function of x alone that equals a function of t alone must be a
constant, we find
1 T ′′ (t) X ′′ (x)
= = λ. (3.2.40)
c2 T (t) X(x)
3.2. SEPARATION OF VARIABLES 59

We then put λ = −µ2 < 0, to obtain

T ′′ + (cµ)2 T = 0, X ′′ + µ2 X = 0, (3.2.41)
and thus
T (t) = A cos µct + B sin µct, X(x) = C cos µx + D sin µx, (3.2.42)
where A, B, C, and D are arbitrary constants. Hence,
u(x, t) = (A cos µct + B sin µct)(C cos µx + D sin µx).
As with the heat equation, the boundary conditions u(0, t) = u(L, t) = 0,
imply that C = 0, and that µ = nπ/L. Consequently, we obtain the series
solution
∞  
X nπct nπct nπx
u(x, t) = An cos + Bn sin sin . (3.2.43)
n=1
L L L
Finally, substituting the series for u(x, t) into the initial conditions gives
∞
X nπx
u(x, 0) = An sin = f (x), (3.2.44)
n=1
L
∞
X nπc nπx
ut (x, 0) = Bn sin = g(x). (3.2.45)
n=1
L L
For An and Bn , the following formulas apply
2 L nπx
Z
An = f (x) sin dx. (3.2.46)
L 0 L
Z L
2 nπx
Bn = g(x) sin dx. (3.2.47)
nπc 0 L
Problem 26. Solve, by separation of variables, the wave equation
utt = 4uxx , 0 < x < π, t > 0,
u(0, t) = u(π, t) = 0,
u(x, 0) = sin 3x, ut (x, 0) = 2 sin 4x.
Problem 27. Solve the homogeneous wave equation
utt = uxx , 0 < x < 1, t > 0,
u(0, t) = ux (1, t) = 0,
ut (x, 0) = 0, u(x, 0) = x2 − 2x.
60 CHAPTER 3. PARTIAL DIFFERENTIAL EQUATIONS

3.2.4 Inhomogeneous Equations

If either the PDE or the boundary conditions are inhomogeneous, then the
classical separation of variables method will not work. In such cases, one
technique that is often successful is to break the desired solution into two
parts, namely, one that is a function of both independent variables and one
that is a function of only one of these. More specific, one can often convert
the original problem into a homogeneous PDE with homogeneous boundary
conditions that can be solved by separation of variables, and a common ODE
that can be solved by traditional methods.
As a particular example, consider heat flow within a laterally insulated
rod of length L. Let the left and right ends of the rod be attached to a
thermostat, so that the temperature at x = 0 and x = L is fixed at α and
β degrees, respectively. Moreover, let also a heater be attached to the rod,
so that a constant heat q is added to the rod per unit length. If the initial
temperature is zero, then the situation is governed by the initial-boundary
value problem

ut = kuxx + q, 0 < x < L, t > 0, (3.2.48)

u(0, t) = α, u(L, t) = β, (3.2.49)
u(x, 0) = 0. (3.2.50)

Following the above guidelines, we try to solve this problem by writing

the temperature u(x, t) as the sum of a transient solution v(x, t) and a steady
state solution s(x)
u(x, t) = v(x, t) + s(x). (3.2.51)
Substituting this expression into (3.2.48) yields

vt = kvxx + ks′′ + q, (3.2.52)

so if we let
q
s′′ (x) = − , (3.2.53)
k
then v(x, t) satisfies the homogeneous equation vt = kvxx . However, we also
desire homogeneous boundary conditions for v(x, t). To accomplish this, we
note that

u(0, t) = v(0, t) + s(0) = α, (3.2.54)

u(L, t) = v(L, t) + s(L) = β, (3.2.55)
3.2. SEPARATION OF VARIABLES 61

which suggest that we put s(0) = α and s(L) = β, to get v(0, t) = 0 and
v(L, t) = 0, as desired.
As a last step, the ansatz (3.2.51) is inserted into the initial condition,
i.e.,
u(x, 0) = v(x, 0) + s(x) = 0. (3.2.56)
yielding the initial condition for the v(x, t) problem as
v(0, t) = −s(x). (3.2.57)
Hence, the original inhomogeneous problem has been converted into two
problems, each of which is straightforward to solve. First, s(x) is found by
solving
q
s′′ (x) = − , s(0) = α, s(L) = β. (3.2.58)
k
Next, v(x, t) is determined by applying separation of variables to
vt = vxx , 0 < x < L, t > 0, (3.2.59)
v(0, t) = 0, v(L, t) = 0, (3.2.60)
v(x, 0) = −s(x). (3.2.61)
Finally, the solutions to the two separate problems given by (3.2.58) and
(3.2.59) are added to give the desired space-time temperature distribution,
u(x, t).
Problem 28. Solve the heat equation
ut = uxx + sin 3πx, 0 < x < 1, t > 0,
u(0, t) = 0, u(1, t) = 0,
u(x, 0) = sin πx.
Problem 29. Solve the problem
ut = uxx + sin πx, 0 < x < 1, t > 0,
u(0, t) = 0, u(1, t) = 0,
u(x, 0) = 1.
Problem 30. Solve the wave equation
uxx = utt − 1, 0 < x < 1, t > 0,
u(0, t) = 0, u(1, t) = 1,
u(x, 0) = x, ut (x, 0) = 0.
62 CHAPTER 3. PARTIAL DIFFERENTIAL EQUATIONS
Appendix A

Answers to Exercises

Problem 1
a. 1/s2 b. 2/s3
c. 6/s4 d. n!/sn+1
e. 1/s2 + 1/s f. 2/s3 − 2/s2 + 1/s
g. 24/s5 + 24/s4 + 12/s3 + 4/s2 + 1/s h. do not exist
i. 1/(s + 1) j. e4 /(s − 3)
k. 1/(s − 1)2 l. do not exist
2
m. s/(s − 1) n. s/(s2 + 1)
o. 2/(s2 + 4) p. 2/(s(s2 − 4))

Problem 2
a. (s − a)/((s − a)2 + b2 ) b. e−s /s
c. e−(s+1) /(s + 1) d. 2(3s2 + 1)/(s2 − 1)3
e. 6/(s − 1)4 f. s(2 + s)/(s2 + 2s + 2)2
g. (ω cos α + s sin α)/(s2 + ω 2) h. s/(s2 + 1/4)2
1
i. do not exist j. 2
ln(1 + 1/s2)
k. s3 /(s4 + 4) l. (2 + s2 )/(s(s2 + 4))

Problem 3

63
64 APPENDIX A. ANSWERS TO EXERCISES

1
a. e−t b. 2
sin 2t c. cos t + sin t d. sinh t

e. 3 − 2e−4t f. e−2t (1 − 2t) g. 61 t2 e3t (4t + 3) h. θ(t − 1)

Problem 4

a. 41 e−t + 34 e3t b. e−2t cos t c. 31 e2t sin 3t

d. − 61 − 15
2 −3t 3 2t
e + 10 e e. 2e−4t + e2t f. 1
2
+ 21 e−2t − e−t

Problem 5

a. y(t) = 5e−2t − e−3t

b. y = e2t − 2et

c. y(t) = 21 e−t t(t + 2)

d. y(t) = 15 e−t − 15 e−2t (cos 3t + 2 sin 3t)

e. e2t − cos 2t + 21 sin 2t

f. y = 51 (cos t − 2 sin t + 4et cos t − 2et sin t)

3 1 −4t
g. y = 2
− 10
e + 53 et

h. 1 − e−t (cos t + sin t)

Problem 6 y(t) = 2e−t/2

Problem 7

a. y(t) = 21 (e−t − te−t + cos t)

b. y(t) = (−1 + 2 cos t + sin t)e−t

c. y(t) = et + e2t + 5e−3t

 65

Problem 8
∞
1 X 1 − (−1)n
f (x) = + sin nx
2 n=1 nπ
 
1 2 1 1
= + sin x + sin 3x + sin 5x + . . .
2 π 3 5
Problem 9
∞ ∞
π X (−1)n − 1 X (−1)n+1
f (x) = + cos nx + sin nx
4 n=1 πn2 n=1
n
 
π 2 1 1
= − cos x + cos 3x + cos 5x + . . .
4 π 9 25
 
1 1 1
+ sin x − sin 2x + sin 3x − sin 4x + . . .
2 3 4
Problem 10
∞
π2 X (−1)n
x2 = +4 2
cos nx
3 n=1
n
π2
 
1 1
= + 4 − cos x + cos 2x − cos 3x + . . .
3 4 9
Let x = 0 to get
∞
X (−1)n+1 1 1 1 π2
=1− + − + ... = .
n=1
n2 4 9 16 12

Problem 11 No, it contradicts the Riemann-Lebesgue lemma.

Problem 12
∞
X (−1)n+1
x=2 sin nx
n=1
n
 
1 1 1
= 2 sin x − sin 2x + sin 3x − sin 4x + . . .
2 3 4
Use the Parseval formula to conclude that
∞
X 1 1 1 1 π2
2
= 1 + + + + . . . = .
n=1
n 4 9 16 6
66 APPENDIX A. ANSWERS TO EXERCISES

Problem 13

a. -

b.
 
8 πx 1 3πx 1 5πx 1 7πx
|x| = 1 − 2 cos + cos + cos + cos + ...
π 2 9 2 25 2 49 2

c. π 2 /8 (put x = 0)

d. -

Problem 14
a. Neither odd nor even b. Even c. Neither odd nor even

d. Even e. Even f. Neither odd nor even

Problem 15
∞
1 X (−1)n − 1
f (x) = + 2 cos nπx
2 n=1
n2 π 2

Problem 16
∞
π2 X (−1)n
f (x) = +4 2
cos nx
3 n=1
n

Problem 17

a. -
∞
X sin nx
b. f (x) = −2
n=1
n

c. π/4

Problem 18 a. Linear, b. Non-linear, c. Linear.

Problem 19
 67

a. Hyperbolic, except at the point (0, 0) where it is parabolic.

b. Hyperbolic if x < 0, parabolic if x = 0, and elliptic if x > 0.

Problem 20
∞
X 4 −n2 π2 t
u(x, t) = e sin nπx
n=1,3,5,...
nπ

Problem 21

u(x, t) = 3e−28t sin 2x − 6e−175t sin 5x

Problem 22
∞
4 X (−1)k − 1 (2k+1)2 π2 t
u(x, t) = e 4 sin(2k + 1)πx
π 2 k=0 (2k + 1)2
 
4 1
− π2 t 1 − 9 π2 t 1 − 25 π2 t
= 2 e 4 sin πx − e 4 sin 3πx + e 4 sin 5πx − . . .
π 9 25

Problem 23
2 +1) t
u(x, t) = e−(π sin πx

Problem 24
∞
X 4 −(n2 π2 +1)t
u(x, t) = e sin nπx
n=1,3,5,...
nπ

Problem 25
∞
1 4 X 1 −n2 π2 t
u(x, t) = − 2 2
e cos nπx
2 π n=1,3,5,...
n

Problem 26
1
u(x, t) = cos 6t sin 3x + sin 8t sin 4x
4
68 APPENDIX A. ANSWERS TO EXERCISES

Problem 27
∞
16 X 1
u(x, t) = − 3 3
(sin(k + 21 )π(x + t) + sin(k + 21 )π(x − t))
π k=0 (2k + 1)

Problem 28
2
−π 2 t 1 − e−9π t
u(x, t) = e sin πx + sin 3πx
9π 2
Problem 29
2t
!
4 −π2 t 1 − e−π
u(x, t) = e + sin πx
π π2
∞
4X 1 2 2
+ e−(2k+1) π t sin(2k + 1)πx
π 2k + 1
k=0

Problem 30
∞
x2 3x X 4
u(x, t) = − + − 3 3
sin nπx cos nπt.
2 2 n=1,3,5,...
π n