Você está na página 1de 39

Ma 195a.

6 Notes Mark Tolentino

1. Fourier Series

Proposition 1.1. Let m, n be non-negative integers. Then:



Z L  0 n 6= m
nπx mπx
cos cos dx = L n = m 6= 0 (1.1)
−L L L
2L n=m=0


Z L  0 n 6= m
nπx mπx
sin sin dx = L n = m 6= 0 (1.2)
−L L L
0 n=m=0

Z L
nπx mπx
sin cos dx = 0 ∀m, n (1.3)
−L L L
Proof. We only prove (1.1) and leave the others as exercises. First, by the
product-to-sum identity, we can write
 
nπx mπx 1 (n − m)πx (n + m)πx
cos cos = cos + cos .
L L 2 L L

• If n 6= m, then
L  L 
(n − m)πx
Z Z
nπx mπx 1 (n + m)πx
cos cos dx = cos + cos dx
−L L L 2 −L L L
" (n−m)πx
#L
L sin L sin (n+m)πx
L
= +
2π n−m n+m
−L
= 0.

• If n = m 6= 0, then
Z L Z L
nπx mπx nπx
cos cos dx = cos2 dx
−L L L −L L
Z L 
1 2nπx
= 1 + cos dx
2 −L L
 L
1 L 2nπx
= 2L + sin
2 2nπ L −L
= L.

• The case n = m = 0 is easy since the integrand reduces to 1.

1
Definition 1.2. A function f (x), −L < x < L, is even if f (−x) = f (x) for all
−L < x < L. A function f (x), −L < x < L, is odd if f (−x) = −f (x) for all
−L < x < L.
Proposition 1.3. 1. The product of two even functions is an even function,
the product of an odd function and an even function is an odd function,
and the product of two odd functions is an even function.
Z L
2. If f (x), −L < x < L, is an odd function, then f (x)dx = 0.
−L
Z L Z L
3. If f (x), −L < x < L, is an even function, then f (x)dx = 2 f (x)dx.
−L 0

Motivation: Recall that given an infinitely differentiable function f (x), then



X f (i) (a)
f (x) = (x − a)i
i=1
i!

in some interval containing a. The series on the right is called the Taylor series
expansion of f centered at a. This expansion of f makes use of power functions
((x − a)i ).
Questions:
• Does f have a series expansion that makes use of functions other than
power functions?
• In particular, does f have an expansion that makes use only of the sine
and cosine functions?
Definition 1.4. Let f (x), −L < x < L, be a real-valued function. The Fourier
series of f is the trigonometric series
∞ 
X nπx nπx 
A0 + An cos + Bn sin (1.4)
n=1
L L

where the coefficients are defined by


Z L
1
A0 = f (x)dx, (1.5)
2L −L
1 L
Z
nπx
An = f (x) cos dx n = 1, 2, ..., (1.6)
L −L L
1 L
Z
nπx
Bn = f (x) sin dx n = 1, 2, .... (1.7)
L −L L

2
Proposition 1.5. 1. If f (x), −L < x < L, is an even function, then the
Fourier series of f (x) has Bn = 0 for n = 1, 2, ....
2. If f (x), −L < x < L, is an odd function, then the Fourier series of f (x)
has An = 0 for n = 0, 1, 2, ....
Example 1.6. Find the Fourier series of f (x) = x, −L < x < L.
Solution. The function f (x) is odd; therefore, its Fourier series has An = 0 for
n = 0, 1, .... To compute Bn , we note that f (x) sin(nπx/L) is an even function;
thus

1 L
Z
nπx
Bn = x sin dx
L −L L
2 L
Z
nπx
= x sin dx.
L 0 L

We integrate by parts with u = x, dv = sin(nπx/L)dx; hence,


L Z L !
2 L nπx L nπx
Bn = −x cos + cos dx .
L nπ L 0 nπ 0 L

The last integral is zero, and we have Bn = −(2L/nπ) cos nπ = (2L/nπ)(−1)n+1 .


Therefore, the Fourier series of f (x) = x, −L < x < L, is

2L X (−1)n+1 nπx
sin .
π n=1 n L

Remark 1.7. Note carefully that we are not saying yet that

2L X (−1)n+1 nπx
x= sin , x ∈ (−L, L)
π n=1 n L

although this is what we intend to show when we talk about convergence later
on. In this discussion, we can only look at how the above Fourier series behave
graphically and make some intuitive observations regarding convergence. We will
make more precise statements on convergence later.
Recall: The value of a function given by an infinite series

X
f (x) = an ϕn (x)
n=0

3
is determined via the definition
`
X
f (x) = lim an ϕn (x),
`→∞
n=0

if this limit exists. Hence, the maximal domain of f (x) is the set of all x for
which the infinite series converges.
Definition 1.8. Let ϕ and ψ be two real-valued functions. The distance be-
tween ϕ and ψ on the interval [a, b] is defined as
s
Z b
d(ϕ, ψ) := (ϕ − ψ)2 dx.
a

Remark 1.9. While the complete motivation of Definition 1.8 is beyond our
scope, the definition is appropriate, in one sense, because

d(ϕ, ψ) = 0 ⇔ ϕ = ψ a.e. on [a, b];

that is, the distance between ϕ and ψ is 0 if and only if they are equal a.e. on
[a, b].
Proposition 1.10. Let f (x), −L ≤ x ≤ L, be a real-valued function and ϕ1 , ..., ϕN
be distinct functions in the collection
n nπx o n nπx o
cos : n ∈ Z+ ∪ sin : n ∈ Z+ .
L L
0πx
Moreover, set ϕ0 ≡ 1 = cos . Then the minimum of the expression
L
d (f, c0 ϕ0 + c1 ϕ1 + · · · + cN ϕN )

is attained when
Z L Z L
1 1
c0 = f dx and ci = f ϕi dx, i = 1, ..., N (1.8)
2L −L L −L

Proof.

[d (f, c0 ϕ0 + c1 ϕ1 + · · · + cN ϕN )]2
Z L
= [f − (c0 ϕ0 + c1 ϕ1 + · · · + cN ϕN )]2 dx
−L
 
Z L XN N
X X
= f 2 − 2ci f ϕi + c2i ϕ2i − ci cj ϕi ϕj  dx
−L i=1 i=1 i6=j

4
Z L
c20 − 2c0 f dx

+
−L
Z L N
X Z L N
X Z L
= f 2 dx − 2ci f ϕi dx + c2i ϕ2i dx
−L i=1 −L i=1 −L

X Z L Z L
− ci cj ϕi ϕj dx + 2c20 L − 2c0 f dx
i6=j −L −L
Z L N
X Z L N
X
2
= f dx − 2ci f ϕi dx + L c2i
−L i=1 −L i=1
Z L
+ 2c20 L − 2c0 f dx
−L
 !2 
N
2ci L L
Z Z
X
2 1
= L ci −
 f ϕi dx + 2 f ϕi dx 
i=1
L −L L −L

N
!2
Z L Z L
1X2
+ f dx − f ϕi dx
−L L i=1 −L
 !2 
Z L Z L
2 c0 1
+ 2L c0 − f dx + f dx 
L −L 2L −L
!2
Z L
1
− f dx
2L −L

N
!2 !2
Z L Z L
X 1 1
= L ci − f ϕi dx + 2L c0 − f dx
i=1
L −L 2L −L
Z L N Z L !2 Z L !2
2 1X 1
+ f dx − f ϕi dx − f dx
−L L i=1 −L 2L −L
Z L N Z L !2 Z L !2
2 1X 1
≥ f dx − f ϕi dx − f dx .
−L L i=1 −L 2L −L

Clearly, the minimum holds when (1.8) holds.


Remark 1.11.

• Proposition 1.10 gives the best “approximation” of the given function


f (x), −L ≤ x ≤ L, by means of linear combinations of ϕ1 , ..., ϕN − a
trigonometric series with a finite number of terms!
• Thus, the coefficients of a trigonometric series with a finite number of terms

5
can be chosen so that the distance between the given function f and the
trigonometric series is minimized.
• In the case of an infinite trigonometric series, our guess is that the same
coefficients will work; this motivates the definition of the coefficients of the
Fourier series.
Definition 1.12. The indicator function or characteristic function of a
subset A of a set X is the function

1, x ∈ A,
χA (x) :=
0, x 6∈ A.

Example 1.13. Find the Fourier series of

f (x) = χ(0,L) − χ(−L,0] , −L < x < L.

Solution. Note that



1, x ∈ (0, L),
f (x) =
−1, x ∈ (−L, 0].

This function is an odd function; hence, the Fourier series of f (x) has An = 0
for n = 0, 1, 2, .... For the Bn ’s, we have

1 L
Z
nπx
Bn = f (x) sin dx
L −L L
"Z #
L Z 0
1 nπx nπx
= sin dx − sin dx
L 0 L −L L
" L 0 #
1 L nπx L nπx
= − cos + cos
L nπ L 0 nπ L −L
1
= [−(cos nπ − 1) + (1 − cos nπ)]

2
= [1 − (−1)n ] .

Therefore, the Fourier series of f (x) is

X 2 nπx
[1 − (−1)n ] sin , x ∈ (−L, L).
n=1
nπ L

Example 1.14. Find the Fourier series of

g(x) = (L + x)χ(−L,0] + (L − x)χ(0,L) , −L < x < L.

6
Proof. Note that 
L + x, x ∈ (−L, 0],
g(x) =
L − x, x ∈ (0, L).
This function is an even function; hence, the Fourier series of g(x) has Bn = 0
for n = 1, 2, .... For the An ’s, we have
Z L
1
A0 = f (x)dx
2L −L
1 L
Z
= (L − x)dx
L 0
L
x2

1
= Lx −
L 2 0
L2
 
1 2
= L −
L 2
L
=
2
and
Z L
1 nπx
An = f (x) cos dx
L −L L
ZL
2 nπx
= (L − x) cos
dx
L L
"0 Z #
L Z L
2 nπx nπx
= L cos dx − x cos dx
L 0 L 0 L
 L "  2 #L
L nπx 2 xL nπx L nπx
=2 sin − sin + cos
nπ L 0 L nπ L nπ L
0
 2
2 L
=− [(−1)n − 1]
L nπ
2L
= 2 2 [(−1)n+1 + 1].
n π
Therefore, the Fourier series of f (x) is

L 2L X (−1)n+1 + 1 nπx
+ 2 cos , x ∈ (−L, L).
2 π n=1 n2 L

Example 1.15. Find the Fourier series of f (x) = ex , −L < x < L.

7
Solution. The function f (x) is neither even nor odd. We compute for the Fourier
coefficients as follows:
Z L
1 eL − e−L
• A0 = ex dx =
2L −L 2L
• For n ≥ 1,
" L #
1 L x
Z L
1 Lex
Z
nπx nπx L x nπx
e cos dx = sin − e sin dx
L −L L L nπ L −L nπ −L L
" L Z L #
1 ex L nπx L nπx
=− − cos + ex cos dx
nπ nπ L −L nπ −L L
Z L
(eL − e−L )L(−1)n L nπx
= − 2 2 ex cos dx
n2 π 2 n π −L L
By transferring the second term on the right side to the left side, we get
!
L2 1 L x (eL − e−L )L(−1)n
  Z
nπx
1+ 2 2 e cos dx = .
n π L −L L n2 π 2
Hence, for n ≥ 1, we have
(eL − e−L )L(−1)n
An = .
L2 + n2 π 2
• For n ≥ 1, we have
1 L x
Z
nπx
Bn = e sin dx
L −L L
" L Z L #
1 Lex nπx L x nπx
= − cos + e cos dx
L nπ L −L nπ −L L
!
(e−L − eL )(−1)n 1 L x
Z
L nπx
= + e cos dx
nπ nπ L −L L
(e−L − eL )(−1)n L
Bn = + An
nπ nπ

Example 1.16. Find the Fourier series of f (x) = |x|, −L < x < L.
Solution. The function f (x) is even; therefore, its Fourier series has Bn = 0 for
n = 1, 2, .... To compute An , we note that f (x) cos(nπx/L) is an even function;
thus for n ≥ 1,
1 L
Z
nπx
An = |x| cos dx
L −L L

8
Z L
2 nπx
= x cos dx.
L 0 L
We integrate by parts with u = x, dv = cos(nπx/L)dx; hence,
L Z L !
2 L nπx L nπx
An = x sin − sin dx
L nπ L 0 nπ 0 L
 2 !
2 L nπx L
= 0+ cos
L nπ L 0

2L
= [(−1)n − 1]
n2 π 2
for n ≥ 1. For n = 0, we have
Z L Z L
1 1 L
A0 = |x|dx = xdx = .
2L −L L 0 2

Therefore, the Fourier series of f (x) = |x|, −L < x < L, is



L 2L X (−1)n − 1 nπx
+ 2 cos .
2 π n=1 n2 L

Since (−1)n − 1 is equal to 0 for even n and is equal to −2 for odd n, we can
simply take n = 2m − 1 for m = 1, 2, ... and the series becomes

L 4L X cos[(2m − 1)πx/L]
− 2 .
2 π m=1 (2m − 1)2

Problem: How do we find the Fourier series of a function f (x) defined only the
interval 0 < x < L.
Solution: We extend f to a new function f˜ defined the interval −L < x < L;
then we get the Fourier series of f˜.
Definition 1.17. Let f (x) be a function on the interval 0 < x < L. The function

 f (x) 0<x<L
fO (x) = −f (−x) −L < x < 0
0 x=0

is called the odd extension of f to (−L, L). It is an odd function, and therefore
its Fourier coefficients are given as follows:

An = 0 n = 0, 1, ...

9
Z L Z L
1 nπx 2 nπx
Bn = fO (x) sin dx = f (x) sin dx, n = 1, 2, ... (1.9)
L −L L L 0 L

We define the Fourier sine series of f (x) as the series



X nπx
Bn sin , 0<x<L
n=1
L

where Bn is given by (1.9).


Definition 1.18. Let f (x) be a function on the interval 0 < x < L. The function

 f (x) 0<x<L
fE (x) = f (−x) −L < x < 0
0 x=0

is called the even extension of f to (−L, L). It is an even function, and therefore
its Fourier coefficients are given as follows:

Bn = 0 n = 1, 2, ...
Z L
1 L
Z
1
A0 = fE (x)dx = f (x)dx (1.10)
2L −L L 0
1 L 2 L
Z Z
nπx nπx
An = fE (x) cos dx = f (x) cos dx (1.11)
L −L L L 0 L

We define the Fourier cosine series of f (x) as the series



X nπx
A0 + An cos , 0<x<L
n=1
L

where A0 and An are given by (1.10) and (1.11).


Remark 1.19. To ensure that fO (x) is odd, it is necessary to set fO (0) = 0.
On the other hand, such a requirement is not present in ensuring that fE (x) is
even. Hence, the definition fE (0) = 0 is completely arbitrary.
For instance, we could have set fE (0) = limx→0+ f (x) if this limit exists.
Example 1.20. Find the Fourier sine and Fourier cosine series of

f (x) = xχ(0,π) , x ∈ (0, π).

Solution.
1. Fourier sine:

10
Using formula (1.9) and by our computations in Example 1.6, we have

2 π
Z
2
Bn = x sin nxdx = (−1)n+1 .
π 0 n

Therefore, the Fourier sine series of f (x) is



X 2
(−1)n+1 sin nx.
n=1
n

2. Fourier cosine:
Using formulas (1.10) and (1.11) and by our computations in Example 1.16,
we have
1 π
Z
π
A0 = xdx =
π 0 2
and Z π
2 2
An = x cos nxdx = [(−1)n − 1].
π 0 n2 π
Therefore, the Fourier cosine series of f (x) is

π X 2
+ [(−1)n − 1] cos nx,
2 n=1 n2 π

or equivalently,

π 4 X cos[(2m − 1)x]
− .
2 π m=1 (2m − 1)2

Remark 1.21. Note that the Fourier sine and Fourier cosine series in the preced-
ing example are the same as the ones we obtained in Example 1.6 and Example
1.16. This is because the odd extension of f (x) is g(x) = x, −π < x < π while
the even extension of f (x) is h(x) = |x|, −π < x < π.
Example 1.22. Find the Fourier sine and Fourier cosine series of

f (x) = χ(−π,−π/2) , x ∈ (−π, 0).

Solution. Note that



1, −π < x < −π/2,
f (x) =
0, −π/2 ≤ x < 0.

11
1. Fourier sine:
The odd extension of f (x) is

 1, −π < x < −π/2,
g(x) = 0, −π/2 ≤ x < π/2,
−1, π/2 ≤ x < π.

We now compute the Fourier series of g(x). Since g(x) is odd, An = 0 for
all n = 0, 1, 2, .... For the Bn0 s, we have
1 π
Z
Bn = g(x) sin nxdx
π −π
Z π
2
= g(x) sin nxdx
π 0
Z π
2
= (−1) sin nxdx
π π/2
 π
2 1
= cos nx
π n π/2
2 h nπ i
= (−1)n − cos .
nπ 2
Therefore, the Fourier series of g(x) is

X 2 h nπ i
(−1)n − cos sin nx.
n=1
nπ 2

This is the Fourier sine series of f (x).


2. Fourier cosine:
The even extension of f (x) is

 1, −π < x < −π/2,
h(x) = 0, −π/2 ≤ x < π/2,
1, π/2 ≤ x < π.

We now compute the Fourier series of h(x). Since h(x) is even, Bn = 0 for
all n = 1, 2, .... For A0 , we have
Z π
1
A0 = h(x)dx
2π −π
1 π
Z
= h(x)dx
π 0
Z π
1
= 1dx
π π/2

12
1 π
= x|
π π/2
1
= .
2
For the A0n s, we have

1 π
Z
Bn = h(x) cos nxdx
π −π
Z π
2
= h(x) cos nxdx
π 0
Z π
2
= (1) cos nxdx
π π/2
 π
2 1
= sin nx
π n π/2
2 h nπ i
= 0 − sin .
nπ 2
Therefore, the Fourier series of h(x) is

1 X nπ 
− sin cos nx.
2 n=1 2

This is the Fourier sine series of f (x).

Example 1.23. Find the Fourier sine and Fourier cosine series of

f (x) = xχ[1,2) , x ∈ (0, 2).

Solution. Note that 


0, 0 < x < 1,
f (x) =
x, 1 ≤ x < 2.

1. Fourier sine: Using formula (1.9), we have

2 2
Z
nπx
Bn = f (x) sin dx
2 0 2
Z 2
nπx
= x sin dx
1 2
2 h nπx i2 4 h nπx i2
=− x cos + 2 2 sin
nπ 2 1 n π 2 1
2 h nπ i 4 h nπ i
=− 2(−1)n − cos + 2 2 0 − sin
nπ 2 n π 2

13
2 h nπ i 4 nπ
=− 2(−1)n − cos − 2 2 sin .
nπ 2 n π 2
Therefore, the Fourier sine series of f (x) is
∞  
X 2 h n nπ i 4 nπ nπx
− 2(−1) + cos − 2 2 sin sin .
n=1
nπ 2 n π 2 2

2. Fourier cosine: Using formulas (1.10) and (1.11), we have


Z 2
1
A0 = f (x)dx
2 0
Z 2
1
= xdx
2 1
 2
1 x2
=
2 2 1
 
1 1
= 2−
2 2
3
=
4
and
2 2
Z
nπx
An = f (x) cos dx
2 0 2
Z 2
nπx
= x cos dx
1 2
2 h nπx i2 4 h nπx i2
= x sin + 2 2 cos
nπ 2 1 n π 2 1
2 h nπ i 4 h nπ i
= − sin + 2 2 (−1)n − cos
nπ 2 n π 2
2 nπ 4 h nπ i
=− sin + 2 2 (−1)n − cos .
nπ 2 n π 2
Therefore, the Fourier cosine series of f (x) is
∞  
3 X 2 nπ 4 h n nπ i nπx
+ − sin + 2 2 (−1) − cos cos .
4 n=1 nπ 2 n π 2 2

14
2. Convergence of Fourier Series
Definition 2.1. The partial sum of order N of a trigonometric series (1.4)
as the series
N 
X nπx nπx 
A0 + An cos + Bn sin . (2.1)
n=1
L L
In this section, we will discuss when a function is equal to its Fourier series;
that is, we will discuss when the equality

X nπx nπx
f (x) = A0 + An cos + Bn sin
n=1
L L

occurs. We do this by analyzing the partial sum of order N as N → ∞.


Definition 2.2. A function f (x), a < x < b, is piecewise continuous if there
is a finite set of points a = x0 < x1 < · · · < xp < xp+1 = b such that
f is continuous at x 6= xi , i = 1, ..., p,
lim f (xi + ε) exists, i = 0, ..., p, (2.2)
ε→0+
lim f (xi − ε) exists, i = 1, ..., p + 1. (2.3)
ε→0+

The limit (2.2) is denoted f (xi +) and is called the right-hand limit. Likewise,
the limit (2.3) is denoted f (xi −) and is called the left-hand limit.
Remark 2.3. A function is piecewise continuous if it has only finitely many
discontinuities, all of which are removable or jump discontinuities. Since Fourier
coefficients are defined as integrals and since integrals are insensitive to the func-
tion values at a finite number of points, these discontinuities do not pose a
difficulty when computing the Fourier series of a piecewise smooth function.
Definition 2.4. A function f (x), a < x < b, is said to be piecewise smooth
if f and all of its derivatives are piecewise continuous.
If f (x), a < x < b, is piecewise smooth, then f 0 (x) exists except for x =
x1 , ..., xp . This is the piecewise derivative of f .
Example 2.5. Determine if the function f (x) = |x|, −π < x < π, is piecewise
smooth.
Solution. Clearly, f (x) is continuous on (−π, π). We observe that

−1, x<0
f 0 (x) =
1, x>0
and that f 0 (0) does not exist; however, f 0 (0−) = −1 and f 0 (0+) = 1. Hence,
f 0 (x) is piecewise continuous on (−π, π). Moreover, all higher derivatives are zero
(except at x = 0 where they don’t exist) so these are also piecewise continous on
(−π, π). Therefore, f (x) is piecewise smooth.

15
Example 2.6. Determine if the function

x3 ,

−π < x < 1
f (x) = 3
x − 1, 1≤x<π

is piecewise smooth.

Figure 1: f (x) = x3 if −π < x < 1 and f (x) = x3 − 1 if 1 ≤< x < π

Solution. Clearly, f (x) is discontinuous at x = 1 but f (1−) = 1 and f (1+) = 0


exist; hence, f (x) is piecewise continuous on (−π, π). It is easy to see that for
x ∈ (−π, 1) ∪ (1, π),

f 0 (x) = 3x2 , f 00 (x) = 6x, f 000 (x) = 6

and
f (n) ≡ 0 ∀n ≥ 4.
Thus, all derivatives of f (x) are piecewise continuous on (−π, π). Therefore,
f (x) is piecewise smooth.
1
Example 2.7. Determine if the function f (x) = x2 sin , −π < x < π, is
x
piecewise smooth.

Solution. Clearly, f (x) is discontinuous at x = 0 but by squeeze theorem, f (0−) =


f (0+) = 0; hence, f (x) is piecewise continuous on (−π, π). The first derivative
of f is given by
1 1
f 0 (x) = − cos + 2x sin ,
x x
which is defined for all x 6= 0. However, f 0 (0−) and f 0 (0+) do not exist so f (x)
is not piecewise smooth on (−π, π).

16
1
Figure 2: f (x) = x2 sin
x

Definition 2.8. Suppose f (x), −L < x < L, is piecewise smooth. We define


the 2L-periodic extension of f as the function

f¯(x) = f (x − 2nL)

where n is an integer, dependent on x, so that x − 2nL ∈ (−L, L).


Example 2.9. The periodic extension f¯ of the function f (x) = x, −L < x < L,
is shown in the figure below.

Example 2.10. The periodic extension ḡ of the function g(x) = |x|, −L < x <
L, is shown in the figure below.

17
Theorem 2.11. Let f (x), −π < x < π, be piecewise smooth. Then the Fourier
series of f converges for all x to the value
1 ¯
f (x + 0) + f¯(x − 0) ,

(2.4)
2
where f¯ is the 2π-periodic extension of f .
Let us analyze the implications of Theorem 2.11. First, we focus on the
interval −π < x < π. Let −π = a < x1 < · · · < xp < xp+1 = π be subdivision
points of the interval (−π, π). First, note that if f¯ is continuous at a point x,
then (2.4) is equal to f¯(x). Since f¯ = f on (−π, π) and since f is continuous on
(−π, π) except at x1 , ..., xp , we conclude that the Fourier series of f converges
to f (x) for all x ∈ (−π, π) \ {x1 , ..., xp }. Moreover, at each xi , the Fourier series
converges to the average of the one-sided limits at xi .
Furthermore, from periodicity, we see that

f¯(−π − 0) = f¯(π − 0) = f (π − 0)

and
f¯(π + 0) = f¯(−π + 0) = f (−π + 0).
Therefore,
1 ¯ 1
[f (−π − 0) + f¯(−π + 0)] = [f (π − 0) + f (−π + 0)],
2 2
1 ¯ 1
[f (π − 0) + f¯(π + 0)] = [f (π − 0) + f (−π + 0)];
2 2
that is, the average of the left- and right-hand limits of the periodic extension
at the endpoints agrees with the common average of the original function at the
endpoints.
Using the above observations, we have the following remark.
Remark 2.12. Let f (x), −π < x < π, be piecewise smooth.
• The Fourier series of f converges to f (x) for all x in (−π, π) if f is contin-
uous on the entire interval (−π, π).
• The Fourier series is continuous on R if f¯ is continuous everywhere; in
particular, f must be continuous on [−π, π] and must satisfy f (−π) = f (π).

18
Example 2.13. Recall that the Fouries of f (x) = x, −L < x < L, is

2L X (−1)n+1 nπx
F (x) = sin .
π n=1 n L

By Theorem 2.11, we have

f¯(x),

x 6= (2n + 1)L, n ∈ Z,
F (x) =
0, x = (2n + 1)L, n ∈ Z.

Observe that F (x) is defined everywhere but has jump discontinuities at odd
multiples of L. Moreover, the Fourier series of f (x) equals f (x) in its domain
−L < x < L.
Example 2.14. Recall that the Fourier series of f (x) = |x|, −L < x < L, is

L 4L X cos[(2m − 1)πx/L]
F (x) = − 2 .
2 π m=1 (2m − 1)2

By Theorem 2.11, we have

f¯(x),

x 6= (2n + 1)L, n ∈ Z,
F (x) =
L, x = (2n + 1)L, n ∈ Z.

Observe that F (x) is continuous everywhere; this is because f is continuous on


[−L, L] and satisfies f (−L) = f (L).
We can now use Fourier series to obtain various numerical series.
Example 2.15. The Fourier series of f (x) = |x|, −π < x < π, is

π 4 X cos(2m − 1)x
F (x) = − .
2 π m=1 (2m − 1)2

By Theorem 2.11, F (0) = f (0); that is,

π2
 
π 4 1 1 1 1
− 1+ + + ··· = 0 ⇔ 1+ + + ··· = .
2 π 9 25 9 25 8
Example 2.16. Show that

π2 4 4
x2 = − 4 cos x + cos 2x − cos 3x + · · · + (−1)m 2 cos mx + · · ·
3 9 m
for −π ≤ x ≤ π. Then evaluate the following sums:

X 1
1. (−1)m+1
m=1
m2

19

X 1
2. 2
m=1
m
X 1
3.
m2
m odd

Solution. Recall that the Fourier series of f (x) = x2 , −L < x < L, is



L2 4L2 X (−1)n nπx
+ 2 cos .
3 π n=1 n2 L

Setting L = π, we can use Theorem 2.11 to conclude that f (x) equals this
Fourier series on (−π, π). Furthermore, equality also occurs at the endpoints
because f (x) is continuous on [−π, π] and f (π) = f (−π).
1. Set x = 0.
π2 4 4
0= − 4 + 1 − + · · · + (−1)m 2 + · · ·
3 9 m
4 4 π2
⇔ 4 − 1 + − · · · − (−1)m 2 + · · · =
9 m 3
π2

1 1 1
⇔ 4 1 − + − · · · − (−1)m 2 + · · · =
4 9 m 3
2
1 1 1 π
⇔ 1 − + − · · · − (−1)m 2 + · · · =
4 9 m 12

2. Set x = π.
π2 4 4
π2 = + 4 + 1 + + · · · + (−1)m 2 + · · ·
3 9 m
4 4 π2
⇔ 4 + 1 + + · · · + (−1)m 2 + · · · = π 2 −
 9 m  3 2
1 1 m 1 2π
⇔ 4 1 + + + · · · + (−1) + ··· =
4 9 m2 3
2
1 1 1 π
⇔ 1 + + + · · · + (−1)m 2 + · · · =
4 9 m 6

3. Observe that
∞ ∞
" #
X 1 1 X
m+1 1 X 1
2
= (−1) + .
m 2 m=1
m2 m=1 m2
m odd

It follows that
1 π2 π2 π2
 
X 1
2
= + = .
m 2 12 6 8
m odd

20
Consider the function

−1, −π < x < 0
f (x) =
1, 0≤x<π

The Fourier series of f (x) is


∞  
X 2 4 sin 3x sin 5x
F (x) = [1 − (−1)n ] sin nx = sin x + + + ··· .
n=1
nπ π 3 5

By the convergence theorem, we have

F (x) = f (x), x ∈ (−π, 0) ∪ (0, π),


1 (2.5)
F (0) = [f (0 + 0) + f (0 − 0)] = 0.
2
Set
N
X 2
FN (x) = [1 − (−1)n ] sin nx.
n=1

Then we can write (2.5) in terms of FN as follows:

lim FN (x) = f (x), x ∈ (−π, 0) ∪ (0, π),


N →∞
1
lim FN (0) = [f (0 + 0) + f (0 − 0)] = 0.
N →∞ 2
The first of the above equations describes what is called pointwise convergence;
in this case, we say that FN (x) converges pointwise to f (x) for all x on (−π, 0) ∪
(0, π).
From the graph (use GeoGebra) of f (x) and FN (x), we see that, just before
the discontinuity at x = 0, the partial sums overshoot the right- and left-hand
limits then slope rapidly toward their mean. Observe (again, using GeoGebra)
that this overshoot persists even for large values of N . This behavior is known
as the Gibbs phenomenon.
Hence, even though FN (x) converges pointwise to f (x) on the interval (−π, 0)
∪(0, π), there is always a point on the graph of FN (x) that is “far” from the graph
of f (x). In this case, we say that FN (x) does not converge uniformly to f (x) on
the interval (−π, 0) ∪ (0, π).
Definition 2.17. A sequence of functions FN (x), a < x < b, converges uni-
formly to a function f (x) if

|Fn (x) − f (x)| ≤ εn ∀x ∈ (a, b), n = 1, 2, ...,

where lim εn = 0.
n→∞

21
The above condition is clearly violated in the Gibbs phenomenon discussed
earlier because for x = π/(N + 1), we have seen that Fn (x) is “far” from f (x).
In many problems, it is important that we avoid the Gibbs phenomenon; that
is, we want the Fourier series to converge uniformly. In this section, we present
without proof two criteria for uniform convergence.
Proposition 2.18 (First criterion for uniform convergence). Let f (x), −L <
x < L, be a piecewise smooth function. The Fourier coefficients satisfy

X
(|An | + |Bn |) < ∞
n=1

if and only if the Fourier series of f (x) converges uniformly on (−L, L).
Example 2.19. Let f (x) = |x| over the interval (−π, π). The sum of the
(absolute value of the) Fourier coefficients of f (x) is

π 4 X 1 π 4 π2
+ 2
= + × < ∞;
2 π m=1 (2m − 1) 2 π 8

hence, the Fourier series of f (x) = |x| converges uniformly on (−π, π).
Example 2.20. Let f (x) = x2 over the interval (−L, L). The sum of the
(absolute value of the) Fourier coefficients of f (x) is

L2 4L2 X 1 L2 4L2 π2
+ 2 2
= + 2 × < ∞;
3 π n=1 n 3 π 6

hence, the Fourier series of f (x) = x2 converges uniformly on (−L, L).


Proposition 2.21 (Second criterion for uniform convergence). Let f (x), −L <
x < L, be a piecewise smooth function. The functionf is continuous on (−L, L)
and f (−L + 0) = f (L − 0) if and only if the Fourier series of f (x) converges
uniformly on (−L, L).
Example 2.22. Let f (x) = x over the interval (−L, L). The sum of the (abso-
lute value of the) Fourier coefficients of f (x) is

2L X 1
.
π n=1 n

We know that this sum (the harmonic series) is divergent. So we can conclude
immediately that the Fourier series of f (x) does not converge uniformly. Also,
even though f is continuous on (−L, L), f (−L + 0) 6= f (L − 0); hence, we get
the same conclusion.
This means that we expect the occurrence of the Gibbs phenomenon for the
Fourier series of f (x) = x. (GeoGebra.)

22
Proposition 2.23. Let f (x), −L < x < L, be a continuous piecewise smooth
function with f (L − 0) = f (−L + 0). Then

1 0 X nπ  nπx nπx 
[f (x + 0) + f 0 (x − 0)] = Bn cos − An sin .
2 n=1
L L L

Example 2.24. We can obtain the Fourier series of f (x) = x2 , −L < x < L,
using Proposition 2.23. Since f (x) is even, its Fourier series is of the form

X nπx
A0 + An cos .
n=1
L

Proposition 2.23 implies that



X An nπ nπx
2x = − sin
n=1
L L

for all −L < x < L. Since we know that



2L X (−1)n+1 nπx
x= sin ,
π n=1 n L

we obtain
∞ ∞
4L X (−1)n+1 nπx X An nπ nπx
sin =− sin .
π n=1 n L n=1
L L
The above equation gives

4L(−1)n+1 An nπ 4L2 (−1)n


=− ⇔ An = .
πn L π 2 n2
The coefficient A0 is computed using the definition, which gives
Z L
1 L2
A0 = x2 dx = .
2L −L 3

Therefore, the Fourier series of f (x) = x2 , −L < x < L, is



L2 X 4L2 (−1)n nπx
+ 2 n2
cos .
3 n=1
π L

Example 2.25. Consider the function f (x) = x3 , −L < x < L. Since f (x) is
odd, its Fourier series is of the form

X nπx
Bn sin .
n=1
L

23
Proposition 2.23 implies that

X Bn nπ nπx
3x2 = cos
n=1
L L

for all −L < x < L. Since we have obtained that



L2 X 4L2 (−1)n nπx
x2 = + 2 n2
cos ,
3 n=1
π L

we obtain
∞ ∞
212L2 X (−1)n nπx X Bn nπ nπx
L + 2 cos = cos .
π n=1 n2 L n=1
L L

We see that the left-hand side includes a constant term while the right does not;
thus, in this case, we cannot compute Bn by comparing the coefficients. We will
return to this problem in a later example.
Proposition 2.26. Let f (x), −L < x < L, be a piecewise smooth function with
Fourier series
∞ 
X nπx nπx 
A0 + An cos + Bn sin .
n=1
L L
If −L ≤ x0 < x ≤ L, then
Z x
f (u)du = A0 (x − x0 )+
x0
∞  
X LAn  nπx nπx0  LBn  nπx0 nπx 
sin − sin + cos − cos .
n=1
nπ L L nπ L L

Example 2.27. Recall that



L2 X 4L2 (−1)n nπx
x2 = + 2 n2
cos
3 n=1
π L

for all −L < x < L. Applying Proposition 2.26, we get


Z x" 2 X ∞
Z x #
2 L 4L2 (−1)n nπu
u du = + cos du;
0 0 3 n=1
π 2 n2 L

that is,

x ∞
x3 L2 X 4L3 (−1)n nπu L2 X 4L3 (−1)n nπx
= x+ 3 3
sin = x + 3 n3
sin .
3 3 π n L 3 π L


n=1 0 n=1

24
Since we also know that

2L X (−1)n+1 nπx
x= sin ,
π n=1 n L

we have
∞ ∞
2L X (−1)n+1 nπx X 12L3 (−1)n nπx
x3 = L2 × sin + 3 n3
sin
π n=1 n L n=1
π L

2L3 (−1)n+1 12L3 (−1)n
 
X nπx
= + 3 n3
sin
n=1
nπ π L

for −L < x < L.


Example 2.28. Consider the function f (x) = x4 , −L < x < L. In order to get
the Fourier series of f (x), we integrate the Fourier series of x3 as follows:
Z x
x4 x40
− = u3 du
4 4 x0
∞  3 x
2L (−1)n+1 12L3 (−1)n

X L nπu
= − × + cos
nπ nπ π 3 n3 L

n=1 x0
∞  4 n 4 n+1

X 2L (−1) 12L (−1) nπx
= 2 π2
+ 4 π4
cos
n=1
n n L

X 2L4 (−1)n 12L4 (−1)n+1
 
nπx0
− 2 π2
+ 4 π4
cos
n=1
n n L

Hence,
∞ 
8L4 (−1)n 48L4 (−1)n+1

X nπx
x4 = + cos
n=1
n2 π 2 n4 π 4 L
∞ 
8L4 (−1)n 48L4 (−1) n+1

X nπx0
+ x40 − + cos .
n=1
n2 π 2 n4 π 4 L
| {z }
A0

The value of A0 can be computed easily using the definition:


Z L
1 L4
A0 = x4 dx = .
2L −L 5
Therefore, the Fourier series of f (x) = x4 , −L < x < L, is
∞ 
L4 X 8L4 (−1)n 48L4 (−1)n+1

nπx
+ 2 π2
+ 4 π4
cos .
5 n=1
n n L

25
3. The Method of Sepration of Variables
We will need the following definitions and results.
Theorem 3.1. Let
ay 00 + by 0 + cy = 0 (3.1)
be a second-order homogeneous linear differential equation with constant coeffi-
cients where a, b, c ∈ R and a 6= 0. Let r1 and r2 be the roots of the character-
istic equation ar2 + br + c = 0. Then
1. If r1 , r2 ∈ R with r1 6= r2 , then the general solution to (3.1) is y(t) =
C1 er1 t + C2 er2 t , where C1 and C2 are arbitrary constants.
2. If r1 , r2 ∈ R with r1 = r2 , then the general solution to (3.1) is y(t) =
C1 er1 t + C2 ter1 t , where C1 and C2 are arbitrary constants.
3. If r1 , r2 = λ ± iµ with µ 6= 0, then the general solution to (3.1) is y(t) =
eλt (C1 cos µt + C2 sin µt), where C1 and C2 are arbitrary constants.
Definition 3.2. The hyperbolic sine and cosine functions are defined as
1 x 1 x
sinh x = (e − e−x ), cosh x = (e + e−x ).
2 2
Remark 3.3. The following are properties that follow from the definition:
• ex = cosh x + sinh x, e−x = cosh x − sinh x
• sinh x is odd while cosh x is even.
d d
• sinh x = cosh x, cosh x = sinh x
dx dx
• cosh2 x − sinh2 x = 1
• sinh x has a unique root at x = 0; cosh x is strictly positive
Heat Equation: Homogeneous Boundary Conditions
Consider the following problem involving the heat equation.
ut − Kuxx = 0 0 < x < L, t > 0, (3.2)
u(0, t) = u(L, t) = 0 t ≥ 0, (3.3)
u(x, 0) = f (x) 0 ≤ x ≤ L, (3.4)
where f is a given initial condition, and K is a positive constant.
• By (3.3) and (3.4), we have
f (0) = u(0, 0) = 0 and f (L) = u(L, 0) = 0.
This is called a compatibility condition.

26
• The condition (3.3) is called a Dirichlet condition.

We start our study of the PDE (3.2) by looking for nonzero solutions of the
form
p(x, t) = X(x) · T (t),
where X and T are single variable functions of x and t, respectively. Such a
solutions are called product solutions.
Then
d d2
pt = (X(x) · T (t)) = XTt and pxx = (X(x) · T (t)) = Xxx T.
dt dx2
Assuming p is a solution to (3.2), we get

XTt = KXxx T.

by performing separation of variables, we get


Tt Xxx
= . (3.5)
KT X
Differentiating both sides with respect to t, we get
   
d Tt d Xxx
= = 0;
dt KT dt X

hence, (3.5) is equal to some constant −λ; that is,


Tt Xxx
= = −λ. (3.6)
KT X
Here, λ is called the separation constant. Equation (3.6) leads to the following
system of ODEs:

d2 X
= −λX 0 < x < L, (3.7)
dx2
dT
= −λKT t > 0. (3.8)
dt
The function p satisfies the boundary conditions (3.3) if and only if

u(0, t) = X(0) · T (t) = 0 and u(L, t) = X(L) · T (t) = 0

for all t ≥ 0. Since we want p to be a nonzero solution, T (t) 6= 0 for all t;


hence, X(0) = X(L) = 0. Therefore, the function X should be a solution of the
boundary value problem

d2 X
+ λX = 0, 0 < x < L, (3.9)
dx2

27
X(0) = X(L) = 0. (3.10)

A nontrivial solution of the above problem is called an eigenfunction of the


problem (3.2)-(3.4) with an eigenvalue λ.
Since problem (3.9)-(3.10) is a boundary value problem for an ODE, we are
not sure if a solution exists for any value of λ. But since the equation is a
second-order homogeneous linear ODE,
• we can apply Theorem 3.1 to get a general solution (which depends on λ)
to (3.9);
• then we determine the values of λ for which these solutions also satisfy the
boundary conditions (3.10).

The characteristic equation of (3.9) is

r2 + λ = 0,

which implies that r = ± −λ. Applying Theorem 3.1, we have the following:

1. If λ < 0, then the characteristic equation has two distinct real roots. Hence,
√ √
−λx
X(x) = αe + βe− −λx
.

2. If λ = 0, then the characteristic equation has one real root, which is 0.


Hence,
X(x) = α + βx.

3. If λ > 0, then the characteristic equation has complex roots. Hence,


√ √
X(x) = α cos( λx) + β sin( λx).

In the above, α and β are arbitrary real numbers. We now deal with each case
separately.
Case 1: Negative eigenvalue (λ < 0) First, we rewrite X(x) using the hy-
perbolic sine and cosine functions:
√ √
X(x) = αe −λx + βe− −λx
h √ √ i h √ √ i
= α cosh( −λx) + sinh( −λx) + β cosh( −λx) − sinh( −λx)
√ √
= (α + β) cosh( −λx) + (α − β) sinh( −λx)
| {z } | {z }
α̃ β̃
√ √
X(x) = α̃ cosh( −λx) + β̃ sinh( −λx) (3.11)

28
If we set x = 0, we have
√ √
0 = α̃ cosh( −λ · 0) + β̃ sinh( −λ · 0)
= α̃ + β̃ · 0
= α̃.

Then X(x) = β̃ sinh( −λx). If we set x = L, we get

0 = β̃ sinh( −λL) ⇒ β̃ = 0.

Hence, X(x) is identically zero and p = XT . This means that the problem
(3.9)-(3.10) does not admit a negative eigenvalue.
Case 2: Zero eigenvalue (λ = 0) In this case, X(x) = α + βx. But such a
function cannot satisfy X(0) = X(L) = 0 unless X is identically zero; therefore,
the problem (3.9)-(3.10) also does not admit a zero eigenvalue.
Case 3: Positive eigenvalue (λ > 0) In this case, we have
√ √
X(x) = α cos( λx) + β sin( λx). (3.12)

Setting x = 0, we get
0 = α + β · 0 ⇒ α = 0.
Hence, √
X(x) = β sin( λx).
Setting x = L, we get √
0 = β sin( λL),

which is satisfied only when λL = nπ, where n is a positive integer. Hence, λ
is an eigenvalue if and only if
 nπ 2
λ= , n = 1, 2, 3, ....
L
The corresponding eigenfunctions are
nπx
X(x) = sin ,
L
which are unique up to multiplication by a constant.
In conclusion, the set of all solutions to problem (3.9)-(3.10) is an infinite se-
quence of eigenfunctions, each associated with a positive eigenvalue. To represent
these eigenfunction-eigenvalue pairs, we use the notation
nπx  nπ 2
Xn (x) = sin , λ= , n = 1, 2, 3, ....
L L

29
Note that each λn is simple; that is, the space consisting of its eigenvectors is
1-dimensional.
We now consider the ODE (3.8); its general solution is

T (t) = Ce−Kλt

and since λ must be equal to one of the λn ’s, we obtain


2
Tn (t) = Cn e−K ( ) t,

L n = 1, 2, 3, .... (3.13)

Hence, the we have obtained the following sequence of product solutions:


nπx −K ( nπ 2
L ) t,
pn (x, t) = Xn (x)Tn (t) = Cn sin e n = 1, 2, 3, ... (3.14)
L
Recall that such solutions satisfy only the PDE (3.2) and the boundary conditions
(3.3). By the superposition principle, any linear combination of the p0n s,
N
X nπx −K ( nπ 2
L ) t,
u(x, t) = Cn sin e (3.15)
n=1
L

is a solution that satisfies (3.2) and (3.3).


We now consider the initial condition (3.4). Assuming that the initial condi-
tion f (x) is piecewise smooth, we can write f (x) using its Fourier sine expansion;
that is,

X mπx
f (x) = Bm sin , 0 < x < L,
m=1
L

2 L
Z
mπx
where Bm = f (x) sin dx.
L 0 L
πx 3πx
Example 3.4. Suppose f (x) = sin + 2 sin , 0 < x < L. Then the Fourier
L L
series of f (x) is itself. Since we want u(x, t) to satisfy u(x, 0) = f (x), we must
have  
πx 3πx −K ( nπ 2
L ) t.
u(x, t) = sin + 2 sin e
L L
This is now a solution to the problem (3.2)-(3.4).

x, 0 ≤ x ≤ π/2,
Example 3.5. Suppose L = π and f (x) = . Then
π − x, π/2 ≤ x ≤ π
the Fourier series of f (x) is

4 X (−1)n+1
sin[(2n − 1)x]. (3.16)
π n=1 (2n − 1)2

It is clear that any function of the form (3.15) cannot satisfy u(x, 0) = f (x).

30
As illustrated by the above example, f (x) sometimes has an infinite number
of terms; hence, the series given in (3.15) cannot satisfy (3.4). This suggests that
the solution to (3.2)-(3.4) should be in the form of the infinite series

X nπx −K ( nπ 2
L ) t.
Bn sin e (3.17)
n=1
L

However, such a series already involves an infinite number of terms. Because of


this, the superposition principle does not apply and we need to verify that this
infinite series still satisfies the PDE (3.2).
The verification involves using the following results, which we present without
the proof.

X nπx −K ( nπ 2
L ) t converges for
Proposition 3.6. Suppose the series Bn sin e
n=1
L
some (t, x) = (t0 , x0 ) in [0, ∞) × [0, L]. Then the following statements hold:

X nπx ∂  −K ( nπ 2 
L ) t
1. If the series Bn sin e converges uniformly on [0, ∞)
n=1
L ∂t
for each x ∈ [0, L], then for all (x, t) ∈ [0, L] × [0, ∞), we have
∞ ∞
!
∂ X nπx −K ( nπ
L )
2
t
X nπx ∂  −K ( nπ 2 
L ) t
Bn sin e = Bn sin e .
∂t n=1 L n=1
L ∂t


X ∂2  nπx  −K ( nπ 2
L ) t converges uniformly on [0, L]
2. If the series Bn 2
sin e
n=1
∂x L
for each t ∈ [0, ∞), then for all (x, t) ∈ [0, L] × [0, ∞), we have
∞ ∞
!
∂2 X nπx −K ( nπ )
2
t
X ∂2  nπx  −K ( nπ 2
L ) t.
2
B n sin e L = B n 2
sin e
∂x n=1
L n=1
∂x L

Remark 3.7. The closed intervals in the above proposition can be replaced by
the corresponding open intervals.
Proposition 3.8 (Weierstrass M-Test). Suppose (fn ) is a sequence of real- or
complex-valued functions defined on some set E, and suppose

|fn (x)| ≤ Mn x ∈ E, n = 1, 2, 3, ....


P P
Then fn converges uniformly on E if Mn converges.
Example 3.9. Find a solution to the problem

ut − uxx = 0, 0 < x < π, t > 0, (3.18)

31
u(0, t) = u(π, t) = 0, t ≥ 0, (3.19)

x, 0 ≤ x ≤ π/2,
u(x, 0) = f (x) = (3.20)
π − x, π/2 ≤ x ≤ π.
By (3.17), a candidate solution is of the form

X 2
Bn (sin nx)e−n t .
n=1

where the Bn ’s are the Fourier coefficients of f (x). Hence, based on (3.16), a
candidate solution is

4 X (−1)n+1 2
u(x, t) = 2
sin[(2n − 1)x]e−(2n−1) t .
π n=1 (2n − 1)

Clearly, u(0, t) = u(π, t) = 0 for all t ≥ 0; that is, u(x, t) satisfies the boundary
conditions (3.19). Moreover,

4 X (−1)n+1
u(x, 0) = sin[(2n − 1)x] = f (x);
π n=1 (2n − 1)2

so u(x, t) satisfies the initial condition (3.20).


It remains to show that u satisfies the PDE (3.18). Using Proposition 3.6,
∂u ∂2u
we will show that we can compute and by term-by-term differentiation.
∂t ∂x2
• We have
4 (−1)n+1

−(2n−1)2 t
4 4

π (2n − 1)2 sin[(2n − 1)x]e ≤ π(2n − 1)2 ≤ πn2

4P 1
for all (x, t) ∈ [0, L] × [0, ∞). Since converges, it follows by the
π n2
Weierstrass M-test that u(x, t) converges uniformly on [0, π] × [0, ∞).
• Then u(x, t) converges for all points on (0, π) × (0, ∞). (Since the terms
of the series are continuous, the uniform limit u(x, t) of the series is also
continuous.)
• The term-by-term derivative of u with respect to t is

4X 2
v(x, t) = (−1)n sin[(2n − 1)x]e−(2n−1) t .
π n=1

We have to show that this series converges on the t-interval (0, +∞) for
each fixed x ∈ (0, π). Suppose t ≥ ε > 0 and fix x ∈ (0, π). Then we have

4
(−1)n sin[(2n − 1)x]e−(2n−1)2 t ≤ 4 e−(2n−1)2 ε

π π

32
4 1

π (2n − 1)2 ε
4 1
≤ .
πε n2
4 P 1
Since is convergent, v(x, t) is uniformly convergent on [ε, ∞) for
πε n2
each fixed x ∈ (0, π). Since ε is an arbitrary positive number, we must have
v(x, t) uniformly convergent on (0, ∞) for each fixed x ∈ (0, π). Therefore,
v(x, t) = ut (x, t) for all (x, t) ∈ (0, π) × (0, ∞).
• The second-order term-by-term derivative of u with respect to x is

4X 2
w(x, t) = (−1)n sin[(2n − 1)x]e−(2n−1) t .
π n=1

This is, in fact, equal to v(x, t). Now, fix t ∈ (0, ∞). Then

4
(−1)n sin[(2n − 1)x]e−(2n−1)2 t ≤ 4 1

π πt n2

4 P 1
for all x ∈ (0, π). Since is convergent, w(x, t) is also uniformly
πt n2
convergent on the x interval (0, π) for each fixed t ∈ (0, ∞). Therefore,
w(x, t) = uxx (x, t) for all (x, t) ∈ (0, π) × (0, ∞).
The above computations give ut − uxx = 0 for all 0 < x < π, t > 0; that is,
u(x, t) satisfies the PDE (3.18). Therefore, u(x, t) is a solution to the problem
(3.18)-(3.20).
Example 3.10. Find a solution to the problem

ut = 17uxx , 0 < x < π, t > 0, (3.21)


u(0, t) = u(π, t) = 0, t ≥ 0, (3.22)

 0, 0 < x < π/2,
u(x, 0) = f (x) = 1, x = π/2, (3.23)
2, π/2 < x < π.

By (3.17), a candidate solution is of the form



X 2
Bn (sin nx)e−17n t .
n=1

where the Bn ’s are the Fourier coefficients of f (x). Since the Fourier series of
f (x) is

X 4 h  nπ  i
cos − (−1)n sin nx,
n=1
πn 2

33
a candidate solution is

X 4 h  nπ  i 2
cos − (−1)n (sin nx)e−17n t .
n=1
πn 2

Clearly, u(0, t) = u(π, t) = 0 for all t ≥ 0; that is, u(x, t) satisfies the bound-
ary conditions (3.22). Moreover, u(x, 0) = f (x); so u(x, t) satisfies the initial
condition (3.23).
It remains to show that u satisfies the PDE (3.21). Using Proposition 3.6,
∂u ∂2u
we will show that we can compute and by term-by-term differentiation.
∂t ∂x2
• Fix t ∈ (0, +∞). Then

4 h  nπ 
n
i
−17n2 t
4 −17n2 t
πn cos 2 − (−1) (sin nx)e ≤ πn · 2 · e

8 1
≤ ·
πn 17n2 t
8 1
= · .
17πt n3
8 P 1
for each x ∈ (0, π). Since converges, it follows by the Weier-
17πt n3
strass M-test that u(x, t) converges uniformly on [0, π] for each fixed t ∈
(0, ∞).
Alternative Solution: Use the ratio test to conclude that the series
1 −17n2 t
P
ne is convergent.
• Then u(x, t) converges for all points on (0, π) × (0, ∞). (Since the terms
of the series are continuous, the uniform limit u(x, t) of the series is also
continuous.)

• The term-by-term derivative of u with respect to t is



X −68n h  nπ  i 2
v(x, t) = cos − (−1)n (sin nx)e−17n t .
n=1
π 2

We have to show that this series converges on the t-interval (0, +∞) for
each fixed x ∈ (0, π). Suppose t ≥ ε > 0 and fix x ∈ (0, π). Then we have

−68n h  nπ  i
−17n2 t 68 2
n
· 2 · e−17n ε


π cos − (−1) (sin nx)e ≤
2 πn
136 −17n2 ε
≤ ne .
π

34
2
Consider the series ne−17n ε . We test the convergence of this series using
the ratio test. We have

(n + 1)e−17(n+1)2 ε n + 1 −17(2n+1)ε
lim = lim e

−17n 2ε
n→∞ ne n→∞ n
 
n+1 h i
= lim lim e−17(2n+1)ε
n→∞ n n→∞

=1×0
= 0.
X 2
Therefore, the series ne−17n ε is convergent; so v(x, t) is uniformly con-
vergent on [ε, ∞) for each fixed x ∈ (0, π). Since ε is an arbitrary positive
number, we must have v(x, t) uniformly convergent on (0, ∞) for each fixed
x ∈ (0, π). Therefore, v(x, t) = ut (x, t) for all (x, t) ∈ (0, π) × (0, ∞).
• The second-order term-by-term derivative of u with respect to x is

X −4n h  nπ  i 2
w(x, t) = cos − (−1)n (sin nx)e−17n t .
n=1
π 2

Fix t ∈ (0, ∞). Then



−4n h  nπ 
n
i
−17n2 t
8 −17n2 t
π cos 2 − (−1) (sin nx)e ≤ π ne .

X 2
for all x ∈ (0, π). Since the series ne−17n t is convergent, w(x, t) is
also uniformly convergent on the x interval (0, π) for each fixed t ∈ (0, ∞).
Therefore, w(x, t) = uxx (x, t) for all (x, t) ∈ (0, π) × (0, ∞).

The above computations give ut = 17uxx for all 0 < x < π, t > 0; that is,
u(x, t) satisfies the PDE (3.21). Therefore, u(x, t) is a solution to the problem
(3.21)-(3.23).
Example 3.11. Find a solution to the problem

ut = 2uxx , 0 < x < π, t > 0, (3.24)


u(0, t) = u(π, t) = 0, t ≥ 0, (3.25)
u(x, 0) = f (x) = x 0 < x < π. (3.26)

By (3.17), a candidate solution is of the form



X 2
Bn (sin nx)e−2n t .
n=1

35
where the Bn ’s are the Fourier coefficients of f (x). Since the Fourier series of
f (x) is
∞ ∞
X X (−1)n+1
2 sin nx,
n=1 n=1
n
a candidate solution is

X (−1)n+1 2
2 (sin nx)e−2n t .
n=1
n

Clearly, u(0, t) = u(π, t) = 0 for all t ≥ 0; that is, u(x, t) satisfies the bound-
ary conditions (3.25). Moreover, u(x, 0) = f (x); so u(x, t) satisfies the initial
condition (3.26).
It remains to show that u satisfies the PDE (3.24). Using Proposition 3.6,
∂u ∂2u
we will show that we can compute and by term-by-term differentiation.
∂t ∂x2
• Fix t ∈ (0, +∞). Then
(−1)n+1

−2n2 t
2 −2n2 t
≤ ne
2 (sin nx)e
n
P 1 −2n2 t
for each x ∈ (0, π). Since 2 e converges by the ratio test, it follows
n
by the Weierstrass M-test that u(x, t) converges uniformly on (0, π) for each
fixed t ∈ (0, ∞).
• Then u(x, t) converges for all points on (0, π) × (0, ∞). (Since the terms
of the series are continuous, the uniform limit u(x, t) of the series is also
continuous.)
• The term-by-term derivative of u with respect to t is

X 2
v(x, t) = 4n(−1)n (sin nx)e−2n t .
n=1

We have to show that this series converges on the t-interval (0, +∞) for
each fixed x ∈ (0, π). Suppose t ≥ ε > 0 and fix x ∈ (0, π). Then we have
2
2
4n(−1)n (sin nx)e−2n t ≤ 4ne−2n ε

X 2
The series 4ne−2n ε is convergent by the ratio test; so v(x, t) is uni-
formly convergent on [ε, ∞) for each fixed x ∈ (0, π). Since ε is an ar-
bitrary positive number, we must have v(x, t) uniformly convergent on
(0, ∞) for each fixed x ∈ (0, π). Therefore, v(x, t) = ut (x, t) for all
(x, t) ∈ (0, π) × (0, ∞).

36
• The second-order term-by-term derivative of u with respect to x is

X 2
w(x, t) = 2n(−1)n (sin nx)e−2n t .
n=1

Fix t ∈ (0, ∞). Then


2
2
2n(−1)n (sin nx)e−2n t ≤ 2ne−2n t .

X 2
for all x ∈ (0, π). Since the series 2ne−2n t is convergent, w(x, t) is
also uniformly convergent on the x interval (0, π) for each fixed t ∈ (0, ∞).
Therefore, w(x, t) = uxx (x, t) for all (x, t) ∈ (0, π) × (0, ∞).
The above computations give ut = 2uxx for all 0 < x < π, t > 0; that is,
u(x, t) satisfies the PDE (3.24). Therefore, u(x, t) is a solution to the problem
(3.24)-(3.26).
Other Homogeneous Boundary Conditions
In the previous discussion, we considered homogeneous boundary conditions
of the form
u(0, t) = u(L, t) = 0.
Other homogeneous boundary conditions are of the form

ux (0, t) = 0 or ux (0, t) = hu(0, t)

and
ux (L, t) = 0 or ux (L, t) = −hu(L, t).
All these conditions may be included by using the following conditions:

(cos α)u(0, t) − L(sin α)ux (0, t) = 0

and
(cos β)u(L, t) + L(sin β)ux (L, t) = 0
where 0 ≤ α, β < π. The original conditions at x = 0 can be obtained by setting
α = 0, α = π/2, and cot α = hL, respectively. A similar approach can be done
to obtain the conditions at x = L.
Proposition 3.12. The product solutions of the boundary value problem

ut − Kuxx = 0 0 < x < L, t > 0,


(cos α)u(0, t) − L(sin α)ux (0, t) = 0 t ≥ 0,
(cos β)u(L, t) + L(sin β)ux (L, t) = 0 t≥0

37
are of the form un (x, t) = e−λn Kt φn (x) where λn is an eigenvalue and φn (x) is
an eigenfunction of the Sturm-Liouville eigenvalue problem

φ00 (x) + λφ(x) = 0

with boundary conditions

(cos α)φ(0) − L(sin α)φ0 (0) = 0,


(cos β)φ(L) + L(sin β)φ0 (L) = 0.
RL
These eigenfunctions satisfy the orthogonality relation 0
φn φm = 0 for m 6= n.
Example 3.13. Find all product solutions of the problem

ut − Kuxx = 0 0 < x < L, t > 0,


u(0, t) = ux (L, t) = 0.

Solution. The associated Sturm-Liouville problem is φ00 (x) + λφ(x) = 0 with the
boundary conditions φ(0) = 0, φ0 (L) = 0. Using similar arguments as before, we
can show that when λ ≤ 0, it cannot be an eigenvalue.
For λ > 0, the general solution of the Sturm-Liouville problem is
√ √
φ(x) = A sin λx + B cos λx.

The first boundary condition √requires that B = 0, while the second boundary
condition
√ requires that A cos λL = 0. For a nonzero solution, we must take
L λ = n − 12 π, n = 1, 2, .... Therefore, the solutions are


 
1 πx
φn (x) = sin n −
2 L

1 2

with λn = n − 2 π 2 /L2 . The product solutions are
   "  2 #
1 πx 1 2 Kt
un (x, t) = sin n − exp − n − π 2 , n = 1, 2, ....
2 L 2 L

Summary of Method
Now, consider the initial-value problem

ut − Kuxx = 0 0 < x < L, t > 0,


(cos α)u(0, t) − L(sin α)ux (0, t) = 0 t ≥ 0,
(cos β)u(L, t) + L(sin β)ux (L, t) = 0 t ≥ 0,

38
u(x, 0) = f (x) 0 < x < L,

where f (x), 0 < x < L, is a given piecewise smooth function. To solve this
problem, we first expand f (x) in a series of eigenfunctions of the Sturm-Liouville
problem, in the form

X
f (x) = An φn (x), 0 < x < L.
n=1

The formal solution of the initial-value problem is given by



X
u(x, t) = An φn (x)e−λn Kt .
n=1

To prove that this formal solution is a rigorous solution to the heat equation, we
must check, for each t > 0, that the series for u, ux , uxx , and ut are uniformly
convergent for 0 ≤ x ≤ L.

39

Você também pode gostar