FOURIER SERIES Math 21b, Spring 08

Smooth functions f (x) on [−π, π] form a linear space X. There is an inner product in X defined by

1
Rπ
hf, gi = π −π
f (x)g(x) dx

It allows to define angles, length, projections in X as we did in finite dimensions.

THE FOURIER BASIS.

√
THEOREM. The functions {cos(nx), sin(nx), 1/ 2 } form an orthonormal set in X.

Proof. To check linear independence a few integrals need to be computed. For all n, m ≥ 1, with n 6= m you
have to show:
h √12 , √12 i = 1
hcos(nx), cos(nx)i = 1, hcos(nx), cos(mx)i = 0
hsin(nx), sin(nx)i = 1, hsin(nx), sin(mx)i = 0
hsin(nx), cos(mx)i
√ =0
hsin(nx), 1/ √2i = 0
hcos(nx), 1/ 2i = 0

To verify the above integrals in the homework, the following trigonometric identities are useful:

2 cos(nx) cos(my) = cos(nx − my) + cos(nx + my)

2 sin(nx) sin(my) = cos(nx − my) − cos(nx + my)
2 sin(nx) cos(my) = sin(nx + my) + sin(nx − my)

FOURIER COEFFICIENTS. The Fourier coefficients of a function f in X are defined as

√ 1
Rπ √
a0 = hf, 1/ 2i = π −π f (x)/ 2 dx
1
Rπ
an = hf, cos(nt)i = π −π f (x) cos(nx) dx
1
Rπ
bn = hf, sin(nt)i = π −π f (x) sin(nx) dx

FOURIER SERIES. The Fourier representation of a smooth function f is the identity

a0 P∞ P∞
f (x) = √ 2
+ k=1 ak cos(kx) + k=1 bk sin(kx)

We take it for granted that the series converges and that the identity holds for all x where f is continuous.

ODD AND EVEN FUNCTIONS. Here is some advise which can save time when computing Fourier series:

If f is odd: f (x) = −f (−x), then f has a sin series.

If f is even: f (x) = f (−x), then f has a cos series.
If you integrate an odd function over [−π, π] you get 0.
The product of two odd functions is even, the product between an even and an odd function is odd.

EXAMPLE 1. Let f (x) = x on [−π, π]. This is an odd function (f (−x)+f (x) = 0) so that it has a sin series: with
Rπ n+1
bn = π1 −π x sin(nx) dx = −1 /n, we get x = n=1 2 (−1)n
2 π n+1
P∞
π (x cos(nx)/n + sin(nx)/n |−π ) = 2(−1) sin(nx).
For example
π
2 = 2( 11 − 1
3 + 1
5 − 71 ...)

is a formula of Leibnitz.
EXAMPLE 2. Let f (x) = cos(x) + 1/7 cos(5x). This trigonometric polynomial is already the Fourier series.
There is no need to compute the integrals. The nonzero coefficients are a1 = 1, a5 = 1/7.

EXAMPLE 3. Let f (x) = 1 on [−π/2, π/2] and f (x) = 0 else. This is an even function f (−x) − f (x) = 0 so
√ R π/2 π/2
that it has a cos series: with a0 = 1/( 2), an = π1 −π/2 1 cos(nx) dx = sin(nx) 2(−1)m
πn |−π/2 = π(2m+1) if n = 2m + 1 is
odd and 0 else. So, the series is

1 2 cos(x) cos(3x) cos(5x)

f (x) = + ( − + − ...)
2 π 1 3 5
Remark. The function in Example 3 is not smooth, but Fourier theory still works. What happens at the
discontinuity point π/2? The Fourier series converges to 0. Diplomatically it has chosen the point in the middle
of the limits from the right and the limit from the left.

1.2

FOURIER APPROXIMATION. For a smooth function f , the 1

Fourier series of f converges to f . The Fourier coefficients are 0.8

the coordinates of f in the Fourier basis.

0.6

a0 Pn Pn
The function fn (x) = √
2
+ k=1 ak cos(kx) + k=1 bk sin(kx) is 0.4

called a Fourier approximation of f . The picture to the right

0.2
plots a few approximations in the case of a piecewise continuous
even function given in example 3). -3 -2 -1 1 2 3

-0.2

THE PARSEVAL EQUALITY. When evaluating the square of the length of f with the square of the length of
the series, we get
P∞
||f ||2 = a20 + k=1 a2k + b2k .

EXAMPLE. We have seen in example 1 that f (x) = x = 2(sin(x) − sin(2x)/2 + sin(3x)/3 Rπ − sin(4x)/4 + ...
Because the Fourier coefficients are bk = 2(−1)k+1 /k, we have 4(1 + 1/4 + 1/9 + ...) = π1 −π x2 dx = 2π 2 /3 and
so
1 1 1 1 1 π2
1 + 4 + 9 + 16 + 25 + ... = 6

Isn’t it fantastic that we can sum up the reciprocal squares? This formula has been obtained already by
Leonard Euler.

HOMEWORK: (this homework is due either Wednesday or Thursday. Remember that there is a Mathematica
lab due on Thursday or Friday, the last day of class.)
√
1. Verify that the functions cos(nx), sin(nx), 1/ 2 form an orthonormal family.
2. Find the Fourier series of the function f (x) = |x|.

3. Find the Fourier series of the function cos2 (x) + 5 sin(7x) + 55. You may find the double angle formula
cos2 (x) = cos(2x)+1
2 useful.
4. Find the Fourier series of the function f (x) = | sin(x)|.
5. In the previous problem 4) you should have gotten a series
 
2 4 cos(2x) cos(4x) cos(6x)
f (x) = − + 2 + 2 + ...
π π 22 − 1 4 −1 6 −1

Use Parseval’s identity to find the value of

1 1 1
+ 2 + 2 + ···
(22 − 1)2 (4 − 1)2 (6 − 1)2
HEAT AND WAVE EQUATION Math 21b, Spring 08

FUNCTIONS OF TWO VARIABLES. We consider functions f (x, t) which are for fixed t a piecewise smooth
function in x. Analogously as we studied the motion of a vector ~v (t), we are now interested in the motion of a
function f in time t. While the governing equation for a vector was an ordinary differential equation ẋ = Ax
(ODE), the describing equation is now be a partial differential equation (PDE) f˙ = T (f ). The function
f (x, t) could denote the temperature of a stick at a position x at time t or the displacement of a string
at the position x√at time t. The motion of these dynamical systems will be easy to describe in the orthonormal
Fourier basis 1/ 2, sin(nx), cos(nx) treated in an earlier lecture.

PARTIAL DERIVATIVES. We write fx (x, t) and ft (x, t) for the partial derivatives with respect to x or t.
The notation fxx (x, t) means that we differentiate twice with respect to x.

Example: for f (x, t) = cos(x + 4t2 ), we have

• fx (x, t) = − sin(x + 4t2 )
• ft (x, t) = −8t sin(x + 4t2 ).
• fxx (x, t) = − cos(x + 4t2 ).

One also uses the notation ∂f∂x(x,y)

for the partial derivative with respect to x. Tired of all the ”partial derivative
signs”, we always write fx (x, t) for the partial derivative with respect to x and ft (x, t) for the partial derivative
with respect to t.

PARTIAL DIFFERENTIAL EQUATIONS. A partial differential equation is an equation for an unknown

function f (x, t) in which different partial derivatives occur.

• ft (x, t) + fx (x, t) = 0 with f (x, 0) = sin(x) has a

solution f (x, t) = sin(x − t).
• ftt (x, t) − fxx (x, t) = 0 with f (x, 0) = sin(x) and
ft (x, 0) = 0 has a solution f (x, t) = (sin(x − t) +
sin(x + t))/2.

THE HEAT EQUATION. The temperature distribution f (x, t) in a metal bar [0, π] satisfies the heat equation

ft (x, t) = µfxx (x, t)

This partial differential equation tells that the rate of change of the temperature at x is proportional to the
second space derivative of f (x, t) at x. The function f (x, t) is assumed to be zero at both ends of the bar and
f (x) = f (x, t) is a given initial temperature distribution. The constant µ depends on the heat conductivity
properties of the material. Metals for example conduct heat well and would lead to a large µ.

REWRITING THE PROBLEM. We can write the problem as

d
dt f = µD2 f

We will solve the problem in the same way as we solved linear differential equations:
d
dt ~
x = A~x

where A is a matrix - by diagonalization .

We use that the Fourier basis is just the diagonalization: D2 cos(nx) = −n2 cos(nx) and D2 sin(nx) =
−n2 sin(nx) show that cos(nx) and sin(nx) are eigenfunctions to D2 with eigenvalue n2 . By a symmetry
trick, we can focus on sin-series from now on.
SOLVING THE HEAT EQUATION WITH FOURIER THEORY. The heat equation ft (x, t) = µfxx (x, t) with
smooth f (x, 0) = f (x), f (0, t) = f (π, t) = 0 has the solution

2
Rπ
−n2 µt
P∞
f (x, t) = n=1 bn sin(nx)e
bn = π 0
f (x) sin(nx) dx

2
Proof: With the initial condition f (x) = sin(nx), we have the evolution f (x, t) = e−µn t sin(nx). If f (x) =
−µn2 t
P∞ P∞
n=1 bn sin(nx) then f (x, t) = n=1 bn e sin(nx).

A SYMMETRY TRICK. Given a function f on the interval [0, π] which is zero at 0 and π. It can be extended
to an odd function on the doubled integral [−π, π].

The Fourier series of an odd function

Rπ is a pure sin-series. The
Fourier coefficients are bn = π2 0 f (x) sin(nx) dx.

The odd
The function
symmetric
is given on
extension on
[0, π].
[−π, π].

EXAMPLE. Assume the initial temperature distribution f (x, 0) is

a sawtooth function which has slope 1 on the interval [0, π/2] and
slope −1 on the interval [π/2, π]. We first compute the sin-Fourier
coefficients of this function.
4 (n−1)/2
The sin-Fourier coefficients are bn = n2 π (−1) for odd n and 0 for even n. The solution is
∞
X 2
f (x, t) = bn e−µn t sin(nx) .
n

The exponential term containing the time makes the function f (x, t) converge to 0: The body cools. The higher
frequencies are damped faster: ”smaller disturbances are smoothed out faster.”

VISUALIZATION. We can plot the graph of the function f (x, t) or slice this graph and plot the temperature
distribution for different values of the time t.

f (x, 0) f (x, 1) f (x, 2) f (x, 3) f (x, 4)

THE WAVE EQUATION. The height of a string f (x, t) at time t and position x on [0, π] satisfies the wave
equation

ftt (t, x) = c2 fxx (t, x)

where c is a constant. As we will see, c is the speed of the waves.

REWRITING THE PROBLEM. We can write the problem as

d2
dt2 f = c2 D 2 f

We will solve the problem in the same way as we solved

d2
dt2 ~
x = A~x

d2
If A is diagonal, then every basis vector x satisfies an equation of the form dt2 x = −c2 x which has the solution
x(t) = x(0) cos(ct) + x′ (0) sin(ct)/c.

SOLVING THE WAVE EQUATION WITH FOURIER THEORY. The wave equation ftt = c2 fxx with f (x, 0) =
f (x), ft (x, 0) = g(x), f (0, t) = f (π, t) = 0 has the solution

f (x, t) = 2
Rπ
P∞ an = π f (x) sin(nx) dx
a
n=1 n sin(nx) cos(nct) + 2
R0π
bn
sin(nx) sin(nct) bn = π 0 g(x) sin(nx) dx
nc

Proof: With f (x) = sin(nx), g(x) = 0, the solution is fP (x, t) = cos(nct) sin(nx). With
P∞f (x) = 0, g(x) = sin(nx),
the solution is f (x, t) = 1c sin(ct) sin(nx). For f (x) = n=1 an sin(nx) and g(x) = n=1 bn sin(nx), we get the
∞

formula by summing these two solutions.

VISUALIZATION. We can just plot the graph of the function f (x, t) or plot the string for different times t.

f (x, 0) f (x, 1) f (x, 2) f (x, 3) f (x, 4)

TO THE DERIVATION OF THE HEAT EQUATION. The tem-

perature f (x, t) is proportional to the kinetic energy at x. Di-
vide the stick into n adjacent cells and assume that in each time
step, a fraction of the particles moves randomly either to the
right or to the left. If fi (t) is the energy of particles in cell i
at time t, then the energy fi (t + dt) of particles at time t + dt
is proportional to fi (t) + (fi−1 (t) − 2fi (t) + fi+1 )(t)). This is a
discrete version of the second derivative because dx2 fxx (t, x) ∼
(f (x + dx, t) − 2f (x, t) + f (x − dx, t)).

TO THE DERIVATION OF THE WAVE EQUATION. We can

model a string by n discrete particles linked by springs. Assume
that the particles can move up and down only. If fi (t) is the
height of the particles, then the right particle pulls with a force
fi+1 − fi , the left particle with a force fi−1 − fi . Again, (fi−1 (t) −
2fi (t)+fi+1 )(t)) which is a discrete version of the second derivative
because fxx (t, x)dx2 ∼ (f (x + dx, t) − 2f (x, t) + f (x − dx, t)).
OVERVIEW: The heat and wave equation can be solved like ordinary differential equations:

Ordinary differential equations Partial differential equations

xt (t) = Ax(t) ft (t, x) = fxx (t, x)

xtt (t) = Ax(t) ftt (t, x) = fxx (t, x)

Diagonalizing A leads for eigenvectors ~v Diagonalizing T = D2 with eigenfunctions f (x) =

sin(nx)
Av = −c2 v
T f = −n2 f

to the differential equations

leads to the differential equations
vt = −c2 v
vtt = −c2 v ft (x, t) = −n2 f (x, t)
ftt (x, t) = −n2 f (x, t)

which are solved by

2
which are solved by
v(t) = e−c t v(0)
2
t
v(t) = v(0) cos(ct) + vt (0) sin(ct)/c f (x, t) = f (x, 0)e−n
f (x, t) = f (x, 0) cos(nt) + ft (x, 0) sin(nt)/n

NOTATION:

f function on [−π, π] smooth or piecewise smooth. T f = λf Eigenvalue equation analog to Av = λv.

t time variable ft partial derivative of f (x, t) with respect to time t.
x space variable fx partial derivative of f (x, t) with respect to space x.
D the partial differential operator Df (x) = f ′ (x) = fxx second partial derivative of f twice with respect
d/dxf (x). to space x.
T linear transformation, like T f = D2 f = f ′′ . µ heat conductivity
c speed of the wave. f (x) = −f (−x) odd function, has sin Fourier series

HOMEWORK. This homework is due until Tuesday morning May 6, 2008 in the mailboxes of your CA:

6) Solve the heat equation ft = µfxx on [0, π] with the initial condition f (x, 0) = | sin(3x)|.

We want to see in exercises 2-4 how to deal with solutions to the heat equation, where the boundary values are
not 0. Problems 2-4 belong together:

7) Verify that for any constants a, b the function h(x, t) = (b − a)x/π + a is a solution to the heat equation.

8) Assume we have the problem to describe solutions f (x, t) to the heat equations, where f (0, t) = a and
f (π, t) = b. Show that g(x, t) = f (x, t)−h(x, t) is a solution of the heat equation with g(0, t) = 0 and g(π, t) = 0.

9) Solve the heat equation with the initial condition f (x, 0) = f (x) = sin(5x) + 3 sin(x) + x/π and satisfying
f (0, t) = 0, f (π, t) = 1 for all times t. This is a situation, when the stick is kept at constant but different
temperatures on the both ends.

10) A piano string is fixed at the ends x = 0 and x = π and is initially undisturbed. The piano hammer induces
an initial velocity ft (x, 0) = g(x) onto the string, where g(x) = sin(2x) on the interval [0, π/2] and g(x) = 0 on
[π/2, π]. Find the motion of the string.