Fourier Series Solution of the Heat

Equation
Physical Application; the Heat Equation In the early nineteenth century Joseph Fourier, a French scientist and mathematician
who had accompanied Napoleon on his Egyptian campaign, introduced
the idea of expanding functions in trigonometric series as a device to
solve the heat conduction equation he had developed in his treatise Theorie Analytique de la Chaleur (Analytic Theory of Heat). The subject
was greatly extended beyond Fouriers original contributions by later
mathematicians but has continued to bear Fouriers name. Our treatment of Fourier series would clearly be incomplete without reference to
the subject of heat conduction which led to Fouriers discovery.
We consider a long thin bar of heat conducting material; the length
coordinate may be taken to be x in the interval [0, L]. We will suppose
that the specific heat per unit length (a scalar indicating the capacity of
a unit length of the material to hold heat) is and the heat conductivity
is . We let N be a positive integer, h = L/N , and xk = kh, k =
0, 1, 2, , N . For the present discussion we think of the bar formed into
a ring. Thus x 0 = 0 is identified with xn = L and we use the wraparound convention with respect to indices k lying outside the range 0
through N .
We suppose the temperature in the subinterval Ik = [xk1 , xk ] at
a given time t can be adequately approximated by the scalar function
Tk (t). The heat contained in Ik is then h Tk (t). The heat conductivity coefficient expresses the relationship between the rate of flow of
heat and the temperature differential per unit length, i.e., T
x . Since
T
our model is spatially discrete so far, we approximate x (xk ) by the
k1
k
difference quotient Tk T
. The rate of heat flow into Ik is then h dT
h
dt
while the flow of heat into Ik from Ik+1 and Ik1 is h (Tk+1 (t) Tk (t))
1

and h (Tk1 (t) Tk (t)), respectively. Assuming heat is conserved we

obtain
dTk

h
= (Tk+1 (t) 2 Tk (t) + Tk1 (t)) .
dt
h
Dividing by h we have

dTk
Tk+1 (t) 2 Tk (t) + Tk1 (t)
=
.
dt
h2

If we assume the actual heat distribution is a function T (x, t) of both

time, t, and space, x, the fraction on the right, a second difference
2
divided by h2 , may be regarded as an approximation to xT2 (xk ). Letting
N , h = L/N tends to zero and, in the limit, we obtain for T (x, t)
the partial differential equation
T
2T

(x, t) = 2 (x, t), x [0, L], t > 0.

t
x
If there are external heat sources, or losses which we can represent by
a function f (x, t) the equation is augmented to the more general form
T
2T

(x, t) = 2 (x, t) + f (x, t).

t
x
In addition to the partial differential equation it is necessary to give an
initial heat distribution. We may suppose the initial instant corresponds
to t 0 = 0 and then stipulate the initial data
T (x, 0) = T 0 (x)
with the initial heat distribution function T 0 (x) at least piecewise continuous as a function of x.
Sophisticated mathematical techniques are required to show that
partial differential equations such as these actually have solutions, that
the solutions are uniquely determined by the given initial data, etc.
These questions are important but we do not pursue them here; our
2

task is to describe a method for solution of equations of this type with

the use of Fourier series techniques.
The Method of Separation of Variables Let us divide the partial
differential equation shown earlier by the positive number , define
/ and rename f (x, t) as f (x, t) again. Then we have
T
2T
(x, t) = 2 (x, t) + f (x, t).
t
x
We begin with the homogeneous case f (x, t) 0. To implement the
method of separation of variables we write T (x, t) = z(t) y(x), thus
expressing T (x, t) as the product of a function of t and a function of
dy
d2 y
0
00
and
y
,
y
to
denote
,
, respectively, we
x. Using z to denote dz
dt
dx
dx2
obtain
z(t)
y(x) = z(t) y 00(x).
Assuming z(t), y(x) are non-zero, we then have
z(t)

y 00 (x)
=
.
z(t)
y(x)
Since the left hand side is a constant with respect to x and the right
hand side is a constant with respect to t, both sides must, in fact, be
constant. It turns out that constant should be taken to be non-positive,
so we indicate it as 2 ; thus
z(t)

y 00(x)
=
= 2
z(t)
y(x)
and we then have two ordinary differential equations
z(t)

= 2 z(t), y 00 (x) = 2 y(x).

We first deal with the second equation, writing it as
y 00 (x) + 2 y(x) = 0.
3

The general solution of this equation takes the form

y(x) = c cos x + d sin x.
Since we want y(x) to be periodic with period L the choices for are
=

2k
, k = 0, 1, 2, .
L

The choice k = 0 is only useful for the cosine; cos 0 = 1. Indexing the
coefficients c, d to correspond to the indicated choices of , we have
solutions for the y equation in the forms
c 0 (a constant);
ck cos

2kx
2kx
+ dk sin
, k = 1, 2, .
L
L

the z equation takes the

Now, for each indicated choice = 2k
L
form
4 2 k2
z(t)

=
z(t)
L2
which has the general solution
4 2 k2

z(t) = c exp
t .
L2

Absorbing the constant c appearing here into the earlier ck , dk we have

solutions of the homogeneous partial differential equation in the form
T (x, t) = c 0 ,
4 2 k2
2kx
2kx

T (x, t) = exp
t
c
cos
+
d
sin
, k = 1, 2, .
k
k
L2
L
L

Since we are working at this point with a linear homogeneous equation, any linear combination of these solutions will also be a solution.
4

This means we can represent a whole family of solutions, involving an

infinite number of parameters, in the form
T (x, t) = c 0 +

k=1

4 2 k2
2kx
2kx

exp
t
c
cos
sin
.
+
d
k
k
L2
L
L

It should be noted that this expression is a representation of T (x, t) in

the form of a Fourier series with coefficients depending on the time, t:
T (x, t) = c 0 +

k=1

2kx
2kx
ck (t) cos
+ dk (t) sin
,
L
L
!

where
4 2 k2
4 2 k2

ck (t) = ck exp
t , dk (t) = dk exp
t .
L2
L2

The coefficients ck (t), dk (t), k = 1, 2, 3, in the above representation of T (x, t) remain undetermined, of course, to precisely the
extent that the constants ck , dk remain undetermined. In order to
obtain definite values for these coefficients it is necessary to use the initial temperature distribution T 0 (x). This function has a Fourier series
representation
T 0 (x) = a 0 +

k=1

2kx
2kx
ak cos
+ bk cos
,
L
L
!

where
1
a0 =
L

2 ZL
2kx
T 0 (x) dx, ak =
cos
T 0 (x) dx,
0
L 0
L
2 ZL
2kx
bk =
sin
T 0 (x) dx.
L 0
L
To obtain agreement at t = 0 between our Fourier series representation
of T (x, 0)and this Fourier
series representation of T 0 (x) we require,

4 2 k 2
since exp L2 0 = 1,
Z L

c 0 = a 0, ck = ak , dk = bk , k = 1, 2, 3, .
5

Thus we have, in fact

T (x, t) = a 0 +

k=1

4 2 k2
2kx
2kx

+ bk sin
exp
t ak cos
,
2
L
L
L

where a 0 , ak , bk , k = 1, 2, 3, are the Fourier coefficients of the

initial temperature distribution T 0 (x).
Example 1 In the section on real Fourier series we have computed
the coefficients of the Fourier series for the function f (x) = |x |
on the interval [0, 2] . Taking this function as the initial temperature
distribution T 0 (x) for the spatially periodic heat conduction process
discussed above, we have

4
T 0 (x) =
cos kx.

2
2
k
k=1
(k odd)

Accordingly, the solution of the periodic heat equation with this initial
temperature distribution is

4 k2 t
T (x, t) =
e
cos kx.

2
2
k
k=1
(k odd)

Since the non-negative odd integers can be represented as k = 2j +

1, j = 0, 1, 2, ..., we can write this without the qualifier k odd in the
form
T (x, t) =

j=0

4
(2j+1)2 t
e
cos(2j + 1)x.
(2j + 1)2