MATH 4330/5330, Fourier Analysis

Section 1, The Heat Equation on the Line

We imagine we have an infinitely long rod and that we have fixed a point on the
rod that we call 0. We suppose that a function u of two variables t and x is such
that the value u(t, x) is the temperature at the point x on the rod at time t. We
would like to be able to predict how this temperature function changes with time.
That is, we consider the following so-called “initial value problem.”
We suppose that we know the values u(0, x) ≡ f (x) for all points x. These are the
initial values (temperatures). Is that enough information for us to be able to figure
out the values u(t, x) for a later time t? That is, is the evolution of the temperature
function uniquely determined by what it is at a starting time? Can we find out
what the temperature was at an earlier time, given what it is now? Is there an
explicit formula for u(t, x) in terms of this initial function f ? Moreover, can we see
what happens as time tends to infinity, i.e., the long-term behavior? Or, can we
determine what happened at t = −∞? That is, can we analyze backwards to figure
out what the temperature was at the very beginning of time?
REMARK. Physicists think that the temperature at a point on the rod is propor-
tional to the velocity of the molecule at that point in the rod. Since the square of
the velocity V 2 is proportional to the kinetic energy mV 2 /2, we presume that the
2
R ∞ |u(t, x)|
function
2
is proportional to the “instantaneous” energy at the point x, and
so −∞ |u(t, x)| dx should represent the total energy at time t. This leads us to our
R∞
first discovery about the function u(t, x). It must satisfy −∞ |u(t, x)|2 dx < ∞ for
every time t. The total energy must be finite at any given time.
Mathematicians say that these functions of x belong Rto L2 .
∞
Said precisely, a function φ is said to belong to Lp if −∞ |φ(x)|p dx < ∞.
Physicists also believe that this temperature function u must satisfy the following
partial differential equation, called the heat equation.

∂u ∂2u
(t, x) = (t, x).
∂t ∂x2

EXERCISE 1.1. Can you think of any solutions to this partial differential equation?
How about u(t, x) = 0 or u(t, x) = 1, or u(t, x) = 2t + x2 ? How about u(t, x) =
2
ek t × ekx ? Do any of these functions satisfy the finite energy (L2 ) requirement?
Can you think of any other solutions of the heat equation?
EXAMPLE 1.1. For all real numbers x and all positive t, define

1 − x2
k(t, x) = √ e 4t .
4πt

EXERCISE 1.2. Verify that this function k satisfies the heat equation. Notice that
this function is not defined for t = 0. Can you figure out what happens to k(t, x)
as t approaches 0?
DEFINITION. The function k(t, x) is called the fundamental solution of the heat
equation on the line, and it is also frequently referred to as the heat kernel.
1
2

2 R∞
EXERCISE 1.3. Let f (x) = e−πx , and let C = −∞
f (x) dx. Justify the following
computations Z ∞
C2 = C f (x) dx
−∞
Z ∞
= Cf (x) dx
−∞
Z ∞Z ∞
= f (y) dyf (x) dx
−∞ −∞
Z ∞Z ∞
2
= e−πy dyf (x) dx
−∞ −∞
Z ∞Z ∞
2
= f (x)e−πy dy dx
−∞ −∞
Z ∞Z ∞
2 2
= e−πx e−πy dydx
−∞ −∞
Z ∞Z ∞
2 2
= e−π(x +y ) dydx
−∞ −∞
Z2π Z ∞
2
= e−πr r drdθ
0 0
2π ∞
−1
Z Z
2
= e−πr (−2πr) drdθ
0 2π 0
Z 2π
−1 2
= e−πr [∞
0 dθ
2π 0
2π
−1
Z
= (0 − 1) dθ
2π 0
= 1.
Conclude that Z ∞
2
e−πx dx = 1.
−∞

R∞ 2
EXERCISE 1.4. (a) For any positive number a, compute −∞ e−ax dx. (Use Ex-
ercise 1.3, and make a change of variables.) R∞
(b) Let k(t, x) be the function in Example 1.1 above. Compute −∞ |k(t, x)| dx
R∞
and −∞ |k(t, x)|2 dx, and investigate what happens to this total energy as t tends
to infinity. Also, what happens to this energy
R∞ as t approaches 0?
2
(c) For any real number b, compute −∞ e−(x+b) dx.
R∞ 2
(d) For a, b, and c real numbers, with a > 0, compute −∞ e−(ax +bx+c) dx.
HINT: Complete the square and use earlier parts of this exercise.
One of our main goals is to figure out where this funny function k(t, x) comes
from. You surely couldn’t have guessed that this would be a solution to the heat
equation. It may not be any good anyhow, because it doesn’t really fit our initial
value problem. Why do we even consider this function?