An Introduction to Partial

Differential Equations
Andrew J. Bernoff

LECTURE 2
Cooling of a Hot Bar:
The Diffusion Equation

2.1. Outline of Lecture

An Introduction to Heat Flow

Derivation of the Diffusion Equation
Examples of Solution to the Diffusion Equation
The Maximum Principle
Energy Dissipation and Uniqueness

2.2. An Introduction to Heat Flow

A classical example of the application of ordinary differential equations
is Newtons Law of Cooling which, basically, answers the question How
does a cup of coffee cool? Newton hypothesized that the rate at which
the temperature, U (t), changes is proportional to the difference with
the ambient temperature, which we call U ,
dU
= (U U ).
dt
Here is a positive rate constant (with units of inverse time) that
measures how fast heat is lost from the coffee cup to the ambient environment. If we specify the initial temperature,
(2.1)

(2.2)

U (0) = U0 ,
1

ANDREW J. BERNOFF, AN INTRODUCTION TO PDE

we can solve for the evolution of the temperature,

(2.3)
U (t) = U + (U0 U )et .

Figure 2.1: (a) A coffee cup (b) Its temperature as a function of time.
(draw your own figure).
If we graph the temperature as a function of time, we see that it decays
exponentially to the ambient temperature, U , at a rate governed by .
When we derived Newtons Law of cooling we made several assumptions most importantly that the temperature in the coffee cup did
not vary with location. If we account for the variation of temperature with location, we can derive a PDE called the heat equation or,
more generally, the diffusion equation. If the temperature, U (x, t) is
a function of a single spatial variable, x, we will show that it satisfies
the diffusion equation,
Ut = DUxx ,
where D is a constant known as the thermal diffusivity. In higher
dimensions, the equation can be written
Ut = D2 U,
where 2 is the Laplacian.

2.3. Derivation of the Diffusion Equation

The diffusion equation will be our second example of a conservation
law; we can derive the equation by accounting for the flow of thermal
energy. Suppose we consider a metal bar, with a uniform cross-sectional
area, A, whose temperature, U (x, t), is a function of time, t, and the

LECTURE 2. COOLING OF A HOT BAR:THE DIFFUSION EQUATION 3

position, x, along the bar (that is we assume the temperature is uniform

in every cross-section).

Figure 2.2: Conservation of heat in a metal bar

of cross-sectional area A.
(draw your own figure).
Let the thermal energy in the region a < x < b is given by
Z
(2.4)

E = 0 cv A

U (x, t) dx
a

The important term in the integral is the temperature, U (x, t), measured in degrees. The remaining constants, A, the cross-sectional area
(with units of [(length)2 ]), 0 , the density [mass/(length)3 ] and cv , the
heat capacity [energy/(degree mass)] are physical properties of the
material think of them as being obligatory for making the units work
out.
We wish to equate the change in thermal energy to the heat flux
out of the bar through the planes at x = a and x = b. To do this we
use Fouriers heat law which states that the flux density of thermal
energy, q(x, t) is proportional to the temperature gradient,
(2.5)

q(x, t) = kUx ,

where the negative sign reflects the fact that heat flows from hot to cold,
just as in Newtons law of cooling, with a constant of proportionality,
k, called the thermal conductivity [(energylength)/(degreestime)].

ANDREW J. BERNOFF, AN INTRODUCTION TO PDE

Figure 2.3: Heat flux is from hot to cold!!

(draw your own figure).
Now, the total flux of thermal energy into the into the region a <
x < b is given by
(2.6)

Q = A[q(a, t) q(b, t)],

where we multiply by the area A to get the total flux through the
cross-section.
By conservation of energy, the rate of change of the energy
between a and b is given by the flux into the region,
dE
= Q.
dt

(2.7)

Once again we can rewrite the flux by a clever application of the fundamental theorem of calculus,
(2.8)
(2.9)

Q = A[q(a, t) q(b, t)] = Aq(x, t)|x=b

x=a
Z b
= A
qx dx.
a

We now rewrite the conservation of energy equation as

(2.10)

 Z b
Z b
Z b
dE
d
=
cv A
U dx =
cv AUt dx = Q = A
qx dx,
dt
dt
a
a
a

LECTURE 2. COOLING OF A HOT BAR:THE DIFFUSION EQUATION 5

or, rearranging
b

cv A Ut + Aqx dx = 0.

(2.11)
a

Since this is true for every interval a < x < b, the integrand must
vanish identically. So
(2.12)

cv AUt + Aqx = 0.

Substituting for the flux function q(x, t) = kUx yields

(2.13)

cv AUt kA(Ux )x = 0.

Rearranging the equation yields the diffusion equation,

(2.14)

Ut = DUxx ,

where the diffusivity, D = k/(cv ), is a constant which is determined

by the geometry and physical properties of the metal bar.
To complete the description of the problem, we need to supplement
the diffusion equation with boundary conditions and initial conditions.
Suppose we consider a bar of finite length L, occupying the region
0 < x < L. At the boundaries of the metal bar we can specify a fixed
temperature,
(2.15)

U (0, t) = U0

U (L, t) = U1 ,

which are usually referred to as Dirichlet boundary conditions. Alternatively, we could specify a heat flux,
(2.16)

q0 = q(0, t) = kUx (0, t)

q1 = q(L, t) = kUx (L, t).

Specifying the gradient across the boundary is referred to as Neumann

boundary conditions.
Finally, we also need to specify the initial temperature distribution,
(2.17)

U (x, 0) = f (x)

0 < x < L.

We will demonstrate below that the solution to this problem (if it

exists) is unique; later in this course we will solve this problem using
the method of separation variables.
For completeness, we also comment here that the problem can be
posed on the infinite line, < x < sometime called the Cauchy
problem in this case one usually replaces the boundary condition
with the specification that the temperature remains bounded as we
approach infinity,
(2.18)

lim |U (x, t)| < C,

ANDREW J. BERNOFF, AN INTRODUCTION TO PDE

for some constant C. This condition may seem superfluous at first

glance, but actually is necessary to stop heat from leaking in from infinity (speaking very, very informally an infinite source of heat infinitely
far away can have a finite effect in a short amount of time). If you are
interested in details, look for the examples of Tychonov in a PDEs
text1.

2.4. Examples of Solution to the Diffusion Equation

We can summarize the last section by restating a well-posed problem
for the diffusion equation on the interval 0 < x < L with Dirichlet
boundary conditions,
The Dirichlet Problem for the Diffusion Equation
(Non-homogeneous Boundary Conditions)

U (0, t) = U0

Ut = DUxx
0 < x < L, t > 0
U (L, t) = U1
t>0
U (x, 0) = f (x)
0 < x < L.

DE
BC
IC

Solving the general problem will have to wait, but we can find some
specific solutions to the problem using the ideas of Separation of
Variables. For the moment, we will restrict ourselves to homogeneous
boundary conditions,
The Dirichlet Problem for the Diffusion Equation
(Homogeneous Boundary Conditions)
Ut = DUxx
0 < x < L, t > 0
U (0, t) = 0 U (L, t) = 0
t>0
U (x, 0) = f (x)
0 < x < L,

DE
BC
IC

If you want, you can skip the derivation for the moment and jump
ahead to Exercise 1, if you dont mind the solution appearing deus ex
machina ( a fancy term for out of thin air).
1See,

for example, T. W. K
orner, Fourier Analysis, Cambridge University Press,
p. 338.

LECTURE 2. COOLING OF A HOT BAR:THE DIFFUSION EQUATION 7

2.4.1. A Solution to the Homogeneous Dirichlet Problem

Let us look for solutions to the homogeneous Dirichlet problem of the
form
(2.19)

U (x, t) = X(x)T (t)

we find from the differential equation (DE) that

(2.20)

XTt = DXxx T

and dividing by XT we find

Tt
Xxx
=
= .
DT
X
where is to be determined. Now because Tt /DT is only a function
of t and Xxx /X is only a function of x we know that must be
independent of x and t respectively, and therefore must be a constant
consequently it is known as the separation constant. We can now
solve the resulting ODE for T (t)
(2.21)

(2.22)

Tt = DT

T (t) = et ,

or some constant multiple of it.

We now look for a solution for the X(x) equation that also satisfies
the homogeneous boundary conditions. From the boundary conditions
(BC), we know that
(2.23)
(2.24)

U (0, t) = X(0)T (t) = 0

U (0, t) = X(L)T (t) = 0

X(0) = 0
X(L) = 0

So finally we conclude that we are looking for solutions to the Boundary Value Problem for X(x),
(2.25)

Xxx + X = 0,

X(0) = 0 X(L) = 0.

Solving the DE, we find that

(2.26)

X(x) = B cos( x) + C sin( x)

and applying the boundary conditions we see that X(0) = 0 implies

that B = 0, and that

(2.27)
C sin( L) = 0.
Consequently, a non-trivial solution (that is a solution for which X(x) 6=
0) for X(x) can be found if and only if
 n 2
(2.28)
= n
for n = 1, 2, 3 . . .
L

ANDREW J. BERNOFF, AN INTRODUCTION TO PDE

for which we find

(2.29)

X(x) = Xn (x) sin

 nx 

for n = 1, 2, 3 . . . ,
L
or some constant multiple of it. These special values of are called
eigenvalues and the associated functions, Xn (x), are known as eigenfunctions.
Multiplying the solution for Xn (x) and T (t) together finally yields
a solution for Un (x, t),
 nx 
n 2
(2.30) U (x, t) = Un (x, t) sin
e( L ) Dt for n = 1, 2, 3 . . . .
L
The method of separation of variables is very powerful it will be
one of our primary tools for finding solutions to PDEs in the coming
lectures.

Exercise 1. Verify that

 nx 
n 2
Un (x, t) sin
e( L ) t for n = 1, 2, 3 . . . ,
L
satisfies the diffusion equation Ut = DUxx and the homogeneous boundary conditions U (0, t) = U (L, t) = 0. Explain why any linear combination of Un ,

 nx 
X
X
n 2
an sin
an Un (x, t) =
U (x, t) =
e( L ) Dt
L
n=1
n=1
also satisfies the diffusion equation and the homogeneous boundary
condition. Does it worry you that this is an infinite sum? What initial
condition, U (x, 0), does this correspond to?

LECTURE 2. COOLING OF A HOT BAR:THE DIFFUSION EQUATION 9

2.4.2. A Solution to the Cauchy Problem

We can also consider a solution to the Cauchy problem for the diffusion
equation, which you hopefully remember is the problem posed on the
entire real line,
The Cauchy Problem for the Diffusion Equation
Ut = DUxx
< x < , t > 0
lim |U (x, t)| < C
t > 0,

U (x, 0) = f (x)

< x < .

DE
BC
IC

While there are many clever derivation for the solution to this problem,
for the moment I will simply give you the most important solution,
usually called the fundamental solution or the diffusion kernel,
x2
1
(2.31)
U (x, t) = G(x, t + ) p
e 4D(t+ ) .
4D(t + )
where is a constant (which we will assume is positive). This solution
can be used to construct a general solution of the diffusion equation
for an arbitrary initial condition, f (x).

Exercise 2. Verify that

G(x, t + ) p

x2

e 4D(t+ ) .

4D(t + )
satisfies the diffusion equation and the boundary conditions for the
Cauchy problem when > 0 . Show that this solution corresponds
to a Gaussian with time varying width and height. How does the
Gaussians width, height and area vary in time?

10

ANDREW J. BERNOFF, AN INTRODUCTION TO PDE

2.5. The Maximum Principle

Looking at solutions to the heat equation, we note that they tend to
average out maximums and minimums. We can develop some intuition
for this by considering what the equation says. Basically, Ut = DUxx
means: The temperature is decreasing when the profile is convex down
and the temperature is increasing when the profile is convex up.

Figure 2.4: The heat equation interperted graphically

(draw your own figure).
From which we conclude that interior maximums in temperature are
decreasing and interior minimums of temperature are increasing. This
reasoning is not quite airtight (how to make it tighter is a good question
to ponder). We can give a rigorous statement (without proof) of the
maximum principle:
Theorem 2.32 (Maximum Principle for the Diffusion Equation). If
u(x, t) satisfies the Dirichlet problem for the diffusion equation in the
semi-infinite strip 0 < x < L, 0 < t, then it assumes its maximum
value (as a function of x and t) either initially (when t = 0) or on the
lateral boundaries (where x = 0 or x = l).
The same is also true of the minimum of u(x, t). A proof can be
found in most advanced PDE texts.
Exercise 3. Interpret the solutions we have found for the diffusion
equation in terms of the maximum principle. Show examples where
the maximum value of u(x, t) occur in the initial condition and on the
lateral boundaries.

LECTURE 2. COOLING OF A HOT BAR:THE DIFFUSION EQUATION11

2.6. Energy Dissipation and Uniqueness

By looking at what is normally known as energy for the diffusion equation, we can show that the solution for the Dirichlet problem is unique.
Note this energy is a mathematical construct, not to be confused with
the thermal energy discussed in the derivation of the diffusion equation.
First, suppose that U (x, t) is a solution to the homogeneous Dirichlet problem,
Ut = DUxx
0 < x < L, t > 0
U (0, t) = 0 U (L, t) = 0
t>0
U (x, 0) = f (x)
0 < x < L,

DE
BC
IC

Lets define, the energy,

Z
1 L 2
(2.33)
W =
U dx,
2 0
which is a function of t dependent on the particular solution U (x, t)
(technically it is a function of t and a functional of U (x, t)). Note
that W 0 with W = 0 only for the trivial solution U (x, t) = 0.
If we differentiate the energy with respect to time, we find
Z L
dW
=
U Ut dx,
dt
0
Z L
= D
U Uxx dx,
0
Z L
=
(Ux )2 dx + U Uxx |x=L
x=0 ,
0

where we have substituted the DE and used integration by parts. Now,

applying the BCs, we fine that the boundary terms from the integration by parts vanish, so,
Z L
dW
=
(Ux )2 dx 0
dt
0
Now, we can conclude that W is decreasing (that is energy is dissipated!!)unless Ux = 0, that is to say that U is constant. As the
only constant solution satisfying the boundary conditions is U = 0, we
might be tempted to conclude that the solution always decays to this
trivial state. This turns out to be true, although one must invest some
analysis to show it rigorously.

12

ANDREW J. BERNOFF, AN INTRODUCTION TO PDE

Figure 2.4: (a) A solution to the homogeneous Dirichlet problem for

the heat equation. (b) The corresponding energy, W, which is
decreasing to zero.
(draw your own figure).
A second conclusion one can reach is that if f (x) = 0, that U (x, t) =
0 for all t > 0. This follows quickly because W = 0 at t = 0, it is nonincreasing and non-negative. While this seems like a trivial result, it
has a very powerful consequence. Suppose we had two solutions to the
non-homogeneous Dirichlet problem, call them V1 and V2 . You should
be able to convince yourself that there difference U = V1 V2 satisfies
the homogeneous Dirichlet problem with f (x) = 0. Consequently, we
know that U (x, t) = 0 for all t > 0, which implies V1 = V2 . From
this we conclude that The solution to the non-homogeneous Dirichlet
problem is unique, a powerful result indeed.

Exercise 4. Convince yourself the energy argument for uniqueness of

solutions in the previous paragraph is correct. Show that a similar
argument can be made for the Neumann problem.

LECTURE 2. COOLING OF A HOT BAR:THE DIFFUSION EQUATION13

2.7. Challenge Problems for Lecture 2

Problem 1. Consider the diffusion equation with homogeneous Neumann boundary conditions.
Ut = DUxx
0 < x < L, t > 0,
Ux (0, t) = 0 Ux (L, t) = 0
t > 0,
U (x, 0) = f (x)
0 < x < L.

DE
BC
IC

(a) Explain physically why this corresponds to the diffusion of heat

in a metal bar with insulated ends. Make sure you understand
what each of the equations corresponds to.
(b) Show that
(i)

U0 (x, t) = 1
 nx 

n 2

e( L )

Dt
n = 1, 2, 3, . . .
L
satisfy both the diffusion equation (DE) and the homogeneous
Neumann boundary conditions (BC).
(c) Write down a general solution as a linear combination of the
solutions you found in part (b). What does this say about
f (x) if we assume that this solution also satisfies the initial
condition (IC)?

(ii)

Un (x, t) = cos

Problem 2. In this problem, we will argue that for the homogeneous

Neumann problem discussed in Problem 1, that the solution approaches
a constant temperature, given by the average of the initial temperature.
(a) Suppose we define the total heat energy in the bar as
Z L
U (x, t) dx.
Q(t) =
0

Show that Q is conserved, that is that it is independent of

time (Hint: compute dQ
).
dt
(b) Use the initial condition to compute Q in terms of f (x).
(c) Modify the energy argument in the previous section show that
the energy is decreasing unless U (x, t) is constant. Use this to
argue that U (x, t) approaches a constant solution as t .
(d) Finally, use parts (a) and (b) of the problem to show that there
is only one possible constant solution for U that is consistent
with the conservation of Q. Show that solution corresponds
to the bar approaching the average temperature of the initial
condition.

14

ANDREW J. BERNOFF, AN INTRODUCTION TO PDE

Problem 3. Previously we showed that the diffusion kernel,

x2
1
U (x, t) = G(x, t + ) p
e 4D(t+ ) ,
4D(t + )
satisfies the diffusion equation with an initial condition
U (x, 0) = G(x, ).
(a) Show that the total heat ,
Z
U (x, t) dx,
Q(t) =

is conserved, and in fact Q(t) = 1, for the heat kernel.

(b) Show that as 0, that G(x, ) 0 for x 6= 0 and that
G(0, ) .
(c) Explain why the solution G(x, t) (i.e. with = 0) corresponds
to introducing a unit amount of heat concentrated at the origin
when t = 0. This is called a -function initial condition.

Problem 4. A curious property of the diffusion equation is that both

derivatives and integrals of solution satisfy the diffusion equation also.
This can be used to generate new solutions from existing ones.
(a) Show that if U (x, t) satisfies the diffusion equation then (x, t) =
Ux satisfies the diffusion equation also (Hint: differentiate both
sides of the diffusion equation with respect to x).
(b) Generate a new solution for the heat equation by differentiating
the heat kernel with respect to x. Graph this solution (MAPLE
may be useful here) what does it look like?
(c) Show thatR if (x, t) satisfies the heat equation, then so does
x
U (x, t) = x0 (, t) d.
(d) Show that
Z x/4Dt
1
1
2
U (x, t) = +
ez dz
2
0
satisfies the heat equation, by showing its derivative is the heat
kernel. Graph the solution at various times what physical
problem does this solution correspond to ?