working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11

11 19:58
 CHAPTER 6
The Diffusion Equation

The diffusion equation, like the wave equation, provides a way to analyse some important
physical processes that require evaluation as a function of space and time. But we shall find
that diffusion solutions have properties that in several ways are very different from wave
solutions.
Examples of physical variables that can diffuse include temperature, and concentration
of pollutants. Certain probabilities also diffuse, such as the probability of finding a drunken
person at a particular position at a particular time, if he or she strays from an initial position
by taking many steps, each of which has an equal chance of being oriented in any direction.
Finding the position of a small particle undergoing Brownian motion is similar to that of
finding the drunk — it is best expressed in terms of probabilities, satisfying a diffusion
equation.
Here, we present the main ideas using heat as an example.

6.1 Heat Flow by Conduction/Diffusion: an Example of the Diffusion Equa-

tion

Let us use the symbols ρ to denote mass density, c to denote the specific heat per unit mass,
and T to denote the absolute emperature. Then the heat per unit volume is ρcT .
Heat can spread by convection, radiation, and conduction. We shall consider only the
latter mechanism, governed by the basic rule that “heat flux is proportional to temperature
gradient,” that is,

q = −K ∇T. (6.1)

Here, q is the rate of flow of heat across unit surface area, so it has the units of heat/(area ×
time). Heat flows from hot regions to cold, hence the minus sign in (6.1). The experimental
fact that heat flux q is proportional to temperature gradient, introduces the need for a constant
of proportionality. Hence the factor K in (6.1). K is called the thermal conductivity.1

∂T
1. More generally: K is a second-order tensor, and q = −K · ∇T , or qi = −K i j . But we shall
∂x j
assume an isotropic conductivity, for which K i j = K δi j , leading to (6.1).
101

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 102 Chapter 6 / THE DIFFUSION EQUATION

q
surface S

dS

volume V

FIGURE 6.1
An arbitrary volume V is shown, together with its surface S. The flux of heat out of S is characterized
by the vector q, which need not be orthogonal to the surface. The rate at which heat crosses the area
element dS, outwards, is given by q · dS.

Conservation of energy requires that

the rate of increase of heat throughout a volume V

= the rate of production of heat from sources within V (6.2)
− the rate of loss of heat out through the surface S of V .

If we denote the heat production per unit volume per unit time by A(x, t), it follows
that
  
∂
(ρcT ) d V = A dV − q · dS (6.3)
∂t V V S

(ignoring convection, radiation, the effects of circulating fluids, and many other phenomena
that in practice may lead to additional terms in this equation). V here is any volume in which
heat is present, and S is the surface of V (see Figure 6.1).
We can use Gauss’s divergence theorem to convert the last term in (6.3) to a volume
integral, via
  
q · dS = ∇ · q d V = − ∇ · (K ∇T ) d V from (6.1).
S V

 ∂T
The first term in (6.3) may be written as V ρc d V , if c is not time dependent, and if
∂t
we use the fact that ρd V is not time dependent (it is just the constant mass of an original
volume element). So all the terms in (6.3) may be collected to give a single volume integral
as
  
∂T
ρc − ∇ · (K ∇T ) − A d V = 0. (6.4)
V ∂t

But V is an arbitrary volume. It follows that the integrand in (6.4) must be zero wherever

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 6.1 Heat Flow by Conduction/Diffusion: an Example of the Diffusion Equation 103

Surface boundary condition: T = constant + T0 e - i ω t on z = 0

z=0

FIGURE 6.2
A horizontal surface is shown, which is subjected to period heating that maintains temperature
as a constant plus a sinusoidal wave with amplitude T0 and frequency ω (so that the period itself
is 2π ÷ ω). Taking the depth direction as the z-axis, and assuming no horizontal variation in the
temperature, we can use the diffusion equation to find temperature as a function of depth.

it is continuous. 2 Hence
∂T
ρc = ∇ · (K ∇T ) + A, (6.5)
∂t
which is an example of a diffusion equation. It is a second-order partial differential equation
with a double spatial derivative and a single time derivative.
If we assume that K has no spatial variation, and if we introduce the thermal diffusivity
κ by

K
κ= , (6.6)
ρc

then the diffusion equation assumes a standard form as

1 ∂T A
= ∇ 2T + . (6.7)
κ ∂t K
We shall solve this equation in four completely different examples of conductive heat
flow, using the solution methods to introduce a number of basic properties of diffusion, and
some useful math methods.

6.1.1 PERIODIC HEATING OF THE GROUND SURFACE

Our first example, shows the way in which the constant κ relates the spatial and temporal
scales of a simple diffusion problem.
The problem itself is shown in Figure 6.2: a horizontal surface is given a periodic
temperature variation, and we wish to find how temperature varies as a function of space
and time within the volume of material below the surface.

2. If the integrand were non-zero at a point P, say, we could find a small volume that includes P
and that violates (6.4).

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 104 Chapter 6 / THE DIFFUSION EQUATION

In the applied boundary condition on z = 0, namely T = constant + T0e−iωt (see

Figure 6.2), the constant is just the average absolute temperature. There is no source term
in the diffusion equation (6.7), so

1 ∂T
= ∇ 2T (6.8)
κ ∂t

and it is reasonable to seek a solution for T = T (x, t) in the special form

T = T (z, t) = constant + e−iωt Z (z). (6.9)

This trial form of solution makes two assumptions: first, that temperature depends spatially
only on the depth, and not on horizontal position; and second that the time variation is
also of the form e−iωt , just like the boundary condition. These are reasonable assumptions,
given the linearity of the diffusion equation, and the fact that the boundary condition has no
horizontal dependence, nor any dependence on initial conditions. (We have simply assumed
that the surface boundary condition has always applied, for all times in the past.)
From the boundary condition on z = 0, we see from (6.9) that Z (z) = T0 when z = 0.
But what is the variation of Z with depth? This is where we use the diffusion equation, for
substitution of (6.9) into (6.8) gives

−iω d2 Z
Z= ,
κ dz 2

which is solved by

Z (z) = Ae−λz + Beλz , (6.10)

−iω
where the constant λ is given by λ2 = .
κ
The square root of −i is a complex number with amplitude 1 and phase − π4 or 3π 4 .
−iπ/4 1−i
Note that e can also be written as √ . Without loss of generality, we then have
 2
ω 1−i −λz
λ= √ and both the solutions e and eλz in (6.10) must be considered.
κ 2
It is an obvious requirement of the temperature, that it not become infinite as z → ∞.
Therefore, B = 0 and only the e−λz solution is needed. We noted previously that the
boundary condition requires Z (0) = T0, so the constant A is simply T0 and the complete
solution to this problem is
 
ω ω
− z i z
T = T (z, t) = constant + T0e 2κ e 2κ e−iωt . (6.11)

The constant here is the same one that appears in the boundary condition. Because of the
last exponential factor in (6.11), the temperature penetrates the ground as an oscillation
with the same period as the boundary condition. This was our assumption, in the trial form
of solution (6.9). But now we have learned from the first exponential factor in (6.11) that

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 6.1 Heat Flow by Conduction/Diffusion: an Example of the Diffusion Equation 105

the amplitude of this oscillation is damped (attenuated) with depth. At the depth z 1 the
2
amplitude of the temperature fluctuation is a half that at the surface, if

ω
− z1 1
e 2κ 2
= .
2

Since log10 0.5 ∼ −0.3 and loge 10 ∼ 2.3, it follows that


ω
− z 1 = −0.3 × 2.3
2κ 2

and hence that the “half amplitude depth” is given by


κ × period
z 1 = 0.69 × (6.12)
2 π

where the period of the oscillation is 2π/ω. Typical values of κ for soil and rock are
approximately 0.01cm2/s. So, the depth at which the temperature oscillation is reduced by
a factor of 2 from the surface value, is about 10 cm for a daily oscillation, about 2 meters
for an annual oscillation, and about 300 m for a very long period oscillation of 20,000 years
(representative of change on the scale of ice ages). Fortunately, when it comes to a decision
about how deeply to bury water pipes to avoid freezing in winter, the average temperature
at latitudes such as New York is well above freezing, so a depth considerably less than 2
meters is OK. We discuss below the way to combine daily and annual temperature changes,
and a general time-dependent boundary condition on z = 0.
If we interpret equation (6.12) as a relation between time and distance over which
temperature travels, then we see that the distance variable (z 1 ) is proportional to the square
2
root of the time variable (period), or distance2 ∝ time. This behavior is very different
from the way a wave propagates, where for example the dependence of properties is on
combinations of space and time such as (t − x/c) or (t − l · x/c). The spatial scale and the
time scale of a simple wave are proportional, and if we know how long it takes the wave
to travel a certain distance, then we expect the wave to travel double the distance in double
the time. In constrast, we see here for a simple diffusion solution, that it is the square of the
spatial scale which is proportional to the time scale.
From the third exponential factor in (6.11) it follows that the oscillation at depth z is
delayed from that at the surface. We can obtain the amount of the delay by writing
  
ω z
i z −iω t− √
e 2κ e−iωt = e 2ωκ , (6.13)

which can be interpreted to say that the change in phase due to depth (the first term on the
z
left-hand side of (6.13)) is equivalent to making a delay in time t by an amount √ .
2ωκ
Because the diffusion equation is linear, we can directly combine solutions of the
type (6.11) to obtain the result of more general boundary conditions. Thus, if

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 106 Chapter 6 / THE DIFFUSION EQUATION

30

C
15
o
T, 0
0 50 100 150 200 250 300 350 days

FIGURE 6.3
An annual variation of the surface temperature is shown — the boundary condition of Figure 6.2,
with constant = 15◦C, amplitude = 10◦C, and ω = 2π/365 radians/day.

T (z, t) = constant + T1e−iω1t + T2e−iω2t
z=0

for fixed values of T1, T2, ω1, ω2, then the temperature at depth z is given by
   
ω1 ω1 ω2 ω2
− z i z − z i z
T (z, t) = constant + T1e 2κ e 2κ e−iω1t + T2e 2κ e 2κ e−iω2t .(6.14)

Figure 6.3 shows the boundary condition for an annual variation, and Figure 6.4 shows
a combined annual and daily variation. Each oscillation is attenuated with depth, in the
combined solution (6.14).
More generally, the surface boundary condition may be a function of time that we can
think of as a summation of its frequency components:


T (z, t) = constant + f (t)
z=0
 (6.15)
1 ∞
−iωt
= constant + f (ω)e dω.
2π −∞

Applying the solution (6.11) to each frequency component and superimposing these
solutions, we see in this case that T as a function of depth and time is given by
 
 ω ω
1 ∞ − z i z
T (z, t) = constant + f (ω)e 2κ e 2κ e−iωt dω. (6.16)
2π −∞

The integrand here requires modification for negative values of frequency, but in practice
this solution can be turned into an integral over positive ω values using the results of
Problem 3.1. Because of the exponential decay of the integrand in (6.16) as frequency
increases, direct numerical integration can easily be made stable.
Typical values of the diffusivity (κ) are: approximately 10−6 m2/s for rocks, and values
in the range 10−4 to 10−5m2/s for metals.

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 6.1 Heat Flow by Conduction/Diffusion: an Example of the Diffusion Equation 107

20

10
C
o
T,

-10
240 250 260 270 280 290 300 days
310

FIGURE 6.4
For days 240 through 310 of Figure 6.3, an additional daily term is added (ω = 2π radians/day),
also with amplitude = 10◦C. This is a surface boundary condition. The question of how deeply
pipes should be buried in winter, to avoid freezing, is crudely answered by finding the value of z in
equation (6.14) that filters these inputs (T1e−iω1t , T2e−iω2t ), and keeps T above 0◦C.

6.1.2 A SIMPLE MODEL OF CRUSTAL HEAT FLOW

From simple observations made on or near the surface of the Earth, we find that radioactive
elements are concentrated in the Earth’s crust, and that therefore the materials making up
our planet have differentiated and become stratified over time. This conclusion is reached
from another solution of the diffusion equation (6.7) which we now discuss, in which we
take account of the source term.
A basic observation, is that temperature increases rapidly with depth inside the Earth,
as has been noted for hundreds of years by miners. The gradient varies from one location to
another, but typical values are on the order of 10◦ or 15◦ C per km depth, as measured near
the Earth’s surface. Presumably this gradient must reduce with depth, otherwise material
inside the Earth would melt below about 100 km.
Beginning almost a hundred years ago, measurements of heat flow coming out of
the Earth have been made by inserting temperature probes into the ground and carefully
T T (z 1) − T (z 2)
measuring the temperature gradient, for example as = . To obtain the
z z1 − z2
heat flow, equation (6.1) can be used. But it is first necessary to know the thermal conductivity
K for samples of the same material in which the temperature gradient is obtained in the
field. Measurements of K may be done on samples that are brought in to a lab, or sometimes
it is possible to measure K directly, in the field.
There are many possible causes of the Earth’s observed heat flow. Primordial heat,
left over from the time of our planet’s formation 4.6 billion years ago, is still coming out
as the interior cools. In the vicinity of volcanoes, or at sub-oceanic sites near mid-ocean
ridges, hot magma can drive water circulation that vents to the surface or the ocean floor.
Various chemical reactions can lead to heating. But a major contributor to the Earth’s heat
flow is derived from the radioactive decay of isotopes of U (uranium), Th (thorium), and K
(potassium).
It has long been observed that there is a simple correlation between measurements of
heat flow, and measurements of the heat productivity of rocks found at the Earth’s surface.
The correlation is indicated in Figure 6.5. In order to interpret observations of the type
shown in this Figure, we shall specify a simple model, and use heuristic methods. (That is,

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 108 Chapter 6 / THE DIFFUSION EQUATION

Q=aA+b
Q

FIGURE 6.5
A schematic of heat flow measurements and measurements of heat productivity A (see the definition,
given just prior to equation (6.3)). The data points are fit approximately by a straight line, which has
the form Q = a A + b with two constants, a and b, being the slope and intercept, respectively. We
shall develop a simple model, in which each of these constants has a physical interpretation.

z=0

1 2 3 4

z=H

Q0 Q0 Q0

FIGURE 6.6
A simple model of the Earth’s interior is assumed, in which heat producing rocks are confined to
the upper layer between z = 0 and z = H , so that the heat productivity A = 0 for z > H . Also, it is
assumed that the upper layer is composed of separate homogeneous blocks, in each of which the heat
productivity A is constant, different for each block (numbered 1, 2, 3, . . .). This productivity can be
sampled at the upper surface of each block along with a measurement of the heat flow at the surface
(thus giving a data point in Figure 6.5). In each block, heat flow is presumed to be vertical and steady
state. Additional heat is due to a deep source which supplies heat flow upwards, across the surface
z = H , of amount Q 0, which does not change laterally.

we will make simplifying assumptions in order to make progress. The assumptions can be
checked later, if necessary.)
Thus, we shall assume the model illustrated in Figure 6.6, in which a series of
homogeneous blocks, all with the same thickness b, each has a constant value of heat
production A — the source term in (6.7). The constant is in general different in each block,
each of which is a crude model of a region specified by its geological and geophysical
properties (these properties being different from one block to another). We shall assume
that in each block the heat flow is vertical, independent of horizontal position within the
block, and also independent of time. Thus our original diffusion equation (6.5) reduces to

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 6.1 Heat Flow by Conduction/Diffusion: an Example of the Diffusion Equation 109
 
d dT
K = −A
dz dz

in each block. Since we are assuming the heat flow is vertical, we can write the heat flux
as q = (0, 0, −Q). The minus sign is introduced, to acknowledge that we are taking the
direction of z increasing as the depth direction, but the heat flow is upwards. Therefore the
scalar Q has a positive value, and (6.1) reduces to the scalar equation

dT
Q=K .
dz

These last two equations give

dQ
= −A
dz

for each block, showing that there is a constant heat flow gradient in each block, due to the
distribution of heat-producing radioisotopes throughout the volume of the block. We can
integrate over the gradient to give the solution

Q(z) = −Az + constant. (6.17)

At z = H , let us assume there is an upward contribution to the flow of heat as shown in

Figure 6.6, of amount Q 0, coming f r ombelow. The constant in (6.17) is therefore AH + Q 0
and the solution for each block is Q(z) = A(H − z) + Q 0. It follows that


Q(observed) = Q(z) = AH + Q 0. (6.18)
z=0

When this relation is applied to different blocks, it gives the linear relation between observed
values of heat flow Q at the surface, and the observed values of heat productivity, shown
schematically in Figure 6.5. Furthermore, we see that the intercept b of this Figure is the
same as the constant Q 0 of Figure 6.6, and that the slope a of Figure 6.5 equals the layer
thickness H of Figure 6.6.
Surprisingly, the slope value a from data plotted as in Figure 6.5 turns out to be only
around 5 to 10 km — and definitely less than typical thicknesses of the continental crust.
Interpreting this value of a as the thickness of the blocks in our simple model (Figure 6.6),
we see that heat productivity has somehow been concentrated into shallow depths within the
Earth. This general conclusion is consistent with the gross observation of a high temperature
gradient near the Earth’s surface, and also with the requirement, noted above, that this
temperature gradient must be greatly reduced at lower depths (otherwise the Earth’s interior
would be largely molten).
The model we have analysed (Figure 6.6) is very simple, and our conclusions may be
modified if we make better assumptions. (For example, A could vary with depth in each
block in some plausible way, and we need to see how much the heat flow can be affected by
topography, and by a component of lateral flow near the vertical boundary between blocks.)
But still it appears from observations such as Figure 6.5 that radioactive U, Th, and K, may

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 110 Chapter 6 / THE DIFFUSION EQUATION

be concentrated not just in the crust, but in the upper part of the crust — and also that heat
flow at the surface has a significant component Q 0 derived from much greater depth.
Q was once measured in “heat flow units” which were microcalories per cm2 per s.
Values in the range around 1 or 2 of these units were common, with higher values (sometimes,
much higher) in regions associated with volcanic activity. Today, Q is measured typically in
milliwatts per square meter. One classic heat flow unit corresponds to 42 mW/m2. Pollack
et al. (1993)3 describe and analyse more than 20,000 measurements of Q. They report a
global heat loss of 44.2 terawatts, half of which comes from geologically young oceanic
lithosphere. The mean heat flows of continents and oceans are 65 and 101 mW m−2,
respectively. With the primitive model discussed in this subsection, Q 0 for young oceanic
lithosphere is a dominant fraction of the total observed heat flow emerging at the top of
oceanic sediments.
The broad interpretation of heat flow measurements can be based upon (6.18).
In the case of measurements made on continental crust, we can think of Q 0 as roughly
constant and A as varying laterally for different provinces. The total continental heat flow
is derived mostly from heat-producing isotopes that concentrate in the upper crust, though
the contribution from below the crust (Q 0) is a significant fraction of the total observed at
the surface.
In the case of measurements made on oceanic crust (usually, from measuring dT /dz
and K in soft sediments at the ocean floor), again we can use (6.18) but now A is quite small
and the dominant contribution is from Q 0, which varies laterally. Pollack et al. (1993)3 note
that measured values of total heat flow for the oceanic crust are observed to vary with crustal
age (t, say) according to Q(t) = Ct −1/2 for t in millions of years, where the constant C is
about 510 mW m−2 = Q(1).

6.1.3 DYKE INJECTION

Our third solution of the diffusion equation (6.7) introduces a method of analysis that has
wide application to the analysis of linear partial differential equations with a source term.
We suppose that a thin sheet of molten rock has been injected on the plane z = 0 at time
t = 0, and that the amount of heat is C joules per square meter. The heat from this molten
rock is conducted away over time, and we seek to obtain the resulting value of temperature
T as a function of position z (distance from the source), and time t. We shall assume the
sheet of inserted material is in a horizontal plane, and take the z-axis as the depth direction.
When the sheet has solidified, and cooled, the inserted rock is referred to as a dyke.
Our first task is to interpret the above description of the source, in terms of a specific
form for the heat production term A used in the diffusion equation (6.7). Recall that A(x, t)
is heat production per unit volume per unit time. Therefore

A=0 if z = 0 or if t = 0, (6.19)

since the heat in the present problem is introduced only at z = 0 and t = 0. But although

3. Pollack, H. N., S. J. Hurter, and F. R. Johnson, Heat flow from the Earth’s interior: analysis of
the global data set, Reviews of Geophysics, 31, 267–280, August 1993.

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 6.1 Heat Flow by Conduction/Diffusion: an Example of the Diffusion Equation 111

BOX 6.1
Dirac delta functions

There are many situations where we would like to idealize values of a function of time or
space, so that the function is somehow concentrated at a particular time or place (or both).
Typically, we want to express the function in a natural way with a notation that describes
when or where the concentration occurs, and what is the overall effect of the concentration.
Paul Dirac introduced key ideas on this subject, and we acknowledge his contributions by
referring to the idealized functions as Dirac delta functions.
Take, for example, the mass density ρ(x) that describes a point mass m located at position
x = ξ . We normally think of ρ as the mass per unit volume defined by the relation
δM
ρ(x) = lim
δV →0 δV
where δ M is the mass contained in a volume δV , which is centered on x. It follows that the
mass M contained in a volume V is given by

M= ρ(x) d V.
V

But for a point mass m located at position x = ξ , the idealized mass density has the following
two properties. First, it is zero almost everywhere, so

ρ(x) = 0 if x = ξ . (1)

Second, the mass density is strong enough at the point x = ξ to give a finite result when ρ
is integrated over a region of space that includes the point mass itself. Specifically, we have


0 if ξ is outside V
ρ(x) d V = (2)
V m if ξ is inside V .

The result here for ξ outside V is an obvious consequence of (1), because the integrand
must be zero everywhere. The result with ξ inside V is Dirac’s inspiration, which he saw
intuitively. The integrand in (2) now has some type of singularity at the point x = ξ , and
though the integrand is zero everywhere else, the singularity at this one point is strong
enough to give a finite result for the integration.
In practice we don’t need to inquire deeply into the nature of the infinity of the integrand
at x = ξ , because we can directly get everything we need from the two equations (1)
and (2). These equations become the defining property of the Dirac delta function in three-
dimensional space. Thus, we introduce the notation δ(x − ξ ) for a function with the two
properties

δ(x − ξ ) = 0 if x = ξ , (3)

and


0 if ξ is outside V
δ(x − ξ ) d V = (4)
V 1 if ξ is inside V .

In terms of this normalized Dirac delta function, the mass density ρ(x) corresponding to a
point mass m located at position x = ξ is given by

ρ(x) = mδ(x − ξ ). (5)

With this notation, we have obtained the desired device for describing where the concen-
tration of mass occurs (it is at x = ξ ), and the total amount of the integrated concentration
(it is m). If the point mass were at the origin (ξξ = 0), we would just write ρ(x) = mδ(x).

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 112 Chapter 6 / THE DIFFUSION EQUATION

BOX 6.1 (continued)

In one dimension, we can write δ(x − ξ ) for the function of the scalar independent
variable x that has the properties
 b
δ(x − ξ ) = 0 if x = ξ, and δ(x − ξ ) d x = 1 if a < ξ < b (6)
a

(that is, if the range of integration includes the value ξ ). Using cartesian coordinates, it
follows that

δ(x − ξ ) = δ(x1 − ξ1)δ(x2 − ξ2)δ(x3 − ξ3). (7)

To obtain an intuitive understanding of the Dirac delta function in the time domain,
consider the integrated effect of a force F(t) applied at a point
 t over a period of time from
t = t1 to t = t2. The impulse of this force is defined as P = t12 F(t) dt. What then is the
impulse of a hammer blow, applied at t = T ? And what is the corresponding description of
the applied force F(t)? To obtain a finite impulse P from a hammer blow that is associated
with a force F applied only at one instant, we can write F(t) = Pδ(t − T ) where the
time-domain Dirac delta function is defined by the two properties
 t2
δ(t − T ) = 0 if t = T, and δ(t − T ) dt = 1 if t1 < T < t2. (8)
t1

Elsewhere in these notes, we have pointed out additional properties of Dirac delta
functions. For example, in Section 3.4 we noted the substitution property, which for
continuous functions of time can be written as
 t2
f (t)δ(t − T ) dt = f (T ) if t1 < T < t2. (9)
t1

That Section noted various parallels between Kronecker delta functions and Dirac delta
functions, and showed how a formal definition of a Dirac delta function can be obtained
as the limit of functions that have a finite range of non-zero values. In Box 2.2 we briefly
noted that the Dirac delta function in the time domain, δ(t − T ), can be thought of as the
time derivative of the Heaviside step function H (t − T ) (which is zero for t < T , and unity
for T < t). Thus,
d
[H (t − T )] = δ(t − T ).
dt
The Fourier transform of δ(t − T ) is easily obtained from an application of (9), and we find
 ∞
δ(t − T )eiωt dt = eiωT .
−∞

It follows that the Fourier transform of δ(t) is just 1. This is an extreme example of the
uncertainty principle discussed in Section 3.3.1: if information is totally concentrated in
the time domain (like δ(t)), then it must be completely spread out in the frequency domain
(equal to a constant value 1, the same for all frequencies).
The final property that we may note for Dirac delta functions, is the practical one that
their physical dimension is given by the reciprocal of the independent variable over which
these functions are integrated to give a unit result. Thus, from (8), since δ(t − T ) dt is
dimensionless, it follows that δ(t − T ) has the physical dimension of 1/time, or frequency.
Similarly from (6) it follows that δ(x − ξ ) has the physical dimension of 1/length, or
wavenumber; and from (4) we see that δ(x − ξ ) or δ(x) has the dimension of 1/volume. The
term δ(x − ξ ) in the right-hand side of (5) therefore contributes not only the concentrated
part of ρ (specifying where ρ is not zero), but also the “per unit volume” part of the physical
units.

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 6.1 Heat Flow by Conduction/Diffusion: an Example of the Diffusion Equation 113

A is zero almost everywhere, as described by (6.19), A is sufficiently strong at z = 0 and

t = 0 that we get a finite amount of heat when integrating A over ranges of z and t that
include the place and time for which heat emplacement occurs. That is,
 z 2  t2
A(z, t) dz dt = C (6.20)
z1 t1

provided z 1 < 0 < z 2 and t1 < 0 < t2.

The two properties given in (6.19) and (6.20) are the defining properties of the Dirac
delta function which is concentrated at z = 0 and t = 0. From the discussion in Box 6.1,
we can write

A = C δ(z) δ(t). (6.21)

Note that the physical dimensions of (6.21) make sense. A is the heat production rate, in
units of energy per unit volume per unit time. And on the right-hand side of the equation
we see that the units are those of C (energy per unit area), per unit length (from δ(z)), per
unit time (from δ(t)). So the units match on both sides of the equation. The interpretation
of the right-hand side, is that the delta functions serve to characterize the concentrated
(infinite) densities with which energy is pumped into the rock mass, per unit length in the
z direction (the spatial density is zero everywhere except at z = 0 where it is unbounded),
and per unit time (the rate, or temporal density, is zero at all times except at t = 0 where it
is unbounded). The only detail that we need to know, of the unbounded nature of the source
at the special place (z = 0) and time (t = 0) where it is concentrated, is the integrated effect
given in (6.20). Equation (6.21) is simply a restatement, using Dirac delta function notation,
of the two earlier equations (6.19) and (6.20).
The relevant diffusion equation, (6.7), has derivatives perpendicular to the z axis that
can be ignored in the present problem, since heat is conducted only in the z direction. Since
T depends only on z and t, our task is reduced to solving the equation

1 ∂T ∂ 2T C
= 2 + δ(z) δ(t), (6.22)
κ ∂t ∂z K
given that T = constant for all z, and t < 0. We shall take this initial constant value as zero.
(If it were not zero, we can use the same methods we are about to use, and add the constant
back at the end.)
To solve (6.22) we shall take the following three steps:

(i) transform z → k and t → ω by Fourier methods;

(ii) solve algebraically for the transformed solution T = T (k, ω); and
(iii) invert back to the (z, t) domain by evaluating the inverse Fourier transforms (two of
them).

Let us discuss each of these steps in turn.

(i) Equation (6.22) applies for all z and all t (provided we interpret the delta functions
as described in Box 6.1). So we can multiply each side of the equation by e−i(kz−ωt)

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 114 Chapter 6 / THE DIFFUSION EQUATION

BOX 6.2
On Fourier transforms of spatial and temporal derivatives

Working with the Fourier transform pair

 ∞  ∞
1
f (ω) = f (t) eiωt dt and f (t) = f (ω) e−iωt dω, (1)
−∞ 2π −∞

we can differentiate the second equation here to obtain

 ∞
df 1
= (−iω) f (ω) e−iωt dω.
dt 2π −∞
df
From examination of the integrand here, it follows that the Fourier time transform of is
dt
(−iω) f (ω).
With the spatial transform pair
 ∞  ∞
1
g(k) = g(t) e−ikz dz and g(z) = g(k)) eikz dk, (2)
−∞ 2π −∞
the second equation differentiates to give
 ∞
dg 1
= (ik) g(k) eikz dk.
dz 2π −∞
dg
Hence the Fourier space transform of is (+i f ) g(k).
dz
Applying these results to a function of both t and z, where
 ∞ ∞
f (k, ω) = f (z, t) e−i(kz−ωt) dz dt (3a)
−∞ −∞

and
 ∞  ∞
1
f (z, t) = f (k, ω) e+i(kz−ωt) dk dω, (3b)
(2π)2 −∞ −∞

∂
the double Fourier transform of f (z, t) is (−iω) f (k, ω); and the double Fourier
∂t
∂2
transform of f (z, t) is (ik)2 f (k, ω) = −k 2 f (k, ω). These are the results we need
∂z 2
to interpret the differentiated terms in equation (6.22) in the (k, ω) domain.
Note that different sign conventions for the Fourier transforms (1) and (2) may give
slightly different results (for example, multiplication by +iω instead of −iω). Throughout
these notes we are using the mixed convention given in (1) and (2) and discussed in Box 3.1.

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 6.1 Heat Flow by Conduction/Diffusion: an Example of the Diffusion Equation 115

and integrate over all real values of z and t. The reason to do this, is that partial
differentiation (with respect to t and z) becomes scalar multiplication (by −iωt and
+ikz) in the transform domain, as explained in Box 6.2, leading to an equation for
T = T (k, ω) that is easy to solve as a specific algebraic expression. Specifically, the
three terms in (6.22) transform as follows:
 ∞ ∞ 
1 ∂T (−iω)
e−i(kz−ωt) dz dt = T (k, ω);
−∞ −∞ κ ∂t κ
 ∞ ∞ 2 
∂ T
e−i(kz−ωt) dz dt = (−k 2)T (k, ω);
−∞ −∞ ∂z 2

and
 ∞  ∞  
C C
δ(z) δ(t) e−i(kz−ωt) dz dt = .
−∞ −∞ K K

This last result is obtained from the “substitution” property of the Dirac delta function,
illustrated earlier by (3.22) and equation (9) of Box 6.1. More fundamentally, it can
A
be obtained directly from (6.19) and (6.20) applied to in (6.7).
K
(ii) Using the three results listed above, equation (6.22) has now been reduced to

−iω C
T (k, ω) = −k 2 T (k, ω) + . (6.23)
κ K
This is an algebraic equation for the double transform of T , easily solved to obtain

iκC 1
T (k, ω) = . (6.24)
K ω + iκk 2
(iii) In order to obtain T (z, t) from T (k, ω) we must evaluate the two inverse transform
integrals. We shall go first from ω to t, then from k to z. Thus
 ∞
1 iκC e−iωt
T (k, t) = dω. (6.25)
2π K −∞ ω + iκk 2

The integrand in (6.25) has a simple pole at ω = −iκk 2. The quickest way to evaluate
this integral explicitly, is to use the methods given in Chapter 5. Thus, although the
integration in (6.25) is over purely real values of ω, we shall allow ω to become a
complex variable and use the result that

f (ω) 0 if ω0 is outside the circuit C, and

dω = (6.26)
C ω − ω 0 2πi f (ω0 ) if ω0 lies inside the circuit.

[This is a basic result in the calculus of residues. C is a closed circuit in the complex
ω-plane, taken in the counterclockwise direction. See (5.17.)]
First, assume t < 0. To the real axis path of integration in (6.25), we add the
upper semicircle as shown on the left in Figure 6.7. (The semicircle has arbitrarily
large radius). Since e−iωt = e[ω]t , the integrand of (6.25) is exponentially small in

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 116 Chapter 6 / THE DIFFUSION EQUATION

Imaginary ω Imaginary ω

C1

The circuit needed when t > 0

Real ω
Real ω
Pole at ω = −i κ k 2
Pole at ω = −i κ k 2

C2
The circuit needed when t < 0

FIGURE 6.7
Closed circuits in the complex ω-plane, used for evaluating (6.25). For t < 0 we use the path C1
shown on the left, on which e−iωt is exponentially small. For t > 0 we use the path C2 shown on the
right. Note that we adopt the convention of taking circuit integrals in the counterclockwise direction.

the upper half of the complex ω plane, where [ω] > 0, if t < 0. It follows for a large
enough semicircle that
 ∞
e−iωt e−iωt
dω = dω = 0
−∞ ω + iκk 2 C1 ω + iκk 2

because the pole lies outside the circuit. Hence T (k, t) = 0 for t < 0 — confirming
a result we expect.
Second, assume t > 0. To complete the circuit, we now must add a return path
in the lower half plane (because in this case, with t > 0, e−iωt is exponentially small
if [ω] < 0). See the right-hand side of Figure 6.7. So, from (6.26)
 ∞
e−iωt e−iωt
dω = −2πi e−κk t
2
dω = −
−∞ ω + iκk 2 C2 ω + iκk 2

(minus signs arise because the circuit C2 is by convention taken counterclockwise,

and we must reverse this direction to get the integral from negative to positive values
along the real ω axis).
At this point we have an explicit solution for the first inverse transform, namely

Cκ −κk 2t
T (k, t) = e H (t) (6.27)
K
where H (t) is the Heaviside function, sometimes called the unit step function
(H (t) = 0 for t < 0, and = 1 for t > 0).
Finally, we must invert from k to z. We consider only t > 0, and our integral is
 ∞
Cκ
e−κtk eikz dk.
2
T (z, t) = (6.28)
2π K −∞

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 6.1 Heat Flow by Conduction/Diffusion: an Example of the Diffusion Equation 117

BOX 6.3
Equivalence of two paths of integration

To verify the result given in (6.29), we need to show that

 iz
∞− 2κt  ∞
e−κtl dl = e−κtl dl.
2 2
(1)
iz
−∞− 2κt −∞

The path of integration of the left-hand side of (1) is not the real l-axis, but a line below
it. (The line extends all the way from infinity on the left, through the indicated point in the
−i z
complex l-plane given by l = , then all the way to infinity on the right.)
2κt

Imaginary l- axis

Real l- axis

−iz
l = ___
2κt

But instead of the path of integration shown above, we can take the following path:

Imaginary l- axis

Real l- axis

−iz
l = ___
2κt

Since the integrand e−κtl is analytic in the region between the paths shown in these
2

two Figures, we can apply (5.7) and equate the result of integrating along the path shown
in the upper Figure, to the result using the path shown in the lower Figure. But along the
−i z
vertical sections, from l = −∞ − to l = −∞ on the real l-axis, and then from the real
2κt
−i z
l-axis again at l = ∞ to l = ∞ − , there is neglible contribution because the integrand
2κt
is arbitrarily small. Throwing away these two segments, it follows that (1) is valid.
This result, like many applications of the properties of complex functions of a complex
variable including our discussion from (6.25) to (6.30), takes a whole lot longer to
demonstrate the first time you see it, than to use in practice.

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 118 Chapter 6 / THE DIFFUSION EQUATION

We can force this integrand to look like something we can integrate if we make the
substitution

z2 iz
−κtk 2 + ikz = −κtl 2 − where l = k − ,
4κt 2κt

z2
for then the term involving is a constant with respect to the integration, and we
4κt
have something like

 ∞
−z 2  −z 2 
−κtk 2 +ikz π
e−κtl dl = e 4κt
2
e dk = e 4κt (6.29)
−∞ κt

(the last equality here uses a result given in Box 3.2). The only problem with (6.29)
is with the limits of integration. For the first equality of (6.29) to be true, the limits
iz iz
of the integral over l are from −∞ − to +∞ − . For the second equality
2κt 2κt
of (6.29) to be true, the limits of the integral over l need to be from −∞ to +∞. Does
it matter, that these limits of integration are different?
The answer is no, for reasons detailed in Box 6.3.
Putting the results (6.27)–(6.29) together, we obtain at last the result we seek in
the space-time domain:

z2
−
Cκ e 4κt
T (z, t) = √ H (t). (6.30)
2K π κt

Having executed all of the steps (i), (ii), and (iii), we next comment briefly on two
properties of the solution, (6.30).
First, we should expect that heat is conserved in vertical columns for all times t > 0.
That is, we expect
 ∞
ρcT (z, t) dz = C. (6.31)
−∞

Indeed the solution (6.30) has this property, for

 ∞
ρc Cκ z2 ρc Cκ √
Left-hand side of (6.31) = √ e− 4κt dz = √ 4κt π
2K π κt −∞ 2K π κt
=C (from Box 3.2 and K = ρcκ)
= Right-hand side of (6.31).

Second, in Figure 6.8 we show the main space-time properties of the solution. Note that
at any fixed value of z, the temperature begins to rise instantly as soon as t becomes positive.
In this sense, the propagation speed for conduction is infinite. However, in practical terms
we don’t see this effect because the temperature remains infinitesimal until t has become
z2
large enough for − to become small, thus avoiding exponentially small values of T .
4κt

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 6.1 Heat Flow by Conduction/Diffusion: an Example of the Diffusion Equation 119

-4
Dyke injection.
Hot molten rock is injected
256 hours at depth z = 0
d -3
e at time t = 0.
p The subsequent temperature profile
t is shown at five different times.
h
-2 After 1 hour, the temperature change
is still confined to less than – 0.5 meter.
i 64 hours After 256 hours, the temperature change
n is still confined to the range – 3 meters.
-1
m 16 hours
e 4 hours
1 hour
t
e
r 0
s
2
—z
temperature = constant x e 4 κ t H(t)
1
πκt

2
κ = 0.000001 meter /second
2
z in meters

t in seconds
3
At fixed time, the temperature as
a function of z has an inflection
given by z = 2 κ t . This is
another example of the general rule,
2
for diffusion, that time scales with length .
4
0 2 4 6 8 10

temperature/constant
FIGURE 6.8
Values are shown, of the temperature solution (6.30) for the problem of dyke injection. The constant
referred to in this Figure is Cκ/(2K ).

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 120 Chapter 6 / THE DIFFUSION EQUATION

At any fixed time t, the temperature has a gaussian distribution as a function of depth z.
The width of the gaussian increases for increasing values of t, and the height of the gaussian
correspondingly decreases, thus preserving the total heat (the area under each gaussian), as
indicated by (6.31). As noted in Figure 6.8 and repeatedly throughout this chapter, time in
the diffusion solution scales in proportion to the square of the distance.

6.1.4 RADIATION TEMPERATURE AT THE SURFACE OF A SPHERICAL CONDUCTING

SOLID WITH INTERNAL HEATING

Our fourth and final discussion of a particular solution to the 3D diffusion equation, (6.7),
concerns the temperature distribution throughout a homogeneous spherical solid with a
constant internal source of heat. The discussion below is based on Menke and Abbott
(1990), and is a very simple model of a planet or planetesimal in space.
Thus, we consider a spherical planetary object with radius, say, 6371 km (equal to r⊕,
the radius of the Earth). But we are using only a very crude model, which everywhere has the
same conductivity (K ), mass density (ρ), specific heat (c), and strength of internal heating
(A). Furthermore, we assume the temperature has stabilized, is spherically symmetric, and
is no longer time dependent. In this case, the diffusion equation
1 ∂T A
= ∇ 2T + ((6.7) again)
κ ∂t K
reduces to
 
1 d 2 dT
K 2 r = −A (6.32)
r dr dr

because the general dependence of T on space and time is replaced by T = T (r ). (We have
interpreted the spatial derivatives in the Laplacian as given for spherical polars in (4.3), in
the especially simple case of dependence only on the radial coordinate, r .)
Equation (6.32) is easily integrated, first to give
dT 1
Kr2 = − Ar 3 + B
dr 3
for some constant B. After dividing through by r 2 we can integrate a second time, obtaining
1 B
K T = − Ar 2 − + C
6 r
where C is a second integration constant.
We require B = 0 since there nothing special about the temperature at the center of
the sphere which would make T singular there. And we can express C in terms of the
temperature at r = r⊕, finding that
1
T= 2
A(r⊕ − r 2) + T (r⊕). (6.33)
6K
It is known that the radiation into outer space is proportional to the fourth power of
T (r⊕) (expressed in degrees Kelvin, ◦K). The constant of proportionality for this relationship

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58
 Problems 121

is called the Stefan-Boltzmann constant, σ (= 5.7 × 10−8 W m−2 ◦K−4). The heat flux, q,
which in spherical polars has components (q(r ), 0, 0) with scalar q, is given in terms of the
dT
vertical temperature gradient by q = −K . So from (6.33) we find that
dr

A
q = r⊕ = σ T (r⊕)4.
r =r⊕ 3

The last equality here allows us to obtain the surface temperature in terms of the strength
of the internal heat source, A, and the planetary radius. The result
 1
Ar⊕ 4
T (r⊕) = . (6.34)
3σ

Even if we use a high value for A (say, 10 microwatts per cubic meter), and assume that it
applies throughout the Earth), (6.34) gives a temperature of only about 140 degrees above
absolute zero. This basically is why planetary surfaces are so cold — as long as they are
not exposed to solar radiation and have no atmosphere. Our own temperate planet would
be in bad shape if life had to rely upon internal sources of heat.

Suggestions for Further Reading

Menke, William, and Dallas Abbott. Geophysical Theory, New York: Columbia University
Press, 1990 (pp 204–217).
Carslaw, Horatio S., and J. C. Jaeger. Conduction of Heat in Solids, Oxford University
Press. Second edition reprinted, 1973. Second edition in paperback, 1986.

Problems

6.1 Prove the equality (7) of Box 6.1. That is, show that the right-hand side of this
equation has the defining qualities of the 3D Dirac delta function.
6.2 Given that the total heat flow out of the Earth’s interior is about 44.2 terawatts,
what is the average heat flux through the surface of the Earth? Show that this
value has the expected properties of (1) lying between the average continental
value (65 mW m−2) and the average oceanic value (101 mW m−2), and (2) being
somewhat closer to the average oceanic value (oceans have greater surface area
than continents).
6.3 Work through some relevant values of A and the value of σ (Stefan-Boltzmann)
to see what the surface temperature given by (6.34) actually is.

working pages for Paul Richards’ class notes; do not copy or circulate without permission from PGR 2004/11/11 19:58