The diffusion equation

20.5 THE DIFFUSION EQUATION
term is a little less obvious. It can be viewed as representing the accumulated
transverse displacement at position x due to the passage past x of all parts of
the initial motion whose effects can reach x within a time t, both backward and
forward travelling.
The extension to the three-dimensional wave equation of solutions of the type
we have so far encountered presents no serious difficulty. In Cartesian coordinates
the three-dimensional wave equation is
∂2 u ∂2 u
1 ∂2 u
∂2 u
+ 2 + 2 − 2 2 = 0.
2
∂x
∂y
∂z
c ∂t
(20.32)
In close analogy with the one-dimensional case we try solutions that are functions
of linear combinations of all four variables,
p = lx + my + nz + µt.
It is clear that a solution u(x, y, z, t) = f(p) will be acceptable provided that
µ2 d2 f(p)
= 0.
l 2 + m2 + n2 − 2
c
dp2
Thus, as in the one-dimensional case, f can be arbitrary provided that
l 2 + m2 + n2 = µ2 /c2 .
Using an obvious normalisation, we take µ = ±c and l, m, n as three numbers
such that
l 2 + m2 + n2 = 1.
In other words (l, m, n) are the Cartesian components of a unit vector n̂ that
points along the direction of propagation of the wave. The quantity p can be
written in terms of vectors as the scalar expression p = n̂ · r ± ct, and the general
solution of (20.32) is then
u(x, y, z, t) = u(r, t) = f(n̂ · r − ct) + g(n̂ · r + ct),
(20.33)
where n̂ is any unit vector. It would perhaps be more transparent to write n̂
explicitly as one of the arguments of u.
20.5 The diffusion equation
One important class of second-order PDEs, which we have not yet considered
in detail, is that in which the second derivative with respect to one variable
appears, but only the first derivative with respect to another (usually time). This
is exemplified by the one-dimensional diffusion equation
κ
∂u
∂2 u(x, t)
,
=
∂x2
∂t
695
(20.34)
PDES: GENERAL AND PARTICULAR SOLUTIONS
in which κ is a constant with the dimensions length2 × time−1 . The physical
constants that go to make up κ in a particular case depend upon the nature of
the process (e.g. solute diffusion, heat flow, etc.) and the material being described.
With (20.34) we cannot hope to repeat successfully the method of subsection
20.3.3, since now u(x, t) is differentiated a different number of times on the two
sides of the equation; any attempted solution in the form u(x, t) = f(p) with
p = ax + bt will lead only to an equation in which the form of f cannot be
cancelled out. Clearly we must try other methods.
Solutions may be obtained by using the standard method of separation of
variables discussed in the next chapter. Alternatively, a simple solution is also
given if both sides of (20.34), as it stands, are separately set equal to a constant
α (say), so that
α
∂2 u
= ,
∂x2
κ
∂u
= α.
∂t
These equations have the general solutions
u(x, t) =
α 2
x + xg(t) + h(t)
2κ
and u(x, t) = αt + m(x)
respectively and may be made compatible with each other if g(t) is taken as
constant, g(t) = g (where g could be zero), h(t) = αt and m(x) = (α/2κ)x2 + gx.
An acceptable solution is thus
u(x, t) =
α 2
x + gx + αt + constant.
2κ
(20.35)
Let us now return to seeking solutions of equations by combining the independent variables in particular ways. Having seen that a linear combination of
x and t will be of no value, we must search for other possible combinations. It
has been noted already that κ has the dimensions length2 × time−1 and so the
combination of variables
η=
x2
κt
will be dimensionless. Let us see if we can satisfy (20.34) with a solution of the
form u(x, t) = f(η). Evaluating the necessary derivatives we have
df(η) ∂η
2x df(η)
∂u
=
=
,
∂x
dη ∂x
κt dη
2 2
2x
∂2 u
d f(η)
2 df(η)
+
=
,
2
∂x
κt dη
κt
dη 2
x2 df(η)
∂u
=− 2
.
∂t
κt dη
Substituting these expressions into (20.34) we find that the new equation can be
696
20.5 THE DIFFUSION EQUATION
written entirely in terms of η,
4η
d2 f(η)
df(η)
= 0.
+ (2 + η)
dη 2
dη
This is a straightforward ODE, which can be solved as follows. Writing f (η) =
df(η)/dη, etc., we have
⇒
f (η)
1
1
=−
−
f (η)
2η 4
η
ln[η 1/2 f (η)] = − + c
4
−η A
⇒
f (η) = 1/2 exp
4
η
η
−µ dµ.
⇒
f(η) = A
µ−1/2 exp
4
η0
If we now write this in terms of a slightly different variable
ζ=
η 1/2
x
,
=
2
2(κt)1/2
then dζ = 14 η −1/2 dη, and the solution to (20.34) is given by
ζ
exp(−ν 2 ) dν.
u(x, t) = f(η) = g(ζ) = B
(20.36)
ζ0
Here B is a constant and it should be noticed that x and t appear on the RHS
only in the indefinite upper limit ζ, and then only in the combination xt−1/2 . If
ζ0 is chosen as zero then u(x, t) is, to within a constant factor,§ the error function
erf[x/2(κt)1/2 ], which is tabulated in many reference books. Only non-negative
values of x and t are to be considered here, so that ζ ≥ ζ0 .
Let us try to determine what kind of (say) temperature distribution and flow
this represents. For definiteness we take ζ0 = 0. Firstly, since u(x, t) in (20.36)
depends only upon the product xt−1/2 , it is clear that all points x at times t such
that xt−1/2 has the same value have the same temperature. Put another way, at
any specific time t the region having a particular temperature has moved along
the positive x-axis a distance proportional to the square root of t. This is a typical
diffusion process.
Notice that, on the one hand, at t = 0 the variable ζ → ∞ and u becomes
quite independent of x (except perhaps at x = 0); the solution then represents a
uniform spatial temperature distribution. On the other hand, at x = 0 we have
that u(x, t) is identically zero for all t.
§
Take B = 2π −1/2 to give the usual error function normalised in such a way that erf(∞) = 1. See
the Appendix.
697
PDES: GENERAL AND PARTICULAR SOLUTIONS
An infrared laser delivers a pulse of (heat) energy E to a point P on a large insulated
sheet of thickness b, thermal conductivity k, specific heat s and density ρ. The sheet is
initially at a uniform temperature. If u(r, t) is the excess temperature a time t later, at a
point that is a distance r ( b) from P , then show that a suitable expression for u is
α
r2
,
(20.37)
u(r, t) = exp −
t
2βt
where α and β are constants. (Note that we use r instead of ρ to denote the radial coordinate
in plane polars so as to avoid confusion with the density.)
Further, (i) show that β = 2k/(sρ); (ii) demonstrate that the excess heat energy in the
sheet is independent of t, and hence evaluate α; and (iii) prove that the total heat flow past
any circle of radius r is E.
The equation to be solved is the heat diffusion equation
k∇2 u(r, t) = sρ
∂u(r, t)
.
∂t
Since we only require the solution for r b we can treat the problem as two-dimensional
with obvious circular symmetry. Thus only the r-derivative term in the expression for ∇2 u
is non-zero, giving
∂u
∂u
k ∂
r
= sρ ,
(20.38)
r ∂r
∂r
∂t
where now u(r, t) = u(r, t).
(i) Substituting the given expression (20.37) into (20.38) we obtain
2
2
sρα
r2
r2
r
r
2kα
= 2
,
− 1 exp −
− 1 exp −
2
βt
2βt
2βt
t
2βt
2βt
from which we find that (20.37) is a solution, provided β = 2k/(sρ).
(ii) The excess heat in the system at any time t is
∞
∞
r2
r
bρs
dr
u(r, t)2πr dr = 2πbρsα
exp −
t
2βt
0
0
= 2πbρsαβ.
The excess heat is therefore independent of t and so must be equal to the total heat
input E, implying that
α=
E
E
=
.
2πbρsβ
4πbk
(iii) The total heat flow past a circle of radius r is
∞
∞
r2
∂u(r, t)
E
−r
exp −
dt
−2πrbk
dt = −2πrbk
∂r
4πbkt βt
2βt
0
0
∞
r2
= E exp −
=E
for all r.
2βt 0
As we would expect, all the heat energy E deposited by the laser will eventually flow past
a circle of any given radius r. 698