Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

The One Dimensional Heat Equation

Kurt Bryan
1 Informal Derivation in One Dimension
Consider a one-dimensional bar lying along the x axis from x = 0 to
x = 1. Let t denote time and u(t, x) the temperature of the bar at time t
and position x. We want to nd an equation that governs the temperature
u in the interior of the bar. In what follows well use the notation u
x
for
u
x
,
u
xx
=

2
u
x
2
, etc.
Well think of u as a measurement of the density of thermal energy
in the bar. Specically, consider a portion of the bar stretching from x to
x +x, where x is small; well refer to this part of the bar as the control
volume. Well assume that the amount of thermal energy in the control
volume is given approximately by cu(t, x)x where c is some constant. Note
that cu(t, x)x has dimensions of energy. The constant c is related to the
specic heat of the material out of which the bar is made.
Its intuitively clear that heat energy ows from hotter regions to colder
regions. Well assume that the rate at which energy ows through or past a
point x from left to right (increasing x) in the bar is given by (x)u
x
(t, x),
where (x) > 0 is some function, called the thermal conductivity of the bar
at point x. If the bar is homogeneous (most typical) then (x) is actually
a constant. The quantity (x)u
x
(t, x) has dimension energy per time, and
the ow of thermal energy is downhill, from hot to cold, since (x) < 0.
Well require that (x) be twice continuously dierentiable for now.
Now well invoke conservation of energy: The rate at which energy ows
INTO the left end of the control volume is given by (x)u
x
(t, x) (make sure
you believe the sign is correct). The rate at which energy ows INTO the
right end of the control volume is given by (x + x)u
x
(t, x + x) (again,
make sure you believe the sign is correct). All in all thermal energy is owing
into the control volume at a rate of
rate in = (x + x) (u
x
(t, x + x) (x)u
x
(t, x)) . (1)
But by conservation of energy this means that the thermal energy in the
control volume is increasing at this rate. The rate at which the total energy
of the control volume increases is given by
rate of increase cu
t
(t, x)x (2)
1
If we equate the right sides of equations (1) and (2), divide by x, and let
x approach zero we nd that u(t, x) should satisfy
u
t
((x)u
x
)
x
= 0, (3)
the so-called heat equation, where (x) =
(x)
c
is called the thermal diusivity
of the bar. For simplicity well often take = 1.
The heat equation requires boundary and initial conditions in order to
possess a unique solution. One typically species and initial conditions like
u(0, x) = u
0
(x) for some specied initial temperature u
0
(x). For the bound-
ary conditions at the endpoints one possibility is to specify u(t, 0) = f
0
(t)
and u(t, 1) = f
1
(t) (so the endpoint temperatures are specied, so-called
Dirichlet boundary conditions). Or one can specify the rate at which heat
energy is entering the ends of the bar. This leads to boundary conditions
(0)u
x
(t, 0) = g
0
(t) and (1)u
x
(t, 1) = g
1
(t) (these are the Neumann
boundary conditions). If g
i
(t) 0 for i = 0 or i = 1 then we have an
insulating boundary condition at that end of the bar.
One can show that under reasonable assumptions on the initial temper-
ature u
0
and boundary conditions f
0
and f
1
, or g
0
and g
1
, the heat equation
has a unique solution.
2 Modeling Defects: The Forward Problem
Consider a one-dimensional bar stretching from x = 0 to x = 1. For sim-
plicity lets take the constants above to be one, so = c = = 1. With
u(t, x) denoting the temperature in the bar we should have u
t
u
xx
= 0 for
0 < x < 1 and t > 0, with appropriate boundary and initial conditions.
But suppose that the bar contains an interior point defect, that is, some
point x = with 0 < < 1 in the interior of the bar through which the
ow of heat is somehow impeded. For example, at x = there might be a
break in the bar which blocks the ow of heat. At that point wed have
u
x
(t, ) = 0 for all t. But more generally, we can consider a defect which
allows heat to ow past x = , but with some resistance.
To quantify this, lets dene notation
u

(t) = lim
x

u(t, x), u
+
(t) = lim
x
+
u(t, x),
u

x
(t) = lim
x

u
x
(t, x), u
+
x
(t) = lim
x
+
u
x
(t, x),
2
the limiting values of u and u
x
from the left and right sides of the defect at
any given time. Also let [u](t) = u
+
(t) u

(t), the jump in u across the


defect.
Suppose that at x = 0 and x = 1 we input heat ux (or heat energy) g
0
(t)
and g
1
(t), respectively, and the bar starts with initial temperature u
0
(x). One
model for an interior defect is that the temperature u(t, x) should obey the
usual heat equation AWAY from the defect, so that
u
t
u
xx
= 0, x (0, ) (, 1). (4)
We also require the boundary and initial conditions
u
x
(t, 0) = g
0
(t) (5)
u
x
(t, 1) = g
1
(t) (6)
u(0, x) = u
0
(x), x (0, ) (, 1). (7)
We model the heat impeding behavior of the defect at x = as
u

x
(t, ) = u
+
x
(t, ) = F([u]). (8)
The function F depends on the physical nature of the defect, but in any case
it should satisfy the conditions that F(0) = 0, F is strictly increasing and
(at least) continuous. It is possible to show that under these conditions (if,
for example, g
0
, g
1
, and u
0
are continuous) then there is a unique solution to
equations (4)-(8). Note that equation (8) really makes two assertions, that
u

x
(t, ) = u
+
x
(t, ) and that both of these quantities equal F([u]).
In order to motivate this model for how heat ows past a defect, well
rst look more closely at the steady-state version of the problem.
3 The Steady-State Problem in One Dimen-
sion
Consider the steady-state model for the heat conduction problem, in which
the temperature doesnt depend on time, so u = u(x) with no dependence
on t. In this case the heat equation becomes just ((x)u
x
)
x
= 0. The
Neumann boundary conditions become (0)u
x
(0) = g
0
and (1)u
x
(1) = g
1
for some constants g
0
and g
1
. Now ((x)u
x
)
x
= 0 implies that (x)u
x
= m
for some constant m, and in fact from the boundary condition you see that
3
m = g
0
= g
1
, so g
0
= g
1
is a necessary condition for solving the steady-
state version of the problem.
Also, since u
x
= m/(x) this means that u must be of the form
u(x) = m

x
0
dy
(y)
+ b
for some constant of integration b, and in fact b can by ANYTHINGtheres
no unique solution to the problem. Its conventional to specify an additional
condition like u(0) = 0, to nail down one particular solution. Physically,
this is equivalent to choosing the (arbitrary) zero point for our temperature
scalewe just dene the left end of the bar to be at zero degrees.
In the special case that the bar has constant thermal diusivity equal to
1, the heat equation just becomes u

(x) = 0, since u
t
0. The boundary
conditions become u

(0) = g
0
and u

(1) = g
1
, where g
0
and g
1
are con-
stants. The initial condition becomes irrelevant. Its then easy to see that
u(x) = mx + b for constants m and b, and of course u

(x) = m. Again, this


immediately shows that we must have g
0
= g
1
or else no solution is possi-
ble. If this condition holds, there are innitely many solutions, for anything
of the form u(x) = g
1
x + b satises the PDE boundary conditions, for any
choice of b. Again, we nail down a unique solution by specifying an additional
condition, e.g., u(0) = 0, which forces b = 0 here.
4 Motivation for the Defect Model
Now consider a bar of length one, 0 < x < 1, in which the thermal diusivity
is of the form
(x) =

1, x

1
(x), < x < +
1, + x < 1
where > 0 is close to zero. The motivation here is this: the region of the
bar from x = to x = + is corroded, and so has a low thermal diusivity
(doesnt conduct heat very well). In this region the diusivity
1
(x) is close
to zero. But away from this region the diusivity is some xed constant,
which Ive chosen to be 1.
Lets consider the steady-state version of the heat conduction problem
in this bar, in which we put in a ux g at the x = 0 end (so u

(0) = g),
pull the heat out at x = 1 (so u

(1) = g), and set u(0) = 0. As discussed


4
above, from the steady-state heat equation ((x)u
x
)
x
= 0 we have, from one
integration,
u

(x) =
g
(x)
, (9)
and a second integral gives
u(x) = u(0)

x
0
g
(y)
dy. (10)
Its easy to see from equation (9) that away from the defect region we
have u

(x) = g. But within the defect region we have u

(x) = g/(x),
which is of much greater magnitude than g, since (x) > 0 is close to zero,
except maybe at the edges near x = and x = +. The solution will look
something like this:
In fact, if we assume that (x)
0
(a constant) in the defect region, then
u

(x) =
g

0
in this region. As a result, the solution will rise a distance of
g

0
in the short space from x = to x = + . Note that u

= g,
however, just outside either boundary of the defect.
The inverse problem would be to recover the defect, dened by its left
boundary , right boundary + , and diusivity prole
1
(x). But in fact,
we probably wouldnt be interested in the precise prole of the corrosion
diusivity, but rather its rough location () and overall magnitude. Thus
instead of modelling the defect as above, we could consider it to be a point
within the bar, at x = . Instead of worrying about the diusivity of the
defect, we simply capture its behavior with the requirement that u

(x) have
the same limiting value on both side of , and that u have an appropriate
5
jump [u] over that obeys u

() =

0
[u] (roughly equivalent to a rise of
g

0
over a run of ). Since were not interested in
0
or individually, we
just jump them into a single parameter k and use u

() = k[u]. If we want
to model a defect which transmits nonlinearly, we can use u

() = F([u]),
though its not clear under what circumstance this is physically reasonable.
In the full time-dependent problem we simply use u
x
(, t) = F([u]) for
the defect jump condition.
6

You might also like