1. Modeling of Heat Flow.

Heat or Diffusion Equation

Physical experiments show that in a body, heat flows in the direction of decreasing temper-
ature, and the rate of flow is proportional to the gradient of the temperature. This means
that the velocity v of the heat flow in a body is of the form

⃗v = −K∇U, (1.1)

where is U (x, y, z, t) temperature, t is time, and K is called the thermal conductivity of the
body; in ordinary physical circumstances K is a constant. Using this information, we set up
the mathematical model of heat flow, the so-called heat equation or diffusion equation.
Solution. Let V be a region in the body bounded by a surface S with outer unit normal
vector n̂ such that the divergence theorem applies. Then ⃗v · n̂ is the component of ⃗v in the
direction of n̂, and the amount of heat leaving V per unit time is
ZZ
⃗v · n̂ dA.
S

Using
∇ · (∇U ) = ∇2 U = ∆U = Uxx + Uyy + Uzz ,
(where ∇2 or ∆ is the the Laplacian operator), we have by the divergence theorem and (1)
ZZ ZZZ ZZZ
⃗v · n̂ dA = −K ∇ · (∇U )dV = −K ∇2 U dV. (1.2)
S V V

On the other hand, the total amount of heat H in V is

ZZZ
H= σρU dV
V

where the constant σ is the specific heat of the material of the body and ρ is the density of
the material. Hence the time rate of decrease of H is
ZZZ
∂H ∂H
U
− =− σρ dV
∂t V ∂t
and this must be equal to the above amount of heat leaving V. This gives us
ZZZ ZZZ
∂U
− σρ dV = −K ∇2 U dV.
V ∂t V

We rewrite this as ZZZ  

∂U 2
− σρ dV − K∇ U dV = 0.
V ∂t
Since this holds for any region V in the body, the integrand (if continuous) must be zero
everywhere; that is,
∂U
= c2 ∇ 2 U (1.3)
∂t
where c2 = K/(σρ) is called the thermal diffusivity of the material. This partial differential
equation is called the heat equation. It is the fundamental equation for heat conduction.

1
And our derivation is another impressive demonstration of the great importance of the
divergence theorem. The heat equation is also called the diffusion equation because it also
models diffusion processes of motions of molecules tending to level off differences in density
or pressure in gases or liquids.
If heat flow does not depend on time, it is called steady-state heat flow. Then ∂U/∂t = 0
and hence ∇2 U = 0, so that (1.3) reduces to Laplace’s equation.
3. Heat Conduction in Different Domains
In this chapter we discuss the methods to solve the heat equation (or diffusion equation)
in different domains and initial and boundary conditions. We first consider the pure initial
value problem (with no boundaries),

ut (x, t) = c2 uxx (x, t), −∞ < x < ∞, t > 0, u(x, 0) = f (x), −∞ < x < ∞.

We assume that f (x) satisfies the assumptions for its Fourier integral to exist. We take
u(x, t) = X(x)T (t) and obtain
X ′′ T′
= 2 = λ.
X cT
′ 2
For T with get T = λc T. Since the temperature decreases with time, we must have λ =
2 2
−p2 , where p > 0. This gives us T (t) = T0 e−c p t , where T0 is a constant. For X we get
X ′′ + p2 X = 0 which gives X(x) = A cos px + B sin px. Thus,
2 p2 t
u(x, t; p) = (A cos px + B sin px)e−c ,

where we have merged T0 in A and B. Since p > 0 is a continuous variable, we take

Z ∞ Z ∞
2 2
u(x, t) = u(x, t; p)dp = [A(p) cos px + B(p) sin px]e−c p t dp. (3.1)
0 0

To determine A(p) and B(p) we use u(x, 0) = f (x) to get

Z ∞ Z ∞
f (x) = u(x, 0; p)dp = [A(p) cos px + B(p) sin px] dp.
0 0

Hence Z ∞ Z ∞
1 1
A(p) = f (y) cos py dy, B(p) = f (y) sin py dy. (3.2)
π −∞ π −∞

Using (2) in (1), we get

1 ∞
Z Z ∞ 
−c2 p2 t
u(x, t) = f (y) cos p(x − y)e dy dp
π 0 −∞

1 ∞
Z Z ∞ 
−c2 p2 t
= f (y) cos p(x − y)e dy dp.
π −∞ 0
We use the following formula,
Z ∞ √
−s2 π −b2
e cos 2bs ds = e , (3.3)
0 2

2
 √ √
and by putting s = cp t and p(x − y) = 2bs = 2bcp t which gives
x−y
b= √ ,
2c t
we get
∞
(x − y)2
Z  
1
u(x, t) = √ f (y) exp − dy. (3.4)
2c πt −∞ 4c2 t
If we put  
(x − y)
− = z,
4c2 t
then
1
Z ∞ √ 2
u(x, t) = √ f (x + 2cz t)e−z dz. (3.5)
π −∞

Example 1 Let f (x) = U0 , for |x| < 1 and f (x) = 0 for |x| > 1, where U0 is a constant.

Then
1
(x − y)2
Z  
U0
u(x, t) = p √ exp − dy. (3.6)
2c πt −1 4c2 t
We may write (3.6) as √
Z (1−x)/2c t
U0 2
u(x, t) = √ √
e−z dz. (3.7)
π −(1+x)/2c t

The integral (3.7) can be expressed in terms of the error function erf (x),
Z x
2 2
erf (x) = √ e−z dz. (3.8)
π 0

whose values are tabulated.

3
4. Heat Conduction in a Semi-infinite Bar
If we have the following problem of heat conduction in a semi-infinite bar,

ut (x, t) = c2 uxx (x, t), 0 < x < ∞, t > 0

u(x, 0) = f (x), 0 < x < ∞, u(0, t) = 0, t ≥ 0,

then we extend f on (−∞, 0) oddly by defining

o f (x), x > 0,
f (x) =
−f (−x), x < 0,

and consider
ut (x, t) = c2 uxx (x, t), 0 < x < ∞, t > 0
0
u(x, 0) = f (x), 0 < x < ∞.
Using (3.5) we get
1
Z ∞ √ 2
u(x, t) = √ f 0 (x + 2cz t)e−z dz. (4.1)
π −∞
√
We put 2c t = τ > 0, and observe that for 0 < x + τ z < ∞, we have −x/τ < z < ∞ and
for −∞ < x + τ z < 0, we have −∞ < z < −x/τ. Hence
Z ∞ Z −x/τ
1 −z 2 1 2
u(x, t) = √ f (x + τ z)e dz − √ f (−x − τ z)e−z dz
π −x/τ π −∞
Z ∞ Z ∞
1 −z 2 1 2
=√ f (x + τ z)e dz − √ f (−x + τ z)e−z dz
π −x/τ π x/τ

Now, we consider the heat conduction in a bar of finite length L,

ut (x, t) = c2 uxx (x, t), 0 < x < L,

4
 u(x, 0) = f (x), 0 < x < L, u(0, t) = u(L, t) = 0, t ≥ 0.
By taking u(x, t) = X(x)T (t), we get

X ′′ T′
= 2 = λ.
X cT
For X we get the SLP,

X ′′ = λX, 0 < x < L, X(0) = X(L) = 0.

For this SLP to have non-zero solutions we know that λ = −µ2 where µ > 0. We get the
solutions as X(x) = A cos µx + B sin µx. The condition X(0) = 0 gives A = 0 and X(L) = 0
inmplies that µ = nπ/L. Hence λn = −n2 π 2 /L2 and Xn (x) = sin(nπx/L). The second SLP
2 2 2 2
T ′ + (n2 π 2 c2 /L2 )T = 0 gives Tn (t) = T0n e−c n π t/L . Hence
∞
2 n2 π 2 t/L2
X
u(x, t) = Bn e−c sin(xπx/L).
n=1

To determine Bn we use u(x, 0) = f (x). Hence

∞ Z L
X xπx 2 nπx
f (x) = Bn sin =⇒ Bn = f (x) sin dx.
n=1
L L 0 L

Example 2 let 
x, 0 < x < L/2,
f (x) =
L − x, L/2 < x < L.
We have !
Z l/2 Z L
2
Bn = x sin(nπx/L) dx + (L − x) sin(nπx/L) dx .
L 0 l/2

It follows that Bn = 0 if n is even and

Bn = 4L/(n2 π 2 ), n = 1, 5, 9, . . . , Bn = −4L/(n2 π 2 ), n = 3, 7, 11, . . . .

Thus,  
4L πx −(c2 π2 /L2 )t 1 3πx −(9c2 π2 /L2 )t
u(x, t) = 2 sin e − sin e + −··· .
π L 9 L
Figure 3 shows that the temperature decreases with increasing t, because of the heat loss
due to the cooling of the ends.

5
 If the ends are insulated instead of keeping at zero temperature, we then have the Neu-
mann conditions at x = 0 and x = L, i.e., ux (0, t) = ux (L, t) = 0 for all t. Then we replace
earlier SLP for X by the SLP,
X ′′ = λX, X ′ (0) = 0 = X ′ (L).
For λ < 0, we get X ≡ 0. For λ = 0 we get X ≡ X0 , a constant solution. For λ > 0, we take
λ = µ2 , µ > 0. We have X(x) = A cos µx + B sin µx. X ′ (0) = 0 gives B = 0 and X ′ (L) gives
µn = nπ/L. Thus,
∞
X 2 2 2 2 nπx
u(x, t) = An e−(c n π /L )t cos .
n=0
L
The condition u(x, 0) = f (x) implies
∞
X nπx
f (x) = An cos .
n=0
L
Hence Z L
1 L
A0 = f (x) dx = ,
L 0 4
Z L
2 nπx 2L  nπ 
An = f (x) cos dx = 2 2 2 cos − cos nπ − 1 , n = 1, 2, , . . . .
L 0 L nπ 2
Hence the solution is
 
L 8L 1 2πx −(4c2 π2 /L2 )t 1 6πx −(36c2 π2 /L2 )t
u(x, t) = − 2 cos e + 2 cos e + ··· .
4 π 22 L 6 L

6
The terms decrease with increasing t, and u → L/4 as t → ∞, this is the mean value of the
initial temperature. This is plausible because no heat can escape from this totally insulated
bar. In contrast, the cooling of the ends in the first part led to heat loss and u → 0, the
temperature at which the ends were kept.
Non-homogeneous Heat Equation
Now we consider the following non-homogeneous heat conduction problem in a bar of finite
length L, 
wt (x, t) − c2 wxx (x, t) = G(x, t), 0 < x < L, t > 0, 
w(x, 0) = g(x), 0 < x < L, (1)
w(0, t) = h1 (t), w(L, t) = h2 (t), t ≥ 0.


We can transform the general problem (GP) to a problem with homogeneous boundary
conditions by taking
 
L−x x
v(x, t) = w(x, t) − h1 (t) + h2 (t)
L L
as 
vt (x, t) − c2 vxx (x, t) = F (x, t), 0 < x < L, t > 0, 
v(x, 0) = f (x), 0 < x < L, (2)
v(0, t) = 0, v(L, t) = 0, t ≥ 0,


where  
L−x ′ x ′
F (x, t) = G(x, t) − h1 (t) + h2 (t) , 0 < x < L, t > 0,
L L
 
L−x x
f (x) = g(x) − h1 (0) + h2 (0) , 0 < x < L,
L L
Then we split the problem (1) into two sub-problems,

∗
vt∗ (x, t) − c2 vxx (x, t) = 0, 0 < x < L, t > 0, 
v ∗ (x, 0) = f (x), 0 < x < L, (3)
∗ ∗
v (0, t) = 0, v (L, t) = 0, t ≥ 0,


and 
vt∗∗ (x, t) − c2 vxx
∗∗
(x, t) = F (x, t), 0 < x < L, t > 0, 
v ∗∗ (x, 0) = 0, 0 < x < L, (4)
v ∗∗ (0, t) = 0, v ∗∗ (L, t) = 0, t ≥ 0,


Then v = v ∗ + v ∗∗ . The first solution v ∗ can be obtained as before and we now use Duhamel’s
Principle to obtain the second solution v ∗∗ .
Duhamel’s Formula
Let W (x, t, s) satisfy,

Wt (x, t, s) − c2 Wxx (x, t, s) = 0, 0 < x < L, t > 0, s ≥ 0, 
W (x, 0, s) = F (x, s), 0 < x < L, s ≥ 0 (5)
W (0, t, s) = 0, W (L, t, s) = 0, t ≥ 0, s ≥ 0.


Then Z t
∗∗
v (x, t) = W (x, t − s, s) ds, (6)
0

7
satisfies (4).
Proof. Clearly, the initial and boundary conditions of (4) are satisfied by v ∗∗ . By Leibniz
rule and (5), Z t
∗∗
vt (x, t) = W (x, 0, t) + Wt (x, t − s, s) ds
0
Z t
vt∗∗ (x, t) = F (x, t) + c2 Wxx (x, t − s, s) ds
0
t
∂2
Z 
2
= F (x, t) + c W (x, t − s, s) ds
∂x2 0

= F (x, t) + c2 v ∗∗ (x, t).

The solution W (x, t, s) of (5) is given by
∞
X n2 π 2 c2 nπx
W (x, t, s) = Bn (s)e− L2
t
sin ,
n=1
L

where, using W (x, 0, s) = F (x, s), we have

2 L
Z
nπξ
Bn (s) = F (ξ, s) sin dξ.
L 0 L
From (6), under suitable conditions on F which allow us to interchange the integration and
summation, we get
∞ Z Z
∗∗ 2 X t L − n2 π22 c2 (t−s) nπx nπξ
v (x, t) = e L sin sin F (ξ, s) dξ ds
L n=1 0 0 L L
Z tZ L
= K(x, ξ, t − s)F (ξ, s) dξ ds,
0 0
where ∞
2 X − n2 π22 c2 t nπx nπξ
K(x, ξ, t) = e L sin sin .
L n=1 L L
Method of Variation of Constants: We now consider, as an alternative, the method of
variation of constants to the non-homogeneous heat conduction problem in a bar of finite
length L, with homogeneous IC and homogeneous BCs (problem (4), with v ∗∗ replaced by
u, in the earlier section),

ut (x, t) − c2 uxx (x, t) = F (x, t), 0 < x < L, t > 0, 
u(x, 0) = 0, 0 < x < L, (1)
u(0, t) = 0, u(L, t) = 0, t ≥ 0.


Assuming that F (., t) has Fourier sine series, we take

∞
X ∞
X
u(x, t) = Bn (t) sin(nπx/L), F (x, t) = Fn (t) sin(nπx/L),
n=1 n=1

8
and put it in the equation to get the system of IVP of ODEs,

n2 π 2 c2
Bn′ (t) + Bn (t) = Fn (t), Bn (0) = 0, n = 1, 2, 3, . . . .
L2
Solving this system of IVPs of ODES, we get
Z t
2 2 2 2
Bn (t) = Fn (τ )e−(n π c /L )(t−τ ) dτ.
0

Thus, Z t 
−(n2 π 2 c2 /L2 )(t−τ )
u(x, t) = Fn (τ )e dτ sin(nπx/L).
0

Now, Z L
2
Fn (t) = F (t, ξ) sin(nπξ/L) dξ.
L 0
Hence
∞ Z t Z L  
2X −(n2 π 2 c2 /L2 )(t−τ )
u(x, t) = F (τ, ξ) sin(nπξ/L) dξ e dτ sin(nπx/L)
L n=1 0 0

Z t "Z L ∞
! #
2X 2 2 2
= sin(nπξ/L) sin(nπx/L)e−(n π c /L)t F (ξ, τ )dξ dτ,
0 0 L n=1
which is the same as we got earlier using Duahmel’s principle.

Maximum Principle for the Heat Equation

For ant T > 0 let Q = (0, L) × (0, T ), Q = [0, L] × [0, T ] and Γ0 = {(x, t) : x = 0 or x =
L} ∪ {(x, t) : t = 0}. Consider a continuous function u(x, t) on Q satisfying the heat equation

ut = c2 uxx , 0 < t ≤ T, 0 < x < L.

Let
MQ = max u(x, t), MΓ0 = max u(x, t)
(x,t)∈Q̄ (x,t)∈Γ0

Then
MΓ0 = MQ .
Proof: Clearly, MΓ0 ≤ MQ . Suppose that MΓ0 < MQ . Then there exists (x0 , t0 ) ∈ Q
such that u(x0 , t0 ) = MQ . Consider the auxiliary function

MQ − MΓ0
v(x, t) = u(x, t) + 2
(x − x0 )2 ,
4R
where R > 0 is large enough so that the disc (x − x0 )2 + (t − t0 )2 ≤ R2 includes Q. Then
v(x0 , t0 ) = MQ and for any (x, t) ∈ Γ0 we have v(x, t) < MQ . Therefore maximum of v(x, t)

9
attained in Q or on {(x, t) : 0 < x < L, t = T }. In Q or on {(x, t) : 0 < x < L, t = T }, we
have
MQ − MΓ0 MQ − MΓ0
−vt + c2 vxx = −ut + c2 uxx + c2 2
= c2 > 0 (∗),
2R 2R2
since ut = c2 uxx there. But if there is (x1 , t1 ) ∈ Q ∪ {(x, t) : 0 < x < L, t = T }, such that

v(x1 , t1 ) = max v(x, t),

Q̄

then
vt (x1 , t1 ) = 0, vxx (x1 , t1 ) ≤ 0, for 0 < t1 < T,
and for t1 = T
vt (x1 , t1 ) = vt (x1 , T −) ≥ 0,
since if vt (x1 , T −) < 0 then by the continuity of vt , for a small δ > 0, we have vt (x1 , s) < 0
for T − δ < s < T and hence
Z T
v(x1 , T ) − v(x1 , T − δ) = vt (x1 , s)ds < 0.
T −δ

Therefore v(x1 , T − δ) > v(x1 , T ), a contradiction to the fact that v(x1 , T ) is the maximum
in Q̄. Hence
−vt (x1 , t1 ) + c2 vxx (x1 , t1 ) ≤ 0,
which is a contradiction to (∗). Hence MQ̄ = MΓ0 .

10