June 20, 2020 APM346 – Week 6 Justin Ko

1 Properties of the Heat Equation on R

Recall that the solution to (
ut − kuxx = f (x, t) x ∈ R, t > 0,
(1)
u|t=0 = g(x) x ∈ R.
is given by
Z ∞ Z tZ ∞ (x−y)2
1 (x−y)2 1
u(x, t) = √ e − 4kt g(y) dy + p e− 4k(t−s) f (y, s) dyds. (2)
4πkt −∞ 0 −∞ 4πk(t − s)

1.1 Well-Posed
Given some minor integrability assumptions on g (bounded and continuous), we can prove the existence
of a C ∞ solution to (1) using (2). We can show the solutions are also unique and stable.
Proposition 1 (Uniqueness of Solutions that Decay at Infinity )
If u and its derivatives decay at infinity, then (1) has a unique solution.

Proof. We use an energy argument.

Difference of Solutions: Suppose u1 and u2 are solutions to (1) that decay at infinity. By linear-
ity, v = u1 − u2 solves (
ut − kuxx = 0 x ∈ R, t > 0,
(3)
u|t=0 = 0 x ∈ R.
To prove uniqueness, it suffices to show that v ≡ 0 on the domain of the solution.

Show the Energy is Zero: We consider the energy of the solution v to (3),
1 ∞ 2
Z
E(t) = v dx.
2 −∞
By the assumptions on the decay of u, we can differentiate under the integral sign with respect to t to
conclude that
Z ∞
E 0 (t) = vt v dx
−∞
Z ∞
=k vxx v dx vt − kvxx = 0
−∞
Z ∞ x=∞
= −k vx2 dx + (vx v) Integrate by Parts

−∞ x=−∞
Z ∞
= −k vx2 dx lim u = 0, lim ux = 0
−∞ x→±∞ x→±∞

≤ 0.
Since E 0 (t) ≤ 0, we can conclude that E(t) is decreasing by the mean value theorem. Furthermore,
the initial conditions imply
1 ∞
Z
E(0) = v(x, 0)2 dx = 0
2 −∞
because v(x, 0) = 0. This implies that E(t) ≤ 0. Combined with the fact E(t) ≥ 0 since it is the
integral of non-negative functions, this implies
0 ≤ E(t) ≤ 0 =⇒ E(t) = 0 for all t.

Page 1 of 7
June 20, 2020 APM346 – Week 6 Justin Ko

Show the Difference is Zero: Since E(t) is the integral of a sum of squares of continuous functions,
each term in the integrand must be 0 so

v 2 (x, t) = 0 for all x ∈ R and t ≥ 0 =⇒ v(x, t) ≡ 0.

Therefore, u1 = u2 , so the solution to (3) is unique.

Remark 1. We can also prove uniqueness for the homogeneous heat equation using by applying the
maximum principle covered in the next section and taking limits. We need to assume some integrability
on g to ensure that this limiting procedure is valid.
Proposition 2 (Stability of Homogeneous Solutions that Decay at Infinity )
If f = 0 and u and its derivatives decay at infinity, then (1) is stable.

Proof. Let u1 be the solution to the homogeneous version of (1) with initial data g1 and u2 be the
solution to the homogeneous version of (1) with and initial data g2 . We consider the energy of the
solution v to (3),
1 ∞ 2
Z
E(t) = v dx.
2 −∞
The computations in the proof of uniqueness imply that E 0 (t) ≤ 0, so E(t) is decreasing. Furthermore,
we have
1 ∞ 1 ∞
Z Z
E(0) = v(x, 0)2 dx = (g1 (x) − g2 (x))2 dx = 0
2 −∞ 2 −∞
because v1 (x, 0) = g1 (x) and v2 (x, 0) = g2 (x). We define the L2 norm of f as
Z ∞ 
2
kf k2 = f (x) dx .
−∞

1
Therefore, E(t) ≤ E(0) = 2 kg1 − g2 k22 by the mean value theorem. For every  > 0, if we take
kg1 − g2 k2 ≤ , then
Z ∞
1 1 1
ku1 − u2 k22 = E(t) = (u1 − u2 )2 dx ≤ kg1 − g2 k22 =⇒ ku1 − u2 k2 ≤ 
2 2 −∞ 2

for all t. This implies stability for all t in terms of the “square error”.
2
Remark 2. The heat equation is not well-posed for t < 0. For example, take un = 1
n sin(nx)e−n kt
.

1.2 Symmetry
It is easy to check (2) implies that the solution u(x, t) inherits the symmetry properties of the initial
conditions and inhomogeneous term,
Proposition 3 (Symmetry )
Let u(x, t) be the solution to (1).
(i) If f and g are even in x then u(x, t) is even in x.

(ii) If f and g are odd in x then u(x, t) is odd in x.

This means we can use odd or even reflections to solve the heat equation on the half line, in exactly
the same way as for the half line wave equation.

Page 2 of 7
June 20, 2020 APM346 – Week 6 Justin Ko

1.3 Example Problems

Problem 1.1. (?) Solve the following IBVP

ut − kuxx = 0
 x > 0, t > 0,
u|t=0 = g(x) x>0

u|x=0 = 0 t > 0.


Solution 1.1. Since we have Dirichlet boundary conditions, we can find a solution using an odd-
extension. Define 
g(x)
 x>0
godd (x) = 0 x=0

−g(−x) x < 0.


For x > 0, the particular solution is given by

Z ∞
1 (x−y)2
u(x, t) = √ e− 4kt godd (y) dy
4πkt −∞
Z ∞ Z 0
1 (x−y)2
− 4kt 1 (x−y)2
= √ e g(y) dy − √ e− 4kt g(−y) dy
4πkt 0 4πkt −∞
Z ∞ Z 0
1 (x−y)2 1 (x+ỹ)2
=√ e− 4kt g(y) dy + √ e− 4kt g(ỹ) dỹ ỹ = −y
4πkt 0 4πkt ∞
Z ∞ 
1 (x−y)2
− 4kt
(x+y)2
− 4kt
=√ e −e g(y) dy.
4πkt 0

Problem 1.2. (?) Solve the following IBVP


ut − kuxx = 0
 x > 0, t > 0,
u|t=0 = g(x) x>0

ux |x=0 = 0 t > 0.


Solution 1.2. Since we have Neumann boundary conditions, we can find a solution using an even-
extension. Define (
g(x) x≥0
geven (x) =
g(−x) x ≤ 0.
For x > 0, the particular solution is given by
Z ∞
1 (x−y)2
u(x, t) = √ e− 4kt geven (y) dy
4πkt −∞
Z ∞ Z 0
1 (x−y)2
− 4kt 1 (x−y)2
=√ e g(y) dy + √ e− 4kt g(−y) dy
4πkt 0 4πkt −∞
Z ∞ Z 0
1 (x−y)2 1 (x+ỹ)2
=√ e− 4kt g(y) dy − √ e− 4kt g(ỹ) dỹ ỹ = −y
4πkt 0 4πkt ∞
Z ∞ 
1 (x−y)2
− 4kt
(x+y)2
− 4kt
=√ e +e g(y) dy.
4πkt 0

Page 3 of 7
June 20, 2020 APM346 – Week 6 Justin Ko

Problem 1.3. (??) Solve the following IBVP


ut − kuxx = 0
 x > 0, t > 0,
u|t=0 = 0 x>0

u|x=0 = p(t) t > 0.


Solution 1.3. We reduce to a problem with homogeneous boundary conditions by doing a change of
variables.

Change of Variables: We define v(x, t) = u(x, t) − p(t). It is easy to check that v solves

0
vt − kvxx = −p (t)
 x > 0, t > 0,
v|t=0 = −p(0) x>0

v|x=0 = 0 t > 0.


Particular Solution: This is an inhomogeneous heat equation with Dirichlet boundary conditions, so
we can solve this using an odd reflection. We define
 
0
−p(0) x > 0
 −p (t) x > 0

godd (x) = 0 x=0 and fodd (x, t) = 0 x=0
  0
p(0) x < 0. p (t) x < 0.
 

By (2),
Z ∞ Z tZ ∞ (x−y)2
1 (x−y)2 1
v(x, t) = √ e − 4kt godd (y) dy + p e− 4k(t−s) fodd (y, s) dyds.
4πkt −∞ 0 −∞ 4πk(t − s)

Proceeding like in Problem 1.1, this simplifies to

Z ∞  Z tZ ∞  
1 (x−y)2 (x+y)2 1 (x−y)2 (x+y)2
−√ e− 4kt − e− 4kt p(0) dy − p e− 4k(t−s) − e− 4k(t−s) p0 (s)dyds.
4πkt 0 0 0 4πk(t − s)
(4)
Original Solution: We now write our solution in terms of u to conclude

u(x, t) = v(x, t) + p(t),

where v(t) was defined in (4).

Remark 3. To solve an inhomogeneous Neumann problem ux |x=0 = p(t), you do the change of
variables v(x, t) = u(x, t) − xp(t) to reduce it to the homogeneous Neumann problem.

Page 4 of 7
June 20, 2020 APM346 – Week 6 Justin Ko

2 Properties of the Heat Equation on Finite Regions

2.1 Maximum Principle
Consider the closed rectangular domain

ΩT = {(x, t) : 0 < x < L, 0 < t ≤ T }.

We define the parabolic boundary of ΩT by

ΓT = {(x, t) : x ∈ [0, L], t = 0 or x = 0, t ∈ [0, T ] or x = L, t ∈ [0, T ]}

to be boundary of ΩT without the top line {0 < x < L} × {t = T }.

t

ΩT

ΓT
x

Remark 4. The restriction that x ∈ [0, L] is not important. The maximum principle will hold for
any finite interval x ∈ [a, b] by applying the maximum principle to v(x, t) = u(x − a, t).
If u is continuous on Ω̄T and satisfies the heat equation ut − kuxx = 0 on ΩT , then the maximum value
of u occurs on ΓT . That is, the maximum of u is determined by the initial and boundary conditions.
Theorem 1 (The Maximum Principle)
If u is continuous and satisfies ut − kuxx = 0 on ΩT , then

max u = max u. (5)

Ω̄T ΓT

Proof. Since u is continuous and Ω̄T is compact, u attains a global maximum value at some (x0 , t0 ) ∈
Ω̄T . We will show that (x0 , t0 ) must be on ΓT .

Perturbing the Solution: We will add a perturbation to the u to force its second derivative in x
to have a sign. For  > 0, consider
v(x, t) = u(x, t) + x2 .
Notice that v satisfies the diffusion inequality on ΩT

vt − kvxx = ut − kuxx − 2k = −2k < 0. (6)

We will use this fact to show that v cannot have an interior maximum.

Location of Maximum: The maximum must occur in one of 3 places.

1. Interior: Suppose that v attains its global maximum at (x0 , t0 ) in the interior of the rectangle
ΩT . Since v(x0 , t0 ) is an interior maximum, the second derivative test implies that

vt (x0 , t0 ) = vx (x0 , t0 ) = 0 and vxx (x0 , t0 ) ≤ 0 =⇒ vt (x0 , t0 ) − kvxx (x0 , t0 ) ≥ 0.

This contradicts the diffusion inequality (6), so v cannot attains its maximum in ΩT .

Page 5 of 7
June 20, 2020 APM346 – Week 6 Justin Ko

2. Top: Similarly, suppose that v attains its global maximum at some point (x0 , T ) on the top of
the rectangle ΩT . On the boundary, the second derivative test and the fact v(x0 , T ) ≥ v(x0 , t)
for any t < T , implies that
vt (x0 , T ) ≥ 0, vx (x0 , T ) = 0 and vxx (x0 , T ) ≤ 0 =⇒ vt (x0 , T ) − kvxx (x0 , T ) ≥ 0.
This contradicts the diffusion inequality (6), so v cannot attain the maximum on the top of ΩT .
3. Parabolic Boundary: The extreme value theorem implies that v must attain its maximum
somewhere, so it must attain its global maximum on ΓT , i.e.
max v = max v.
Ω̄T ΓT

Removing the Perturbation: We have shown that v(x, t) attains a maximum at some point (x0 , t0 ) ∈ ΓT .
Since x ∈ [0, L] so x2 ≤ L2 for all (x, t) ∈ Ω̄T . Since 0 ≤ x2 ≤ L2 on Ω̄T ,
max u ≤ max v ≤ v(x0 , t0 ) ≤ u(x0 , t0 ) + x20 ≤ max u + L2 .
Ω̄T Ω̄T ΓT

The upperbound holds for all  > 0, so taking  → 0 implies

max u ≤ max u =⇒ max u = max u
Ω̄T ΓT Ω̄T ΓT

since ΓT ⊆ Ω̄T .
Remark 5. A minimum principle also hold by applying the maximum principle to v = −u. That is,
if u is continuous and satisfies ut − kuxx = 0 on ΩT , then
min u = min u. (7)
Ω̄T ΓT

Remark 6. It was essential that we have a strict inequality in (6). Without a strict inequality, an
interior maximum might not lead to a contradiction. For example, the constant solution u = 0 is a
continuous solution to the heat equation, but it has a global maximum on its interior.

2.2 Well-Posed
Consider the Dirichlet problem for the heat equation,

ut − kuxx = f (x, t)


0 < x < L, t > 0
u|
t=0 = g(x) 0<x<L
(8)
u|x=0 = p(t)
 t>0

u|x=L = q(t) t > 0.


We will introduce techniques to solve this IBVP in Week 7. We can prove uniqueness and stability
without even solving it.
Corollary 1 (Uniqueness of the Inhomogeneous Dirichlet Problem)
Continuous solutions to (8) are unique.

Proof. Uniqueness is a straightforward consequences of Theorem 1. We fix T > 0.

Uniqueness: Suppose u1 and u2 are solutions to (8). By linearity, v = u1 − u2 solves


vt − kvxx = 0


0 < x < L, 0 < t ≤ T
v|
t=0 = 0 0<x<L
(9)
v|x=0 = 0
 t>0

v|x=L = 0 t > 0.


Page 6 of 7
June 20, 2020 APM346 – Week 6 Justin Ko

Since v solves the heat equation and v = 0 on ΓT , the maximum principle implies for any (x, t) ∈ Ω̄T ,
v(x, t) ≤ max v = max v = 0.
Ω̄T ΓT

Similarly, the minimum principle implies

v(x, t) ≥ min v = min v = 0.
Ω̄T ΓT

Therefore, v(x, t) = 0 on Ω̄T , so u1 = u2 on Ω̄T . Taking T → ∞ implies uniqueness for all t > 0.
Corollary 2 (Stability of the Homogeneous Dirichlet Problem)
If f = 0, then continuous solutions to (8) are stable.

Proof. Stability is a straightforward consequences of Theorem 1. Let u1 be the solution to (8) with
initial data (g1 , p1 , q1 ) and u2 be the solution to (8) with initial data (g2 , p2 , q2 ). If we define v = u1 −u2 ,
then v satisfies (8) with initial data (g1 − g2 , p1 − p2 , q1 − q2 ). The maximum principle implies for any
(x, t) ∈ Ω̄T ,
v = u1 − u2 ≤ max(kg1 − g2 k∞ , kp1 − p2 k∞ , kq1 − q2 k∞ ),
and the minimum principle implies
v = u1 − u2 ≥ − max(kg1 − g2 k∞ , kp1 − p2 k∞ , kq1 − q2 k∞ ).
For every  > 0, if we take max(kg1 − g2 k∞ , kp1 − p2 k∞ , kq1 − q2 k∞ ) ≤ , then
ku1 − u2 kT ≤ .

2.3 Example Problems

Problem 2.1. (?) Let u(x, t) = 1 − x2 − 2t and ΩT = {0 ≤ x ≤ 1, 0 ≤ t ≤ T }. Find the maximum
and minimum values of u on ΩT .

Solution 2.1. It is easy to see that

ut − uxx = −2 + 2 = 0
so u solves the heat equation on ΩT . By the maximum and minimum principles, it suffices to look for
minimizers and maximizers on ΓT , the shaped boundary of the closed rectangle.

We optimize over each of the three sides of the rectangle.

1. {0 ≤ x ≤ 1, t = 0}: We have
u(x, 0) = 1 − x2 0 ≤ x ≤ 1,
has a maximum of 1 when x = 0 and a minimum of 0 when x = 1.
2. {x = 0, 0 ≤ t ≤ T }: We have
u(0, t) = 1 − 2t 0 ≤ t ≤ T,
has a maximum of 1 when t = 0 and a minimum of 1 − 2T when t = T .
3. {x = 1, 0 ≤ t ≤ T }: We have
u(1, t) = −2t 0 ≤ t ≤ T,
has a maximum of 0 when t = 0 and a minimum of −2T when t = T .
From above, we see that the maximum of 1 occurs at the point x = 0, t = 0 and the minimum of −2T
occurs when x = 1, t = T .

Page 7 of 7