1.

1D Heat equation

Suppose that a smooth solution upx, tq satisfies the following differential equation

ut “ u xx pHeat equationq, (1)

in tpx, tq : 0 ď x ď L, 0 ď tu. Then, upx, tq can represent the temperature under the heat flow on a rod
located in t0 ď x ď Lu. In order to solve the equation, we need the initial data

upx, 0q “ gpxq pCauchy conditionq, (2)

and one of the following boundary data

up0, tq “ h1 ptq, upL, tq “ h2 ptq pDirichlet conditionq, (3)

´u x p0, tq “ h1 ptq, u x pL, tq “ h2 ptq pNeumann conditionq, (4)

´u x p0, tq ` αup0, tq “ h1 ptq, u x pL, tq ` αupL, tq “ h2 ptq pRobin conditionq. (5)

2. Uniqueness

We establish the following uniqueness theorem.

Theorem 1 (Uniqueness). Given smooth functions gpxq, h1 ptq, h2 ptq, the heat equation (1) has at most
one smooth solution upx, tq satisfying (2) on t0 ď x ď Lu and (3) on tt ě 0u.

Proof. Suppose that upx, tq and vpx, tq are solutions satisfying the conditions. Then, the smooth func-
tion wpx, tq “ upx, tq ´ vpx, tq satisfies

wt “ ut ´ vt “ u xx ´ v xx “ w xx . (6)

Moreover, we can observe

wp0, tq “ wpL, tq “ 0. (7)

Next, we define an energy

żL
Eptq “ w2 px, tqdx.
0
Then, (6) shows żL żL
d
dt Eptq “ 2wwt dx “ 2 ww xx dx.
0 0
1
2

Using the integration by part and (7),

żL żL
2
d
dt Eptq “ 2ww x |0L ´2 |w x | dx “ ´2 |w x |2 dx ď 0. (8)
0 0

Therefore,
0 ď Eptq ď Ep0q, (9)

for all t ě 0. However, we have wpx, 0q “ 0 by definition, namely Ep0q “ 0. Thus, Eptq “ 0 and
wpx, tq “ 0. Hence, the smooth solution is unique. 

Remark. If h1 ptq “ h2 ptq “ 0, then we can modify the proof above to show

d L 2
ż
u px, tqdx ď 0. (10)
dt 0
Then, it would be a natural question to prove lim sup |upx, tq| “ 0. We will prove this next week,
tÑ`8 0ďxďL
but it’d be good to try to prove it yourself.

3. Review: Fourier series

We recall the Fourier series. In this class, we will use the following fact without proofs.

Given a smooth function f : r´L, Ls Ñ R with f p´Lq “ f pLq, the following holds

lim sup | f pxq ´ S N pxq| “ 0,

NÑ`8 |x|ďL

for the partial sums S N pxq of Fourier series,

8 8
a0 ÿ ÿ
S N pxq “ ` am cospmπx{Lq ` bm sinpmπx{Lq,
2 m“1 m“1

where
żL żL żL
1 1 ´ mπx ¯ 1 ´ mπx ¯
a0 “ f pxqdx, am “ f pxq cos dx, bm “ f pxq sin dx.
L ´L L ´L L L ´L L

Suppose that f : r0, Ls Ñ R is a smooth function satisfying f p0q “ 0. Then,

lim sup | f pxq ´ S N pxq| “ 0,

NÑ`8 0ďďL
 3

holds for the partial sums S N pxq of Fourier sine series,

8
ÿ
S N pxq “ bm sinpmπx{Lq,
m“1

where
żL
2 ´ mπx ¯
bm “ f pxq sin dx.
L 0 L

Suppose that f : r0, Ls Ñ R is a smooth function satisfying f 1 p0q “ 0. Then,

lim sup | f pxq ´ S N pxq| “ 0,

NÑ`8 0ďďL

holds for the partial sums S N pxq of Fourier cosine series,

8
a0 ÿ
S N pxq “ ` am cospmπx{Lq,
2 m“1

where
żL żL
2 2 ´ mπx ¯
a0 “ f pxqdx, am “ f pxq cos dx.
L 0 L 0 L

4. Review: ODE

We recall the some well-known results in ODEs. We will also use them without proofs.

Suppose that a function upxq satisfies the following differential equation

u2 pxq ` µ2 upxq “ 0. (11)

Then,
upxq “ c1 sinpµxq ` c2 cospµxq, (12)

for some constants c1 , c2 depending on initial (or boundary data). For example, if upxq satisfies up0q “
0 and u1 p0q “ 1, then the constants must be c1 “ µ´1 and c2 “ 0.

Suppose that a function upxq satisfies the following differential equation

u1 pxq “ λupxq. (13)

4

Then,
upxq “ ceλx , (14)

for some constant c depending on the initial data.

5. Separation of Variables

In this section, we will SOLVE the Cauchy-Dirichlet problem with the vanishing Dirichlet data.
Namely, given smooth gpxq, we will find the solutions to the heat equation (1) under the conditions
(2) and (3), where h1 ptq “ h2 ptq “ 0.

To begin with, we remind that by the uniqueness theorem 1 there exists at most one solution. Hence,
if we find a solution, then it is the only solution.

Next, we want find a function upx, tq “ vpxqwptq satisfying (1) and (3) with h1 “ h2 “ 0. (Notice
that in this step we do not consider (2), yet.) Then, (1) implies

wt v “ ut “ u xx “ wv xx .

Dividing by vw yields
wt ptq v xx pxq
“ .
wptq vpxq
The left hand side only depends on t, while the right hand side only depends on x. Therefore, there
exists some constant λ P R such that
wt v xx
“ “ λ.
w v
We consider the three cases that λ ą 0, λ “ 0, and λ ă 0.

Case 1: λ ą 0. In this case, by using the Dirichlet condition vp0q “ vpLq “ 0 we can obtain
żL żL żL żL żL
ˇL
0ďλ 2
v dx “ vpλvqdx “ vv xx dx “ vv x 0 ´
ˇ 2
|v x | dx “ ´ |v x |2 dx ď 0, (15)
0 0 0 0 0

namely v “ 0. Thus, u “ 0.

Case 2: λ “ 0. In this case, v xx “ 0 implies vpxq “ ax ` b. Hence, the Dirichlet condition

vp0q “ vpLq “ 0 guarantees v “ 0. Thus, u “ 0.
 5

Case 3: λ “ ´µ2 ă 0. In this case, the equation v xx ` µ2 v “ 0 has non-trivial solutions. By the
results in ODE, vpxq “ A cospµxq ` B sinpµxq holds for the constants A, B satisfying the boundary
conditions

0 “ vp0q “ A cos 0 ` B sin 0 “ A,

0 “ vpLq “ A cospµLq ` B sinpµLq “ B sinpµLq.

Hence, we have sinpµLq, and thus µL “ mπ for a natural number m. Namely, given m P N we have

vm “ c sinpmπx{Lq,

for some constant c. In addition, λ “ ´µ2 “ pmπ{Lq2 gives

d
dt wm “ ´µ2 wm “ ´pmπ{Lq2 wm . (16)

Hence, the ODE result says

wm “ c expp´pmπ{Lq2 tq

for some constant c. In conclusion, for each m P N and any constant Bm P R

um px, tq “ Bm expp´pmπ{Lq2 tq sinpmπx{Lq

satisfies (1) and (3) with h1 “ h2 “ 0.

By the result in the last case, we know that

8
ÿ
u“ Bm expp´pmπ{Lq2 tq sinpmπx{Lq, (17)
m“1

satisfies (1) and (3) with h1 “ h2 “ 0.

Now, we define the coefficients Bm by
żL
2
Bm “ gpxq sinpmπx{Lqdx. (18)
L 0

Then, by the Fourier series theorem above, the function upx, tq in (17) satisfies (2). Namely, it is the
desired solution.