Advection Term; Transport equation

Sujob

July 2025

1. Transport Equation
1.1. Example
Consider the following PDE
(
∂t f + ∂x f = 0
f (x, 0) = f0 (x)

where f is a function from R × [0, ∞) to R. The solution is just f (x, t) = f0 (x − t). Then this
f has the same shape as f0 (x) at time t.

1.2. Example
Consider the following case:
(
∂t f + x∂x f = 0
f (x, 0) = f0 (x)
Then
 
df0 −t
∂t f = (e x) (−e−t x).
dx
   
df0 −t −k df0 −t
x∂x f = (e x) e = (e x) (e−t x).
dx dx

If generalized, to arbitrary function v(t, x) term multiplied to ∂x f ,

(
∂t f + v(t, x)∂x f = 0
f (x, 0) = f0 (x)
where v, f is given as C ∞ .
The problem here is the well-posedness of the equation in C ∞ . The well-posedness problem
consists of two subproblems:
(1) Find a function f ∈ C ∞ (R × [0, ∞)) satisfying the equation on (x, t) ∈ R × [0, ∞) [Exis-
tence];
(2) Prove that the function found in (1) is the only solution to the equation [Uniqueness].
1.3. Theorem

(
∂t f + ∂x f = 0
f (x, t) = f0 (x)
is well-posed in C ∞ .
1
 Proof ) For the uniqueness, f0 (x−t) ∈ C ∞ (R×[0, ∞)) since f0 ∈ C ∞ (R) and f (x, t) = f0 (x−t)
satisfies the equation.
Assume g(x, t) ∈ C ∞ and satisfies the equation. Consider G(x, t) = g(x + t, t) ∈ C ∞ and
take
 
∂G ∂g ∂g
= (+1) + ≡0
∂t ∂x ∂t
giving G(x, t) = G(x, 0) = g(x, 0) = f0 (x) for all x, t. Hence take z = x + t, then x = z − t.

1.4. Definition
Consider functions defined on R.
(1) f is in Cloc (R) if f is continuous on R and bounded on any bounded set of R.
df
(2) f is in C 1 (R) if f is differentiable on R and f and f ′ = dx are continuous and uniformly
1
bounded on R. Define the C - norm of f as

∥f ∥C 1 (R) = ∥f ∥C(R) + ∥f ′ ∥C(R) .

dk f
(3) f is in C k (R) if f ∈ C k−1 (R) and f is k-times differentiable, and dxk
is uniformly bounded
and continuous on R. Define the C k -norm as
dk f
∥f ∥C k (R) = ∥f ∥C k−1 (R) + .
dxk C(R)

(4) f is in C ∞ (R)([Link]
∞
(R)) if f ∈ ∩k≤0 C k (R)([Link]
k
(R)).
k k
(5) f is in Cc (R) if f ∈ C (R) and it is compactly supported.

1.5. Remark
dk f
Does it makes sense to define a norm of C ∞ space? Even in the ∥f ∥C ∞ = ∞
P −k100
1 e dxk
.
C(R)

C ∞ is not normable and also not Cloc

k
. It is due to the fact that functions in these spaces
have no restrictions at their growth at infinity. Below is one example of the function whose
growth is bounded.
1.6. Example
f is in Clin (R) if f ∈ Cloc (R) and there exists M > 0 such that |f (x)| ≤ M (1 + |x|) for all
x ∈ R.

1.7. Example
(1) v(t, x) ≡ v0 ∈ R.
Then the solution f (t, x) is of the form f (t, x) = f0 (x − v0 t) for (t, x) ∈ R × R. The charac-
teristic curve is x − t = C.
(2) v(t, x) ≡ x.
Then f (t, x) = f0 (e−t x) for (t, x) ∈ R × R. The characteristic curve in this case is x = e−t C.

1.8. Definition
Suppose that for all (t, x), there exists a curve containing that point and the curve inter-
sects {t = 0} at a unique point. Moreover, f is invariant on each curves. This f is called a
characteristic or trajectory.

2
 The method of trajectory or characteristics must satisfy
d
X(t, a) = v(t, X(t, a)),
dt
X(0, a) = a.
Assume that for every a ∈ R, X(t, a) exists and is smooth. Assume that f solves the linear
transport problem. Then
d dX
f (t, X(t, a)) = (∂t f )(t, X) + (∂x f )(t, X) ·
dt dt
= (∂t f + v∂x f )(t, X) ≡ 0.
Then f (t, X(t, a)) = f (0, X(0, a)) = f (0, a) = f0 (a). Fix t. Assume that the map a 7→ X(t, a) is
invertible. Denote the inverse by X −1 (t, a): X −1 (t, X(t, a)) = a. Then from f (t, X(t, a)) = f0 (a),
we get f (t, x) = f0 (X −1 (t, x)).
1.9. Remark
In fact, when v is independent of t, then X −1 (t, ·) and X(−t, ·) are equal but in general case,
that is not true. If v ≡ v0 , then X(t, a) = a + v0 t and X −1 (t, a) = a − v0 t ≡ X(−t, a). If v ≡ x,
then X(t, a) = et a. Hence X −1 (t, a) = e−t a ≡ X(−t, a).
This implies that if one attempts to guess the location of the particle prior to some time
period t, then the point is in fact, X −1 (t, a).

From the theory of ODEs,

1.10. Theorem (Well-definedness of the trajectory)
Consider the following differential equations:
(
d
dt
X(t, a) = V (t, X(t, a))
X(0, a) = a.

Assume V ∈ C k (R × R), k ≥ 1. Then

(1) ∀a ∈ R, there uniquely exists X(t, a) satisfying the ODE for all t ∈ R;
(2) for all t ∈ R, the map a 7→ X(t, a) is a C k -diffeomorphism of R.
Proof ) (1) follows from the theory of ODEs.
(2) Consider k = 1. We first show that X is differentiable. First, assume the differentiability.
Then
d
∂a X = (∂x V )(t, X)∂a X
dt
∂a X(0, a) = 1.
must be satisfied. Hence, for the actual proof, consider the ODE
(
d
dt
Z(t, a) = ∂x V (t, X(t, a))Z(t, a)
Z(0, a) = 1.
Now we claim that for every t, a,
1
W (t) = (X(t, a + h) − X(t, a)) ,
h
|W (t) − F (t, a)| → 0 as h → 0.
This W (t) satisfies the differential equations. Let
X(t, a + h) − X(t, a)
∆(t, h) =
3
h
We want to show that limh→0 ∆(t, h) = F (t, a).
First, we find the ODE satisfied by ∆.
d 1
∆(t, h) = [V (t, X(t, a + h)) − V (t, X(t, a))]
dt h
By the Mean Value Theorem applied to V with respect to its second argument, there exists
a ch between X(t, a) and X(t, a + h) such that:
∂V
V (t, X(t, a + h)) − V (t, X(t, a)) = (t, ch ) · (X(t, a + h) − X(t, a))
∂x
Substituting this back, we get the ODE for ∆:
d ∂V X(t, a + h) − X(t, a) ∂V
∆(t, h) = (t, ch ) · = (t, ch ) · ∆(t, h)
dt ∂x h ∂x
Now, let W (t, h) = ∆(t, h) − Z(t, a). Our goal is to show that W → 0 as h → 0. The initial
condition is W (0, h) = ∆(0, h) − Z(0, a) = (a+h)−a
h
− 1 = 0.
The derivative of W is:
dW d∆ dZ
= −
dt dt dt
∂V ∂V
= (t, ch )∆ − (t, X(t, a))Z
∂x ∂x
∂V ∂V
= (t, ch )(W + Z) − (t, X(t, a))Z
∂x  ∂x 
∂V ∂V ∂V
= (t, ch )W + (t, ch ) − (t, X(t, a)) Z
∂x ∂x ∂x
This is a linear ODE for W . By the theorem on continuous dependence on initial conditions
(a weaker result), we know X(t, a + h) → X(t, a) as h → 0. This implies ch → X(t, a). Since
V is C 1 , its partial derivative ∂V
∂x
is continuous, so the term in the brackets goes to zero as
h → 0.
An application of Gronwall’s inequality to the linear ODE for W shows that since W (0, h) =
0 and the non-homogeneous term vanishes as h → 0, we must have W (t, h) → 0 for t ∈ [0, T ].
ntegrating from 0 to t and using the initial condition yields:
Z t
W (t, h) = [A(s, h)W (s, h) + B(s, h)] ds
0

Taking the absolute value and applying the triangle inequality:

Z t
|W (t, h)| ≤ |A(s, h)W (s, h) + B(s, h)| ds
0
Z t
≤ (|A(s, h)||W (s, h)| + |B(s, h)|) ds
0

∂V
Since |A(s, h)| = ∂x
(s, ch )is bounded by a constant L > 0 for t ∈ [0, T ], we have:
Z t Z t
|W (t, h)| ≤ L|W (s, h)| ds + |B(s, h)| ds
0 0
Rt
This expression is in the form u(t) ≤ α(t) + 0 β(s)u(s) ds, where: u(t) = |W (t, h)|, α(t) =
Rt Rt
|B(s, h)| ds, β(s) = L. The Gronwall-Bellman inequality states that u(t) ≤ α(t)e 0 β(s)ds .
0

4
 Applying it gives:
Z t  Rt
|W (t, h)| ≤ |B(s, h)| ds e 0 L ds
Z0 t 
≤ |B(s, h)| ds eLt
0

We define εh as the maximum of the difference term over the interval:

∂V ∂V
εh = max (s, ch ) − (s, X(s, a))
s∈[0,T ] ∂x ∂x
Due to uniform continuity, limh→0 εh = 0. Since Z(t) is bounded by a constant M , we can
bound the integral term:
Z t Z t
|B(s, h)| ds ≤ (εh · M ) ds = εh M t
0 0

Substituting this back into our Gronwall result gives the final bound:
|W (t, h)| ≤ (εh M t)eLt
Taking the limit as h → 0:

lim |W (t, h)| ≤ lim (εh M teLt ) = 0

h→0 h→0

Since |W (t, h)| ≥ 0, we conclude that limh→0 |W (t, h)| = 0.

This proves that limh→0 ∆(t, h) = Z(t, a), so the derivative ∂X∂a
exists. Its continuity with
respect to a follows from the continuous dependence of solutions of linear ODEs (the varia-
tional equation) on their coefficients, which are continuous in a. Thus, X(t, a) is C 1 in a.
Now we show that inf a∈R ∂a X(t, a) > 0. Then X is strictly increasing in a and X is invert-
ible and inverse is continuously differentiable by the inverse function theorem. We know that
∂a X satisfies
d
∂a X = ∂x V (t, X)∂a X,
dt
R 
t ′ ′
thus giving ∂a X(t, a) = ∂a X(0, a) exp 0 ∂x V (t , X(t , a)) dt. Now we can apply the same
method for C k differentiability.

1.11. Theorem
Consider the linear transport equation:
(
∂t f + v(t, x)∂x f = 0
(1)
f (0, x) = f0 (x)

where v(t, x) : R × R → R, f0 : R × R → R are C ∞ . Then the above problem is well-posed.

Proof ) The solution to the transport equation can be obtained by defining f (t, x) = f0 (X −1 (t, x)) ⇔
f (t, X(t, x)) = f0 (x). Clearly, f (0, x) = f0 (x) and ∂t f + v∂x f = 0 by the above theorem.
If there is another solution, then consider
d
[g(t, X(t, a))] ≡ 0
dt
and g(t, X(t, x)) = f0 (x).
5
1.12. Remark
There exists another uniqueness proof; energy method.
Assume that there are two solutions f, g. Then define the energy as
Z ∞
E(t) := |f (t, x) − g(t, x)|2 dx
−∞

and differentiate by t. That is,

Z ∞ Z ∞
d
E=2 (vf − g)(∂t f − ∂g )dx = −2 (f − g)(v∂x f − v∂x g)dx
dt −∞ −∞
Z ∞ Z ∞
2
=− v∂x [(f − g) ]dx = ∂x V (f − g)2 dx
Z−∞
∞
−∞

≤C (f − g)2 dx = CE(t).
−∞

Hence E(t) ≤ E(0)eCt but E(0) = 0. This proof is basically incorrect as one can not use
integration by parts and even assume the finiteness of the energy E(t).

In mitigation of this situation, we introduce a weighted energy.

General Linear Transport Equation

The below problem is a more generalized version of the linear transport equation.
(
∂t f + v∂x f = Sf + G
.
f (0, x) = f0 (x)

given V, S, G ∈ C ∞ (R × R), f0 ∈ C ∞ (R).

1.13. Example
The following equations are called mass preserving transport equation.
(
∂t ρ + ∂x (vρ) = 0
(*)
ρ(0, x) = ρ0 (x).
We claim that the solution to the mass transport problem satisfies
Z b Z X(T,b)
ρ0 (x)dx = ρ(T, a)dx.
a X(T,a)

This is equivalent to showing that

"Z #
X(t,b)
d
ρ(t, x)dx = 0
dt X(t,a)
Z
dX(t, b)
= ∂t ρ + ρ(t, X(t, b))−
dt
1.14. Theorem
Consider the general linear transport problem:
(
∂t f + V ∂x f = Sf + G
.
f (0, x) = f0 (x)

given V, S, G ∈ C ∞ (R × R), f0 ∈ C ∞ (R). Then the solution exists and is unique.

6
 Proof ) Suppose that the solution exists. Then consider F (t, a) := f (t, X(t, a)) and we compute
 
d d
F (t, a) = ∂t f + ∂x f X (t, X(t, a)) = (Sf + G)(t, X(t, a))
dt dt
= S(t, X(t, a))F (t, a) + G(t, X(t, a)).
Then
Z t 
′ ′
f (t, X(t, a)) = F (t, a) = exp S(t , X(t , a))dt F (0, a)
0
Z t Z t 
′ ′ ′′ ′′ ′′
+ G(t , X(t , a)) exp S(t , X(t , a))dt
0 t′

by the Duhamel’s principle. Now insert X −1 (t, a) to a position. Then we obtain the explicit
formula for f . This f is actually a solution.

1.15. Lemma
Let Y be a C 1 function of t satisfying
d
Y ≤ M Y.
dt
Then
d −M t
(e Y (t)) = −M e−M t Y + e−M t Y ′ ≤ −M e−M t Y + e−M t M Y
dt
≤0
Proof ) The proof uses something similar like Gronwell’s formula.
d −M t
(e Y (t)) = −M e−M t Y + e−M t Y ′ ≤ −M e−M t Y + e−M t M Y
dt
≤0

1.16. Theorem
Let v, S, G ∈ C 1 ([−T, T ]) × R and f0 ∈ C 1 (R). Then if f, g are C 1 ([−T, T ] × R) satisfying
(∗) in [−T, T ] × R, then
f ≡ g, ∀x ∈ [−T, T ] × R.
Proof ) Consider the following weighted energy;
Z ∞
1
E(t) = 2
|f (t, x) − g(t, x)|2 dx.
−∞ 1 + x
If f, g ∈ C(R), then E < ∞ and
Z ∞
dE 2
= 2
(f − g)(−v∂x (f − g) + S(f − g))dx
dt −∞ 1 + x
Z ∞
1 2 2S
= ∂x ( 2
v)(f − g)2 + 2
(f − g)2 dx by integration by parts
2 1 + x 1 + x
Z−∞
∞  
∂x v + 2S 4x
= 2
− 2 )2
v (f − g)2 dx
−∞ 1 + x (1 + x
Z ∞
(f − g)2

4x dE
= (∂x v + 2S) − v dx ⇒ ≤ ME
−∞ 1 + x2 1 + x2 dt
For t < 0, E(t) ≤ E(0) exp(M t) = 0 giving the uniqueness.
7
 There may exist nonbounded solutions. Assume f, g are solution which are continuous and
are allowed to grow linearly in space i.e.

∃M : |f (t, x)|, |g(t, x)| ≤ M (1 + |x|).

In this case, we use the localized energy method. Assume for simplicity, that supp(f ), supp(g) ⊂
[−T, T ] × [−1, 1]. Then
Z 1
E(t) = |f (t, x) − g(t, x)|2 dx.
−1

However, problem here is that the energy here can no longer control the energy that escapes
x = 1. In this case we allow the interval of integration to shrink faster than the velocity. That
is,
Z 1−Rt
Eloc (t) = |f (t, x) − g(t, x)|2 dx.
−1+Rt
Z 1−Rt
d 1−Rt
Eloc (t) = (∂x v)(f − g)2 dx + v(f − g)2 |−1+Rt + R(f − g)2 |1−Rt
−1+Rt .
dt −1+Rt

where R ≥ sup(t,x) |v, (t, x)|. Then

d
Eloc (t) ≤ sup(∂x V )Eloc (t).
dt (t,x)

8
2. The Transport Diffusion Equation
In solving Navier-Stokes equations,
(
∂t u + u · ∇u = −∇p + ∆u
∇ · (u) = 0

mathematicians faced three key difficulties; x ∈ R3 ; u must satisfy total three equations at once
; velocity in the transport is an unknown term turning the equaiton to be nonlinear. If v were
given, then the equation would have been linear.

Generalizing the domain to Rn , n ≥ 2

To remedy the first difficulty, mathematicians extended the linear transport equation to
R cases. Given V (t, x) : R × Rn → Rn where V (t, x) = (v1 (t, x), · · · , vn (t, x)) with v ∈
n

(C ∞ (R × Rn ))n and f0 : Rn → R, f0 ∈ C ∞ (Rn ), consider the

(
∂t f + v · ∇f = 0
(*)
f (0, x) = f0

where V · ∇ = v1 ∂x1 + · · · + vn ∂xn .

2.1. Theorem
n
If V ∈ (C ∞(R×R ) )n , then
(
d
dt
X(t, a)
= V (t, X(t, a))
X(0, a) = a

has a unique solution ∀a ∈ Rn , ∀t ∈ R, the map a 7→ X(t, a) : Rn → Rn is a C ∞ diffeomorphism

of Rn .
Proof ) Similar to the transport case, we take the derivative of X(t, a) by a.

d
∇a X(t, a) = ∇x V (t, X(t, a))∇a X(t, a)
dt
∂a1 X1 ∂a1 X2 · · · ∂a1 Xn ∂x1 v1 ∂x1 v2 · · · ∂x1 vn 
   
∂a1 X

d  a2 X1 ∂a2 X2
∂ · · · ∂a2 Xn   ∂x2 v1 ∂x2 v2 · · · ∂x2 vn 
 ...  i.e. d A = DA.

. .. .. ..  = .. .. .. .. 
dt  .. . . .   . . . .  dt
∂an X
∂an X1 ∂an X2 · · · ∂an Xn ∂xn v1 ∂xn v2 · · · ∂xn vn
If we take a determinant in both hands, then
d
det A = Tr(D) det(A)
dt
Let J(t, a) = det ∇a X(t, a) = det A. Then

∇ · (V ) = ∂x1 v! + · · · + ∂xn vn = Tr(A)

d
J(t, a) = ∇ · (V )(t, X(t, a))J(T, a)
dt Z t 
J(t, a) = J(0, a) exp ∇ · (V )(s, X(s, a))ds .
0

Since the Jacobian is never zero, the Inverse function theorem shows that the inverse of
a 7→ X(t, a) is a C ∞ -diffeomorphism.
9
2.2. Theorem
The n -dimensional linear transport equation has the unique solution f0 (Xt−1 (x)).
Proof ) The direct calculation shows that the equation has a solution. Its uniqueness follows
from the characteristic method or the energy method.

Hence the first problem was simply solved considering the obvious n -dimensional general-
ization.

Nonlinear Transport Equation

2.3. Definition (Burger’s Equation in R )
Consider the equation
(
∂t v + v∂x v = 0
v(0, x) = v0 (x),

given v0 ∈ C ∞ (R)

The most common way to show the existence of the solution in the ODE is to use the
iteration. What it means by iteration is to perform below calculations
y (0) (t) ≡ y0
Z t
(1)
y (t) ≡ y0 + F (s, y(s))ds
0
..
.
Z t
(n+1)
y (t) ≡ y0 + F (s, y (n) (s))ds.
0

Then
Z t
|(yn+1 − yn )(t)| ≤ F (s, y (n) (s)) − F (s, y (n−1) (s)) ds
0
Z tZ t
≤ sup |∂y F | y (n) (s) − y (n−1) (s) ds.
x,t 0 0
 
1
If 0 ≤ t ≤ 2 supx,t |∂y F |
= T , then

sup y (n+1) − y (n) (t) ≤ sup y (n) − y (n−1) (s) · sup |∂y F | · T
0≤t≤T 0≤s≤T x,t
n n−1
sup d (t) ≤ sup d (t) · · ·
0≤t≤T 0≤t≤T

2.4. Example
Consider the standard Picard’s iteration from ODE.
(
Ẏ = f (Y )
.
Y (0) = y0

The Picard’s iteration takes 0 < T and recursively defines

Z t
(n+1)
Y (t) := y0 + f (Y (n) (s))ds.
0
10
 
If T is not too large, then Y (n) (t) is Cauchy sequence in C[0, T ] and see the convergence.
However, if T is not small, then f (x) = x2 ,
y0
Y (t) =
1 − ty0
shows that Y → ∞.

If applied to the nonlinear transport equation, let v ( 0)(t, x) = v0 (x) and

Z t
(n)
v (t, x) = v0 (x) − v (0) ∂x v(0)
0

(n)
and this never converges in C ∞ . This is because v (n+1) contains the (∂x v0 ) and there is no
norm in C ∞ that bounds the iteration. If one chooses the norm in C k , then the iteration term
with −∂x u actually requires higher norms. Moreover, there are smooth functions whose n th
deriatives grow arbitrarily large.

Euler-Lagrangian Method
There are two alternatives for the solution; first is to use the iteration with solving trajectories.
That is, to use iteration while finding v1 such that
v (0) (t, x) = v0 (x)
(
∂t v (1) + v (0) ∂x v (1) = 0
.
v (1) (0, x) = v0 (x)

Hence given v(n), we aim to find

(
∂t v (n+1) + v (n) ∂x v (n+1) = 0
.
v (n+1) (0, x) = v0 (x)

However, extra caution is required for nonlinear cases since the solution v might blow up in
infinite time interval or even for finite time interval as can be seen in the example ẏ = y12 .
2.5. Theorem
Let v0 ∈ C ∞ (R). Define A = supx∈R |∂x v0 (x)| = ∥v0 ∥Ċ 1 (R) . Then, take 0 < T < A1 .
There is a unique
solution to (∗) where v(t, x) ∈ C ∞ ([−T, T ] × R) i.e. there is a solution in
∞
Cloc ( −1
A A
, 1 × R) .
Proof ) (1) Uniqueness.
Assume that ui (t, x), i = 1, 2 decay rapidly as |x| → ∞. Define
Z
E(t) = |u1 (t, x) − u2 (t, x)|2 dx.
R

Then
Z
d
E=2 (u1 − u2 ) (−u1 ∂x u1 + u2 ∂x u2 )
dt R
Z
u1
(u1 − u2 ) (−u1 ∂x (u1 − u2 ) − ∂x u2 (u1 − u2 )) = − ∂x (u1 − u2 )2


=2
2
ZR  
∂x u1 ∂x u1
=2 − ∂x u2 (u1 − u2 )2 dx ≤ 2 sup − ∂x u2 E(t)
R 2 x,t 2
11
Hence E(t) = 0 for t > 0 since E(0) = 0. This extends to local energy method.
(2) Existence.
Consider the
d (n)
X (t, a) = v (n) (t, X (n) (t, a))
dt
X (n) (0, a) = a
(n)−1
and v (n+1) (t, x) = v0 (Xt (x)). Then v (n+1) is the unique soution to
∂t v (n+1) = −v (n) ∂x v (n+1)
v (n+1) (0, x) = v0 (x).
  (n)
In this setting, first we find the uniform bound on v (n) (t, x) n≥0
. Then v (t, x) n≥0
is a
Cauchy sequence if we restrict time appropriately.
(a) Uniform bound on v (m) . 
(n)−1
Recall that v (n+1) (t, x) = v0 Xt (x) is equivalent to
(
∂t v (n+1) + v (n) ∂x v (n+1) = 0
.
v (n+1) (0, x) = v0 (x)

Then for every t, v (n+1) (t, ·) C(R)

= ∥v0 ∥C(R) . Hence,

sup sup v (n) (t, ·) C(R)

= ∥v0 ∥C(R) .
n≥0 t∈R

(b) Uniform bound on ∂x v (m) .

Substituting ∂x v (m) into the PDE, we obtain
∂t (∂x v (n+1) ) + v (n) ∂x ∂x v (n+1) = −∂x v (n) ∂x v (n+1)

d 
∂x v (n+1) (t, X (n) (t, a)) = −∂x v (n) ∂x v (n+1) .




dt
For some T > 0, the hypothesis gives that
sup ∂x v (n) (t, ·) C(R)
≤ M (M > A).
0≤t≤T

Then as long as 0 ≤ t ≤ T ′ ,
d
∂x v (n+1) (t, X (n) (t, a)) ≤ M ∂x v (n+1) (t, X (n) (t, a))


dt
using that d 2
dt
f = 2|f |f ′ = 2gf 2 = 2g|f ||f |. Then we arrive to the conclusion that

∂x v (n+1) (t, X (n) (t, a)) ≤ ∂x v (n+1) (0, X (n) (0, a)) eM t
= |∂x v0 (a)|eM t ≤ AeM t .
′
Now we can take supremum over all a and 0 ≤ t ≤ T ′ . If we take T ′ so that AeM T ≤ M ,
then
sup sup ∂x v (n) (t, · · · ) C(R)
≤ M.
n≥0 0≤t≤T ′

(c) The bound extends to ∂x2 v (n) .

We claim that
sup sup ∂x2 v (n) (t, · · · ) C(R)
≤ M2 .
n≥0 0≤t≤T ′
12
By the definition,
∂t v (n+1) = −v (n) ∂x v (n+1)
∂t ∂x2 v (n+1) = −∂x2 (v (n) ∂x v (n+1) ) = −∂x2 v (n) ∂x v (n+1) − 2∂x v (n) ∂x2 v (n+1) − v (n) ∂x (∂x2 v (n+1) )
Then
d  2 (n+1)
)(t, X (n) (t, x)) = (−∂x2 v (n) ∂x v (n+1) − 2∂x v (n) ∂x2 v (n+1) )(t, X (n) (t, x))

(∂x v
dt
and
∂t ∂xk v (n+1) = −∂xk (v (n) ∂x v (n+1) )
= −k∂x v (n) ∂xk v (n+1) − · · · .

The final display shows that the the partial differentials grow exponentially. Choose
sup ∂x2 v (n) (t, x) ≤ 2A2 e2M t
x∈A

for all 0 ≤ t ≤ T ′ . The key issue is to take T ′ independent of k. Let A2 := supx∈R |∂x2 v0 (x)|.
Then
d
|f2 | ≤ 2A2 e2M t M + 2M |f2 |
dt
d −2M t
(e |f2 |) = e−2M t · 2A2 e2M t M, and
dt
e−2M t |f2 | ≤ e−2M ·0 A2 + 2M A2 t ≤ A2 + 2M A2 T ′ = (1 + 2M T ′ )A2 ≤ 2A2 .

Now change
 (n) 2 to k for general k-th step.
(d) v n≥0
is Cauchy in C([0, T ′ ] × R).
Direct calculation gives that subtraction of two equations leads to
(
∂t v (n+1) = −v (n) ∂x v (n+1)
.
∂t v (n) = −v (n−1) ∂x v (n)

Then
(n+1)
∂t (v (n)−v ) = −v (n−1) ∂x (v (n) − v (n+1) ) − ∂x v (n+1) v (n) − v (n−1) .

d (n)
(v − v (n+1) )(t, X (n−1) (t, a)) = −∂x v (n+1) (v (n) − v (n−1) ) (t, X (n−1) (t, a))


dt
Let dn to be sup0≤t≤T ′ (v (n) − v (n−1) )(T, ·) C(R)
. Then

d (n)
v − v (n+1) ≤ M dn , 0 ≤ t ≤ T ′
dt
v (n) − v (n+1) (t, ·)C(R) ≤ M dn t ≤ M T ′ dn
dn+1 ≤ M T ′ dn ≤ (M T ′ )2 dn−1 ≤ · · · (M T ′ )n d1 .

If T ′ < M1 , then v (n) is a Cauchy sequence in C .



(e) v (n) is Cauchy in C 1 ([0, T ′ ] × R) and

 hence inductively, C k for all k > 0 . 
The above argument is equivalent to that ∂x v (n) is Cauchy in C([0, T ′ ]×R) and ∂t v (n) n
is Cauchy in C([0, T ′ ] × R).
∂t v (n) = −v (n−1) ∂x v (n)
13
 is uniformly bounded. Moreover, the second derivative shows that
∂t (∂t v (n) ) = − ∂t v (n−1) ∂x v (n) − v (n−1) ∂x (∂t v (n) )

= +(v (n−2) ∂x v (n−1) )∂x v (n) + v (n−1) ∂x v (n−1) ∂x v (n−1)

is uniformly bounded giving the Cauchy property.

(f) Convergence gives the solution. Simply taking n → ∞, we see that the integral equation
converges to our desired one; since v (n) → v in C 1 ([0, T ′ ] × R),
∂t v (n+1) + v (n) ∂x v (n+1) = 0 → ∂t v + v∂x v = 0.

Propagation of Regularity
Changing the viewpoint, we iterately apply the theorem again; our PDE itself does not
depend on the time variable t ∈ [0, T ] so we can continue the solution by setting t′ = t − T . In
our PDE, the maximum time continuation is
1
T∗ = .
supx∈R |∂x v0 (x)|
2.6. Theorem
For Burger’s equation, let v0 ∈ C ∞ (R) and let v be a solution in C 1 ([0, T ] × R) with v(0, x) =
v0 (x). Then v ∈ C k ([0, T ] × R).
Proof )

2.7. Remark
We proved existence of C 1 solution v using v0 ∈ C 2 . How can we get the existence just by
v0 ∈ C 1 ?

The C 1 is called the critical regularity.

2.8. Example (Eikonal equation)
Consider the following: ∂t u + (∂x u)2 = 0. In this case, C 2 is the critical regularity.

2.9. Example
Consider the viscous Burgers equation:
∂t u + u∂x u = ∂xx u =
0
The above equation has C as the critical regularity.

The idea behind the critical regularity is that equation for higher derivatives is linear in
itself; that is, with our Burger’s equation,
(
∂t v + v∂x v = 0
∥v(t, x)∥C([0,T ]×R) ≤ ∥v0 (x)∥C(R) .
Then ∂t ∂x v = −v∂x (∂x v) − (∂x v)2 and dtd (∂x v)(t, X) = −(∂x v)2 (t, X) is nonlinear but becomes
linear when differentiated multiple times.

Energy Method for Existence; Compactness

Let ∂t u + V ∂x u = 0 but V can depend on u. That is,
(
∂t v + v∂x v = 0
.
v(0, x) = v0 (x) x ∈ R
14
 The energy method is often called as Eulerian method and directly relates with the Navier
Stokes.

15