Lecture 13

Weston Barger

August 7, 2016

1 Continued...
Example 1.1. Let us solve the problem

 x ∈ (–∞, ∞), t ∈ (0, ∞),
ut + 4ux = 0,  u(x , 0) = arctan(x ).

The characteristic curve at the starting point (x0 , 0) is found by solving

dx (t )
= 4, x (0) = x0 .
dt
We get

x (t ) = x0 + 4t .

Since u is constant along (x (t ), t ), we have

u(x (t ), t ) = u(x (0), 0) = u0 (x0 ) = arctan(x0 ).

Thus,

u(x , t ) = arctan(x – 4t ).

1.1 Nonhomoegeneous advection equation

Assume now that we would like to solve

 x ∈ (–∞, ∞), t ∈ (0, ∞),
ut + cux = f , (1.1)
 u(x , 0) = u0 (x ),

1
the nonhomogeneous advection equation. We can still use characteristics to solve (1.1).
We still want to choose x (t ) to satisfy

dx (t )
=c
dt
However, this time we have
!
d dx (t )
û(t ) = ut + ux
dt dt
= ut + cux
=f.

The value of u along (x (t ), t ) is not constant this time, but satisfies the ODE

û 0 (t ) = f (x (t ), t ) = fˆ(t ), û(0) = u(x (0), 0) = u0 (x0 ).

Let us make this more clear with an example.

Example 1.2. Let us solve the problem


 x ∈ (–∞, ∞), t ∈ (0, ∞),
ut + 4ux = 1,
 u(x , 0) = arctan(x ).

The characteristic curve at the starting point (x0 , 0) is found by solving

dx (t )
= 4, x (0) = x0 .
dt
We get

x (t ) = x0 + 4t .

The rate of change of u along x (t ) is

d
û(t ) = ut + 4ux = 1.
dt
Integrating both sides with repect to t gives

û(t ) = t + A,

2
for some constant of integration A. To find A, note that at t = 0,

A = û(0) = u(x0 , 0)
= arctan(x0 ).

Thus,

u(x (t ), t ) = t + arctan(x0 )

i.e.

u(x , t ) = t + arctan(x – 4t ).

2 General linear conservation laws

Characteristics can also be used to solve initial value problems for linear conservation laws
in the form

 x ∈ (–∞, ∞), t ∈ (0, ∞)
ut + c(x , t )ux = 0,
 u(x , 0) = u0 (x ).
The coefficient c(x , t ) is, in general, not constant. We may still apply the same method-
ology. Choose a curve in the xt -plane which starts at x0 i.e.

(x (t ), t ), (x (0), 0) = (x0 , 0).

Suppose that we assert the IVP

dx (t )
= c(x , t ), x (0) = x0 .
dt
Then, by the chain rule
!
d û(t ) dx (t )
= ut + ux ·
dt dt
= ut + ux · (c(x , t ))
= 0.

We have found that along the curve (x (t ), t ), u is constant i.e.

û(t ) = u(x (t ), t ) = K .

As with the advection equation, the value of u at a point (x , t ) can be found by

3
 1.5

1.0

0.5

-10 -5 0 5 10 15

2
Figure 1: Characteristic curves in the xt -plane for x (t ) = x0 e t /2 for –5 ≤ x0 ≤ 5.

1. Finding the characteristic curves by solving

dx (t )
= c(x , t ), x (0) = x0 ,
dt
2. finding the particular initial point (x0 , 0) for the characteristic passing through (x , t ),

3. using x0 to compute u(x , t ) = u(x0 , 0) = u0 (x0 ).

Example 2.1. Suppose we would like to solve the initial value problem

 x ∈ (–∞, ∞), t ∈ (0, ∞),
ut + txux = 0, 1 .
u(x , 0) = 1+x

2

Step 1. Let us choose an x0 ∈ R and solve the IVP

dx (t )
= tx , x (0) = x0 .
dt
Well, using the separability of the above ODE, we get
dx t2 2
= t dt ⇒ ln |x | = +C ⇒ x (t ) = x0 e t /2 .
x 2
These curves can be seen in Figure 1.
Step 2. We now need to find which x0 to associate with (x , t ) i.e. which characteristic
curve passes through the point (x , t ). Well,
2
xe –t /2 = x0 .

Step 3. Since u is constant on characteristic curves,

2
2 1 et
u(x , t ) = u(x0 , 0) = u(xe –t /2 , 0) = 2 = 2 .
1 + x 2 e –t et + x 2

4
3 Nonlinear conservation laws
Let us suppose that we have a conservation law in PDE form and a constitutive relation
i.e

 x ∈ (–∞, ∞), t ∈ (0, ∞)
ut + c(u)ux = 0, (3.1)
 u(x , 0) = u0 (x ),

where c(u) = φ0 (u). We will attack this problem using characteristics. Choose an x0 ∈ R.
We need to solve the ODE
dx
= c(u(x , t )), x (0) = x0 . (3.2)
dt
This doesn’t really seem to help us since our ODE for x (t ) involves u(x , t ), the very func-
tion which we are trying to determine. However, since the PDE in (3.1) is homogeneous,
if we were able to solve (3.2) we would have that u is constant on x (t ). This is because
(once again)

d û(t )
= ut (x (t ), t ) + x 0 (t )ux (x (t ), t )
dt
= ut (x (t ), t ) + (c(u(x , t ))) ux (x (t ), t )
= 0.

Therefore, since u(x (t ), t ) = u(x (0), 0), we can consider the ODE

dx
= c(u(x (0), 0)) = c(u0 (x0 )) x (0) = x0 . (3.3)
dt
We make the following important observation: c(u0 (x0 )) is constant in time. Therefore,
we can easily solve (3.3) to get

x (t ) = c(u0 (x0 ))t + x0 .

Like the advection equation, the characteristics are lines. Unlike the advection equation,
the characteristic lines are not necessarily parallel. The slope of each line is
1
.
c(u0 (x0 ))

The procedure of finding the value of u at a point xt is now:

5
 1. Construct the characteristics x = c(u0 (x0 ))t + x0 using the speed c(u) from ut +
c(u)ux and the initial profile u0 (x0 )

2. Find the initial point (x0 , 0) of the characteristic passing through (x , t ) by solving
x = c(u0 (x0 ))t + x0 . This step often must be done numerically.

3. Use x0 to calculate u(x , t ) = u(x0 , 0) = u0 (x0 ).

Example 3.1. If φ = 21 u 2 , then the conservation law ut + φx = 0 takes the form of the
inviscid Burgers equation with characteristic speed c(u) = u. Let us consider the initial
value problem



 x ∈ (–∞, ∞), t ∈ (0, ∞),
 
ut + uux = 0,  0 x ≤0
 u(x , 0) =
e –1/x x > 0


 

Since c(u0 (x0 )) = u0 (x0 ), we have

x (t ) = u0 (x0 )t + x0 .

The characteristic lines for the is IVP are thus


 x0 x ≤0
x (t ) =
 te –1/x0 + x0 x0 > 0.

We see the characteristics plotted in Figure 2.

When x < 0, we see that the characteristic lines are vertical. Therefore, for x < 0,
x = x0 and thus

u(x , t ) = u(x0 , 0) = u0 (x ) = 0.

For x > 0, the value of u at the point (x , t ) is found from u(x , t ) = u(x0 , 0), where x0 is
the solution to the equation

x = te –1/x0 + x0 .

This cannot be done explicitly, so we cannot get a closed form expression of u by using
characteristics. However, we can find x0 numerically quickly.

6
Figure 2: Here we have the characteristic lines for Burger’s equation with u0 (x ) =
e –1/x Ix >0 .

4 Gradient Catastrophes and Breaking Times

(Chapter 18 of the text)

In the last section, it was shown that the characteristic curves of the initial value
problem

 x ∈ (–∞, ∞), t ∈ (0, ∞)
ut + c(u)ux = 0,
 u(x , 0) = u0 (x ).

are the lines x = c(u0 (x0 ))t + x0 . Along such lines u is constant.
In the special case where c(u) is constant, the characteristic lines are parallel x =
ct + x0 , with slope 1/c. When c(u) is not constant, the characteristic lines are not
necessarily parallel and may cross. This is illustrated in Figure 3.
The value of u is nevertheless the same along characteristic lines. As we approach an
intersection point, ux (x , t ) becomes larger and larger. At the intersection point, u takes
multiple values. Such an event is called a gradient catastrophe.

Example 4.1. Let us consider the problem


 x ∈ (–∞, ∞), t ∈ (0, ∞),
ut + uux = 0, 2 (4.1)
 u(x , 0) = e –x .

7
 Figure 3: Characteristic lines crossing
2.0

1.5

1.0

0.5

-1 0 1 2

Figure 4: The characteristic lines for (4.1).

2
Here, c(u) = u, and u0 (x ) = e –x . We see that the characteristic lines are
2
x = c(u0 (x0 ))t + x0 = e –x0 t + x0 .

A diagram of the characteristics is given in Figure 4.

We see from Figures 5 and 6 what happens to the wave profile as time moves forward.
It is seen that at some time around tb = 1.2 that u fails to be a function.

4.1 Breaking Time

The earliest time tb ≥ 0 at which a breaking catastrophe occurs is called a breaking time.
We will discuss how tb can be computed by calculating ux (x , t ) and finding the first time
at which ux becomes infinite.
By the method of characteristics, the value of u which satisfies

 x ∈ (–∞, ∞), t ∈ (0, ∞),
ut + c(u)ux = 0,
 u(x , 0) = u0 (x ),

8
Figure 5:

Figure 6:

9
at the point (x , t ) is u(x , t ) = u0 (x0 ), where x0 = x0 (x , t ) determines the starting point
(x0 , 0) of the characteristic which passes through (x , t ). We can thus compute by the chain
rule
∂x0
ux (x , t ) = u00 (x0 (x , t )) · .
∂x
Recall that for each (x , t ), the value of x0 (x , t ) is found by as the solution to

x = c(u0 (x0 ))t + x0 .

The derivative ∂x0 /∂x can be computed by implicitly differentiating the above expression.
We get
∂
1= x
∂x
∂
= [c(u0 (x0 ))t + x0 ]
∂x
d ∂x ∂x
=t [c(u0 (x0 ))] 0 + 0 .
∂x0 ∂x ∂x
Solving for ∂x0 /∂x shows that
∂x0 1
= d c(u (x ))
.
∂x 1 + t dx 0 0 0

Therefore,
u00 (x0 )
ux (x , t ) = d c(u (x ))
.
1 + t dx 0 0
0

The smallest time t = tb in which

d c(u (x )) = 0
1 + t dx 0 0
0

is our breaking time. We can solve for t as

1
t =– d .
dx c(u0 (x 0 ))
0

d c(u (x )) ≥ 0 for all x , then a breaking catastrophe never occurs. On

Note that if dx
0 0 0 0
d c(u (x )) < 0. In this case,
the other hand, a breaking time occurs if for some x0 , dx 0 0 0
making dxd c(u (x )) the smallest (i.e. finding the x which makes d c(u (x )) the most
0 0 0 0 dx0 0 0
negative) gives us our breaking time.

10
Example 4.2. Let us return to the problem

 x ∈ (–∞, ∞), t ∈ (0, ∞),
ut + uux = 0, 2
 u(x , 0) = e –x .

We see that the speed of the characteristic starting at (x0 , 0) is

 
2 2
c(u0 (x0 )) = c e –x0 = e –x0 .

Computing the breaking time requires us to find the x0 which makes

d d –x 2 2
F (x0 ) = c(u0 (x0 )) = e 0 = –2x0 e –x0
dx0 dx0

the most negative. To minimize F (x0 ), we take a derivative and set it equal to zero. Well,
2 1
F 0 (x0 ) = (–24x02 )e –x0 = 0 ⇒ x0 = ± √ .
2
We have that
! s
1 2
F ±√ =∓ .
2 e
√
Thus, x0 = 1/ 2 is our desired value. We can thus compute tb as

1 e
r
tb = – d = ≈ 1.16.
c(u0 (x 0 )) 2
dx 0

11