SF2521NPDE lecture2AK

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 66

SF2521: Numerical Solutions of Partial Di↵erential Equations

Lecture 2: Wellposedness, Hyperbolic equations, characteristics, and


discontinuous solutions
Spring Semester 2023

Jennifer K Ryan (jryan@kth.se)

Slides by Anna-Karin Tornberg


Today’s Lecture

I Wellposedness of partial di↵erential equations

I Hyperbolic Equations
• First order equations
• Characteristics & Examples
• Linear Riemann problems

1/7
Wellposedness

2/7
Boundary conditions and initial conditions
Discussed problems follow a pattern:

I elliptic PDEs are coupled with boundary conditions;

I parabolic PDEs require initial conditions and boundary conditions for


all times.
I hyperbolic PDEs require initial conditions and boundary conditions
typically for some part of the boundary
Well-Posed Problems
A problem is called well-posed if
a.) it has a solution
b.) the solution is unique
c.) the solution depends continuously on data and parameters.

• meaning of a) is clear.
• “uniqueness” typically means “unique within a certain class of
functions”.
Example.: a problem might have several solutions, only one of which
is bounded. We’d say: solution is unique in the space of bounded
functions.
• A solution depends continuously on data and parameters if “small”
changes in initial or boundary values (in appropriate norms) and in
parameter values result in “small” changes in the solution (in some
appropriate norm).
Well-Posed Problems
I Notion of well-posedness is important in applied math.

I If you were using an initial-boundary value problem (P) to make


predictions about some physical process, you’d obviously like (P) to
have solution.
I You’d also want the solution to be unique.

I If solution depends continuously on data and parameters, you don’t


have to worry about small errors in measurement producing large
errors in your predictions.
Notation
I Let k · k denote the norm on a linear space V .
I For a function u(x, t) (in space x and time t) we write

ku(t)k := ku(·, t)k

for the norm at fixed time t.


I Example:
Let B ⇢ Rd be a domain. The L1 (B)-norm of u at time t is
Z
ku(t)k := |u(x, t)| dx.
B
Example 1 & 2
Let 0 < " ⌧ 1 and k · k1 be the maximum norm on C 0 (R).

We seek u" (x, t) with

Example 1: @tt u" + @xx u" = 0 for x 2 R and t > 0,

Example 2: @tt u" @xx u" = 0 for x 2 R and t > 0,

and initial conditions


⇣x ⌘
u" (x, 0) = 0 and @t u" (x, 0) = " sin .
"
Example 1
x
With initial conditions u" (x, 0) = 0 and @t u" (x, 0) = " sin " , the
solution is:
⇣x ⌘ ⇣t ⌘
u" (x, t) = "2 sin sinh .
" "
For " = 0, u0 ⌘ 0.
The solutions u0 and u" only differ in the choice of the initial value with

k@t u" (0) @t u0 (0)k1 = ".

But:
⇣t ⌘
ku" (t) u0 (t)k1 = "2 sinh !1 for " ! 0.
"
Since sinh(t/") = 12 (e t/" e t/"
), the error is exponentially large.
Well-Posedness of Example 1?
The solutions u0 and u" only differ in the choice of the initial value with

k@t u" (0) @t u0 (0)k1 = ".

Exponentially large error for any time t > 0

"2 t/" t/"


ku" (t) u0 (t)k1 = e e !1 for " ! 0.
2
I Very small change in initial value results in large change in solution
for positive time.
I Explanation:
• considered problem is elliptic, which naturally describes stationary
equilibriums.
• Pairing of elliptic equation with initial conditions led to an ill-posed
problem.
Example 2
x
With initial conditions u" (x, 0) = 0 and @t u" (x, 0) = " sin " , the
solution is:
⇣x ⌘ ⇣t ⌘
u" (x, t) = "2 sin sin .
" "
For " = 0, u0 ⌘ 0.
The solutions u0 and u" only differ in the choice of the initial value with

k@t u" (0) @t u0 (0)k1 = ".

We have u0 ⌘ 0 and
⇣t ⌘
ku" (t) u0 (t)k1 = "2 sin  "2 ! 0 for " ! 0.
"
Small change in initial data ) small change of solution for t > 0.

This problem is well-posed. The equation is hyperbolic.


Showing Well-Posedness
I Establishing existence can be quite difficult.

I We will do this only with simple problems for which one can write
down a solution.

I Proving uniqueness and continuous dependency is usually easier,


especially for linear problems.
First Order Hyperbolic Equations
@q
+ r · f (q) = 0
@t

3/7
Hyperbolic equations of first order
In the following, we focus on
Hyperbolic first order conservation laws in 1d.
Here, for some smooth flux function f : R ! R we seek

u = u(x, t) : R ⇥ R+ ! R

such that

@t u + @x f (u) = 0 and u(x, 0) = v (x)

for x 2 R and t 2 (0, 1).


Hyperbolic equations
Example 1: Linear advection: f is linear with

f (u) = a u

From the general form @t u + @x f (u) = 0 we obtain

@t u + a @x u = 0

Solution u(x, t) = v (x at).


Initial profile v propagates with (constant) speed:
Hyperbolic equations
Example 2: Burger’s equation.
f describes non-linear advection with
1 2
f (u) = u
2
From the general form @t u + @x f (u) = 0 we obtain

@t u + u @x u = 0

Smooth initial profile can develop discontinuity.


Second order hyperbolic equations
I Second order hyperbolic scalar PDE can be written as first order
system.
I Example: Second order wave equation of the classical form (c
constant)
ptt = c 2 pxx
[NOTES]
Characteristics

4/7
Characteristics
I Characteristics is an important tool to understand Conservation Laws.
I A characteristic is a line/curve in the x t plane along which the solution
is constant.
Example: Linear advection
I Linear advection with constant speed - the simplest example.

@t u + a@x u = 0,
u(x, 0) = v (x),
u(x, t) = v (x at)

Sol. const. along


Characteristic (t) = x0 + at
0
(t) = a. with value v (x0 ).
Characteristics for linear advection - variable speed
@t u + a(x)@x u = 0, x 2 R, t > 0
u(x, 0) = v (x),

I Linear advection with variable coefficient, a = a(x).


I Define a characteristic curve: ( (t), t) in the x t plane, where

@
= a( (t)), (0) = x0 .
@t
I Then u( (t), t) = v (x0 ),
@
since @t u( (t), t) = 0 (show by chain rule) and u( (0), 0) = v (x0 ).

I Note: the characteristics are not straight lines unless a is constant.


I Note: this is not a conservation law if a = a(x).
Characteristics for conservation laws
Consider a solution u 2 C 1 (R ⇥ [0, 1)) to

@t u + @x f (u) = 0 and u(x, 0) = v (x)

where v 2 C 1 (R) and f 2 C 2 (R).

Definition
For every x0 2 R there is a maximum time T > 0 such that (by Picard-Lindelöf
theorem) there is a unique solution to
0
(t) = f 0 u( (t), t) for t 2 (0, T ),
(0) = x0 .

The curve {( (t), t) | t 2 [0, T ]} is called a Characteristic of the Conservation


law.
Definition and properties
(characteristics for conservation laws)

Definition
For x0 2 R and maximum time T > 0, let solve
0
(t) = f 0 u( (t), t) for t 2 (0, T ),
(0) = x0 .

The curve {( (t), t) | t 2 [0, T ]} is called a Characteristic.

General properties:
I The map t 7! u( (t), t) is always constant on [0, T ].
I The function has the form

(t) = f 0 v (x0 ) t + x0 .

Hence, characteristics are straight lines for all flux functions f.


Next, we prove these properties.
Characteristics - Proof: t 7! u( (t), t) is constant.
We have

@t u + @x f (u)= 0
0
(t)= f 0 u( (t), t)
Characteristics - Proof: (t) = f 0 v (x0 ) t + x0 .
Characteristics - Summary
Definition
For x0 2 R and maximum time T > 0, let solve
0
(t) = f 0 u( (t), t) for t 2 (0, T ),
(0) = x0 .

The curve {( (t), t) | t 2 [0, T ]} is called a Characteristic.

General properties:
I The map t 7! u( (t), t) is always constant on [0, T ].
I The function has the form (t) = f 0 v (x0 ) t + x0 .
Why characteristics?
1. Understanding solutions.
2. Constructing solutions.
Characteristics - Understanding solutions
Some observations:

I the characteristic is explicitly known and solving is not necessary

(t) = f 0 v (x0 ) t + x0 ;

I we know u is constant on ( (t), t).

I we only know that a characteristic exists (uniquely) for some


maximum time T ;

I in general, we do not know the value of T ;

I what does that mean for u(t) for t T?


Examples

5/7
Characteristics of Burger’s equation
Example: Understanding Burger’s equation, where f (u) = 12 u 2 :

@t u + u @x u = 0.

For the initial value we assume v 2 C 1 (R) and there are x1 and x2 with

x1 < x2 and v (x1 ) > v (x2 ).

Since f (u) = 12 u 2 we have f 0 (u) = u and hence

1 (t) = v (x1 ) · t + x1 and 2 (t) = v (x2 ) · t + x2 .

What will happen?


Characteristics - understanding solutions
We have

1 (t) = v (x1 ) · t + x1 and 2 (t) = v (x2 ) · t + x2 .

x1 x2
For t0 = v (x1 ) v (x2 ) > 0 we have 1 (t0 ) = 2 (t0 ) and hence

v (x1 ) = u( 1 (t0 ), t0 ) = u( 2 (t0 ), t0 ) = v (x2 )

Since v (x2 ) < v (x1 ) by initial assumption, we have a contradiction.

What does that imply?

For t t0 there exists no classical solution u.


Intersection of characteristics
Intersection of characteristics! Observe
I Physical interpretation: 1 (t) = v (x1 ) t + x1
| {z }
Wave speed

Wave speed higher the larger the value of v (x1 ).


I Particles in the wave peak are fastest
I wave must eventually develop a discontinuity ) “Shock waves”.
I if v (x1 )  v (x2 ) we have t0 < 0 and no contradiction;
Characteristics - smooth initial profile
Example: v (x) = sin(⇡x). It holds

(t) = sin(⇡x0 ) · t + x0 .

Characteristics will cross ) discontinuous solution for t t0 .


Characteristics - Understanding solutions
The examples shows the following result:
Theorem
Conservation laws do not necessarily have classical solutions for all times
t 2 (0, 1). Not even for very regular initial data.
Characteristics: Example 2, Linear advective systems.
Notation:
For u = u(x, t) : R ⇥ [0, 1) ! Rm we denote
0 1 0 1
@t u1 (x, t) @x u1 (x, t)
B
@t u(x, t) := @ .. C
and
B
@x u(x, t) := @ .. C
. A . A.
@t um (x, t) @x um (x, t)
System of linear hyperbolic equations
We consider a system of linear hyperbolic equations.
Find u = u(x, t) : R ⇥ [0, 1) ! Rm with

@t u + A @x u = 0 and u(x, 0) = v(x).

where v = v(x) : R ! Rm .

If the system is hyperbolic we have


I A 2 Rm⇥m is diagonalizable with real non-zero eigenvalues
1  . . .  m;
Notation
If the system is hyperbolic we have
I A 2 Rm⇥m is diagonalizable with real non-zero eigenvalues
1  . . .  m;

We introduce
I the corresponding diagonal matrix:

⇤:= diag( 1, · · · , m );

I eigenvectors: r1 , · · · , rm 2 Rm ; i.e. Arp = p rp for 1  p  m.


I matrix with eigenvectors as columns R := [r1 , . . . , rm ].

and we have
I A = R⇤R 1
.
Derive decoupled system

@t u + A @x u = 0
Characteristics of decoupled linear advective system
1
We get the following system (z = R u)

@t z + ⇤ @x z = 0.

Since ⇤ is a diagonal matrix, we get m decoupled scalar equations.


Hence for p = 1, . . . , m
1
@ t zp + p @ x zp = 0 and zp (0) = (R v)p .

The corresponding characteristic for x0 2 R is given by

p (t) = p t + x0 .

We call p the p-characteristic of the system.


Solution of decoupled system
1
Since @ t zp + p @ x zp = 0 and zp (0) = (R v)p
and since p is the characteristic for x0 2 R with

p (t) = p t + x0 ,

zp is constant along ( p (t), t) and hence with x = p t + x0

zp (x, t) =
Solution of original linear advective system
1
We have @ t zp + p @ x zp = 0 and zp (0) = (R v)p
p = 1, . . . , m, and
1 1
zp (x, t) = zp (x0 , 0) = (R v(x0 ))p = (R v(x p t))p =: z0p (x p t).

and
m
X m
X
u = Rz = zp (x, t)rp = z0p (x p t) rp
p=1 p=1

Hence
I solution u is given as linear combination (superposition) of zp ;
I superposition of waves propagating with speed p , p = 1, . . . , m.
Characteristics: Linear systems. Example: m = 3
t
p (t) = p t + x0 .
x0 + 1t
x0 + Range of influence of x0 :
2t
All points along p (t) = p t + x0
for p = 1, 2, 3.
x0 + 3t “Information” in x0 spreads
along the red lines.
x0 x
t
x⇤
Domain of dependence of x ⇤ at time T :
Set of points
x⇤ 1t
D(x ⇤ , T ) := {x ⇤ p T | p = 1, 2, 3}.

x⇤ 3t x⇤ 2t Solution in (x ⇤ , T ) is given as linear


x
combination of z0p in x ⇤ p T.
Linear system, explicit example
Find u = u(x, t) : R ⇥ [0, 1) ! R2 with

@t u + A @x u = 0 and u(x, 0) = v(x).

Here, for c > 0


✓ ◆ ✓ ◆ ✓ ◆
u1 0 0 c2
u= and v(x) = and A=
u2 v0 (x) 1 0

For the eigenvalues we have 1 = c and 2 =c


✓ ◆✓ ◆ ✓ ◆ ✓ ◆✓ ◆ ✓ ◆
0 c2 1 1 0 c2 1 1
= 1 and = 2
1 0 c 1 c 1 1 0 c 1 c 1
| {z } | {z }
=r1 =r2
Linear system, explicit example contd.
We have
✓ ◆ ✓ ◆ ✓ ◆
c 0 1 1 1 1 1 c
⇤ := , R := 1 1 , R := .
0 c c c 2 1 c

From @t u + A @x u = 0 and using the transformation u = Rz we have

@t z + ⇤ @x z = 0.

z(0, x) =
Linear system, explicit example contd.
From
✓ ◆
c 1
@t z + ⇤ @x z = 0; z(x, 0) = v (x)
2 1 0

and the p-characteristics

1 (t) = c t + x0 and 2 (t) = c t + x0 ,

we conclude that
✓ ◆ ✓ ◆
z(x + ct, 0) c v0 (x + ct)
z(x, t) = =
z(x ct, 0) 2 v0 (x ct)

and hence
✓ ◆✓ ◆ ✓ ◆
c 1 1 v0 (x + ct) 1 cv0 (x + ct) cv0 (x ct)
u(x, t) = Rz(x, t) = 1 1 = .
2 c c
v0 (x ct) 2 v0 (x + ct) + v0 (x ct)
Linear system, explicit example contd.
✓ ◆ ✓ ◆✓ ◆ ✓ ◆ ✓ ◆
u1 0 c2 u1 u1 (x, 0) 0
+ =0 and = .
u2 t 1 0 u2 x u2 (x, 0) v0 (x)

c 1
u1 (x, t) = (v0 (x + ct) v0 (x ct)) and u2 (x, t) = (v0 (x + ct) + v0 (x ct)).
2 2

(t) = x0 ± ct.

u2 (x, t)

If v0 (x) = 2 sin(2⇡x) we have u2 (x, t) = sin(2⇡(x + ct)) + sin(2⇡(x ct)).


At T = 1 and for c = 1 we have u2 (0, T ) = sin(2⇡) + sin( 2⇡) = 0.
Linear Riemann Problems

6/7
The Riemann-problem - Scalar case
I Riemann-problem: conservation law with discontinuous initial
condition.
I In the following we consider the linear Riemann-problem
Scalar case: for a 2 R, find u : R ⇥ [0, 1) ! R with
(
ul for x  0
@t u + a@x u = 0 and u(x, 0) = v (x) :=
ur for x > 0.

Formal issue:
v is discontinuous ) there exists no continuous (classical) solution u.
However: we know how a physically correct solution looks like:
initial value v moves with speed a, i.e.
u is constant along the characteristic (t) = x0 + a t. Hence

u(x, t) = v (x at).
The Riemann-problem - Scalar case
Scalar case: for a 2 R, find u : R ⇥ [0, 1) ! R with
(
ul for x  0
@t u + a@x u = 0 and u(x, 0) = v (x) :=
ur for x > 0.

t x = at
ul x at < 0 ul
u = ul
ur u = ur ur
x at > 0

x x aT x

Initial value v . Characteristic Solution u(x, T ).


(t) = x0 + at
for x0 = 0.

How does this generalize to systems?


The Riemann-problem - Systems
System of conservation laws:
for diagonalizable A 2 Rm⇥m , find u : R ⇥ [0, 1) ! Rm with
(
ul for x  0
@t u + A @x u = 0 and u(x, 0) =
ur for x > 0

Here: ul , ur 2 Rm . Solution in characteristic variables:


1
z=R u,

which solves
(
1
1 R ul for x  0
@t z + ⇤ @x z = 0 and z(x, 0) = R v(x) := 1
R ur for x > 0.
Remember the notation
If the system is hyperbolic we have
I A 2 Rm⇥m is diagonalizable with real non-zero eigenvalues
1  . . .  m;

We introduce
I the corresponding diagonal matrix:

⇤:= diag( 1, · · · , m );

I eigenvectors: r1 , · · · , rm 2 Rm ; i.e. Arp = p rp for 1  p  m.


I matrix with eigenvectors as columns R := [r1 , . . . , rm ].

and we have
I A = R⇤R 1
.
The Riemann-problem - solution with characteristic variables
The system
(
1
zl := R ul for x  0
@t z + ⇤ @x z = 0 and z(x, 0) = 1
zr := R ur for x > 0

decouples into 1  p  m scalar equations with


(
zpl for x  0
@t zp + p @x zp = 0 and zp (x, 0) =
zpr for x > 0

We have
(
zpl for x pt 0
zp (x, t) = zp (x p t, 0) =
zpr for x p > 0
t

Solution:
X X
u(x, t) = Rz = zpl rp + zpr rp
p with p with
x p t0 x p t>0
The Riemann-problem - solution regions
X X
u(x, t) = zpl rp + zpr rp
p with p with
x p t0 x p t>0

We identify m + 1 regions.
Example: If m = 3, we have 4 regions (cases):
1. x 1t  0; x 2t  0 and x 3t  0.
2. x 1t > 0; x 2t  0 and x 3t  0.
3. x 1 t > 0; x 2 t > 0 and x 3 t  0.

4. x 1t > 0; x 2t > 0 and x 3t > 0.


(⇤) Since 1  2  3, we have

x 1t x 2t x 3t

and other cases cannot happen.


The Riemann-problem - sketch of solution regions
Example: m = 3, where 1 < 0 and 2, 3 0.

x= 1t t x= 2t

x= 3t
t=T

2
1 3
4

x
Observe: The characteristics form m + 1 = 4 regions.

X X
u(x, t) = zpl rp + zpr rp
p with p with
x p t0 x p t>0
The solution to the Riemann-problem
Example: m = 3, where 1 < 0 and 2, 3 0.

x= 1t t x= 2t

x= 3t
t=T

2
1 3
4

Region 1 : u(x, t) = z1l r1 + z2l r2 + z3l r3 ,


Region 2 : u(x, t) = z1r r1 + z2l r2 + z3l r3 .
Region 3 : u(x, t) = z1r r1 + z2r r2 + z3l r3
Region 4 : u(x, t) = z1r r1 + z2r r2 + z3r r3
Locations of discontinuities at a given time
Example: m = 3, where 1 < 0 and 2, 3 0.

x= 1t t x= 2t

x= 3t
t=T (x ⇤ , T )

x⇤ 3t <0 x⇤ 2t <0 x⇤ 1t >0 x


Observe: over each p-characteristics x = p (t) = p t, there is a jump in u.
For example: jump over 1 (t) = 1 t:

uregion 1 uregion 2 = z1l r1 + z2l r2 + z3l r3 z1r r1 z2l r2 z3l r3 = (z1l z1r ) r1 .
| {z }
jump 6=0
The Riemann-problem - Explicit example of system
Find u = u(x, t) : R ⇥ [0, 1) ! R2 with

@t u + A @x u = 0 and u(x, 0) = v(x).


>
Here, u = (u1 , u2 )
8 !
>
> 0
>
✓ ◆ <ul =
>
1
for x  0
0 4 !
A= and v(x) = .
1 0 >
> 1
>
>ur = for x > 0.
: 1

Eigenvalues: 1 2 and
= = 2.
✓ ◆ 2 ✓ ◆
2 2
Eigenvectors: r1 = and r2 = .
1 1

Note that only the first component, u1 has a discontinuity.


The Riemann-problem - Explicit example, contd.
We have
✓ ◆ ✓ ◆ ✓ ◆
2 0 2 2 1 1 1 2
⇤ := , R := , R := .
0 2 1 1 4 1 2
1
Look at the system in characteristic variables z = R u:

@ t z1 2 @ x z1 = 0 and @t z2 +2 @x z2 = 0.

We have:
8 ! ! !
>
> 1 1 2 0 1 1
> 1
<R
> ul = 4
1 2 1
= 2
1
for x  0
1 ! ! !
z(0, x) = R v(x) =
>
> 1 1 2 1 1 1
> 1
:R
> ur = 4
1 2 1
= 4
3
for x > 0.
The Riemann-problem - Explicit example, contd.
From 1 = 2, 2 = 2 and
8 !
>
> 1 1
>
> for x  0
<2 1
z(0, x) = !
>
> 1 1
>
> for x > 0.
:4 3

we conclude that the solution in characteristic variables is


( (
1 1
2
for x + 2t  0 2
for x 2t  0
z1 (t, x) = 1
and z2 (t, x) = 3
4
for x + 2t > 0 4
for x 2t > 0

From that we can construct the solution u.


Since there are 2 characteristics, we know that u has 3 stats (3 regions
with different behavior).
Explicit example, solution in all regions
Solution in characteristic variables:
( (
1 1
for x + 2t  0 for x 2t  0
z1 (t, x) = 21 and z2 (t, x) = 2
3
4
for x + 2t > 0 4
for x 2t > 0

Discontinuity of z1 Discontinuity of z2
x = 2t t x = 2t

t=T

2 z1 =z1r
z2 =z2l
1 z1 =z1l z1 =z1r 3
z2 =z2l z2 =z2r

✓ 1◆ ✓ ◆✓ ◆✓ ◆ x
z 2 2 1/2 0
1 u(x, t)= R l2 = = = ul for x + 2t < 0.
zl 1 1 1/2 1
✓ 1◆ ✓ ◆✓ ◆ ✓ ◆
z 2 2 1/4 1/2
2 u(x, t)= R r2 = = for x + 2t 0 and x 2t < 0.
zl 1 1 1/2 3/4
✓ 1◆ ✓ ◆✓ ◆ ✓ ◆
z 2 2 1/4 1
3 u(x, t)= R r2 = = = ur for x 2t 0.
zr 1 1 3/4 1
Initial condition at t = 0 (u1 ).

(
0 for x < 0
u1 (x, 0) = .
1 for x 0

u1 (x, 0)

1 3

t=0
x
Solution at t = T (u1 )
We have
8
<0
> for x + 2t < 0
u1 (x, t) = 1
for x + 2t 0 and x 2t < 0 .
>2
:
1 for x 2t 0

x= 2t t x = 2t

t=T

2
1 3

x
Solution at t = T (u1 )
We have
8
<0
> for x + 2t < 0
u1 (x, t) = 1
for x + 2t 0 and x 2t < 0 .
>2
:
1 for x 2t 0

x= 2t t x = 2t

t=T

2
1 3

x
Solution at t = T (u1 )
We have
8
<0
> for x + 2t < 0
u1 (x, t) = 1
for x + 2t 0 and x 2t < 0 .
>2
:
1 for x 2t 0

u1 (x, 0)

t=T 1

2
1/2
1 3

x
Solution at t = T (u1 )
We have
8
<0
> for x + 2t < 0
u1 (x, t) = 1
for x + 2t 0 and x 2t < 0 .
>2
:
1 for x 2t 0

u1 (x, 0)

t=T 1

2
1/2
1 3

x
u2 : solution at t = T
We have
8
<1
> for x + 2t < 0
u2 (x, t) = 3
for x + 2t 0 and x 2t < 0 .
>4
:
1 for x 2t 0

Same regions as for u1 , but different values.

u2 (x, 0)

t=T 1 1
3/4

1 2 3

x
Suggested Exercises

7/7

You might also like