Fundamentals: Course Block-Structured Adaptive Finite Volume Methods in C++

Conservation laws Finite volume methods Upwind schemes References
Lecture 1
Fundamentals
Course Block-structured Adaptive Finite Volume Methods in C++
Ralf Deiterding
University of Southampton
Engineering and the Environment
Highfield Campus, Southampton SO17 1BJ, UK
E-mail: r.deiterding@soton.ac.uk
Fundamentals 1
Outline
Conservation laws
Mathematical background
Characteristic information
Weak and entropy solutions
Characteristic form of the Euler equations
Navier-Stokes equations
Fundamentals 2
Outline
Conservation laws
Finite volume methods

Basics of finite difference methods
Splitting methods, second derivatives
Fundamentals 2
Outline
Conservation laws

Upwind schemes
The linear Riemann problem
Flux-difference splitting
Flux-vector splitting
Fundamentals 2
Outline
Conservation laws

Upwind schemes
Fundamentals 3
Hyperbolic Conservation Laws
d
∂ X ∂
q(x, t) + fn (q(x, t)) = 0 , D ⊂ {(x, t) ∈ Rd × R+
0} (1)
∂t n=1
∂xn
Fundamentals 4
d
∂ X ∂
q(x, t) + fn (q(x, t)) = 0 , D ⊂ {(x, t) ∈ Rd × R+
0} (1)
∂t n=1
∂xn
q = q(x, t) ∈ S ⊂ RM - vector of state, fn (q) ∈ C1 (S, RM ) - flux functions,
Fundamentals 4
d
∂ X ∂
q(x, t) + fn (q(x, t)) = s(q(x, t)) , D ⊂ {(x, t) ∈ Rd × R+
0} (1)
∂t n=1
∂xn

s(q) ∈ C1 (S, RM ) - source term
Fundamentals 4
d
∂ X ∂
0} (1)
∂t n=1
∂xn

Definition (Hyperbolicity)
A(q, ν) = ν1 A1 (q) + · · · + νd Ad (q) with An (q) = ∂fn (q)/∂q has M real
eigenvalues λ1 (q, ν) ≤ ... ≤ λM (q, ν) and M linear independent right
eigenvectors rm (q, ν).
Fundamentals 4
d
∂ X ∂
0} (1)
∂t n=1
∂xn

Definition (Hyperbolicity)
A(q, ν) = ν1 A1 (q) + · · · + νd Ad (q) with An (q) = ∂fn (q)/∂q has M real
eigenvalues λ1 (q, ν) ≤ ... ≤ λM (q, ν) and M linear independent right
eigenvectors rm (q, ν).
If fn (q) is nonlinear, classical solutions

q(x, t) ∈ C1 (D, S) do not generally exist, not
even for q0 (x) ∈ C1 (Rd , S) [Majda, 1984],
[Godlewski and Raviart, 1996],
[Kröner, 1997]
Example: Euler equations

Fundamentals 4
Characteristic variables
Consider the first-order partial differential equation
∂q ∂q
+ A(q) =0 (2)
∂t ∂x
Fundamentals 5
∂q ∂q
+ A(q) =0 (2)
∂t ∂x
For A = const. Eq. (2) is called linear, for A = A(q(x, t)) it is called
quasi-linear. For a hyperbolic system, A is diagonalizable as
R−1 A R = Λ
Fundamentals 5
∂q ∂q
+ A(q) =0 (2)
∂t ∂x
R−1 A R = Λ
R is the matrix of right eigenvectors (column-wise)

R = ( r1 | · · · |rM )
Fundamentals 5
∂q ∂q
+ A(q) =0 (2)
∂t ∂x
R−1 A R = Λ
R is the matrix of right eigenvectors (column-wise)

R = ( r1 | · · · |rM )
and Λ the diagonal matrix of eigenvalues

 
λ1 0 0
Λ =  0 ...
 
0 
0 0 λM
Fundamentals 5
Characteristic variables - II
Multiplying (2) with R−1 gives
∂q ∂q
R−1 + R−1 A =0
∂t ∂x
Fundamentals 6
∂q ∂q
R−1 + R−1 A =0
∂t ∂x
with R−1 dq = dv this becomes
∂v ∂v
+ R−1 AR =0
∂t ∂x
or
∂v ∂v
+Λ =0
∂t ∂x
Fundamentals 6
∂q ∂q
R−1 + R−1 A =0
∂t ∂x
∂v ∂v
+ R−1 AR =0
∂t ∂x
or
∂v ∂v
+Λ =0
∂t ∂x
which is just a set of decoupled independent advection equations for the
components, i.e.,
∂vm ∂vm
+ λm =0 for m = 1, . . . , M (3)
∂t ∂x
Fundamentals 6
∂q ∂q
R−1 + R−1 A =0
∂t ∂x
∂v ∂v
+ R−1 AR =0
∂t ∂x
or
∂v ∂v
+Λ =0
∂t ∂x
which is just a set of decoupled independent advection equations for the
components, i.e.,
∂vm ∂vm
+ λm =0 for m = 1, . . . , M (3)
∂t ∂x
(3) is a wave equation but note that in the general quasi-linear case the
eigenvalues can dependent on all vm , i.e. λm = λm (v1 , · · · , vM ) Nevertheless,
an analysis as for the wave equations shows
dx
vm = const. for = λm
dt
Fundamentals 6
Wavefronts
The curves dx = λm dt are called wavefronts or characteristics, vm are the
characteristic variables.
The characteristics define how influence spreads in the x − t plane. A point in
the x − t plane is only influenced by points at earlier times in a finite domain of
dependence and influences only points in a finite range of influence.
Typical wave diagram for vector model problem.

Fundamentals 7
Weak solutions
Integral form (Gauss’s theorem):

Z Z
q(x, t + ∆t) dx − q(x, t) dx
Ω Ω
d t+∆t
X Z Z t+∆t
Z Z
+ fn (q(o, t)) σn (o) do dt = s(q(x, t)) dx
n=1 t t
∂Ω Ω
Fundamentals 8
Weak solutions

Z Z
q(x, t + ∆t) dx − q(x, t) dx
Ω Ω
d t+∆t
X Z Z t+∆t
Z Z
n=1 t t
∂Ω Ω
Theorem (Weak solution)

d
q0 ∈ L∞ ∞
loc (R , S). q ∈ Lloc (D, S) is weak solution if q satisfies
Fundamentals 8
Weak solutions

Z Z
q(x, t + ∆t) dx − q(x, t) dx
Ω Ω
d t+∆t
X Z Z t+∆t
Z Z
n=1 t t
∂Ω Ω
Theorem (Weak solution)

d
q0 ∈ L∞ ∞
loc (R , S). q ∈ Lloc (D, S) is weak solution if q satisfies
Z∞ Z " d
# Z
∂ϕ X ∂ϕ
·q+ · fn (q) − ϕ · s(q) dx dt + ϕ(x, 0)·q0 (x) dx = 0
∂t n=1
∂xn
0 Rd Rd
for any test function ϕ ∈ C10 (D, S)
Fundamentals 8
Rankine-Hugoniot relations
Consider the 1d version of (1), s(q) = 0 integrated over interval [x, x + dx] × [t, t + dt]
x+dx
Z x+dx
Z t+dt
Z
q(x 0 , t + dt)dx 0 − q(x 0 , t)dx 0 = − f(q(x + dx, t 0 )) − f (q(x, t 0 )) dt 0

x x t
Fundamentals 9
x+dx
Z x+dx
Z t+dt
Z

x x t
Assume a discontinuity traveling with
speed
dx
S=
dt
Fundamentals 9
x+dx
Z x+dx
Z t+dt
Z

x x t
speed
dx
S=
dt
State on the left of discontinuity is index

with L, on the right with R Inserting the
states into (8) gives
(qL − qR ) dx = − [f(qR ) − f(qL )] dt
Fundamentals 9
x+dx
Z x+dx
Z t+dt
Z

x x t
speed
dx
S=
dt

(qL − qR ) dx = − [f(qR ) − f(qL )] dt
Or using the above speed definition

S(qR − qL ) = f(qR ) − f(qL )
Fundamentals 9
x+dx
Z x+dx
Z t+dt
Z

x x t
speed
dx
S=
dt

(qL − qR ) dx = − [f(qR ) − f(qL )] dt
Or using the above speed definition

S(qR − qL ) = f(qR ) − f(qL )
This is called Rankine-Hugoniot jump relation. Note the form f(qR ) = f(qL ) for S = 0
from which, for instance, the shock relations for Euler equations are derived.
Fundamentals 9
Entropy solutions
Select physical weak solution as lim qε = q almost everywhere in D of
ε→0
d d
∂qε X ∂fn (qε ) X ∂ 2 qε
+ −ε 2
= s(qε ) , x ∈ Rd , t > 0
∂t n=1
∂xn
n=1
∂x n
Fundamentals 10
Entropy solutions
ε→0
d d
+ −ε 2
= s(qε ) , x ∈ Rd , t > 0
∂t n=1
∂xn
n=1
∂x n
Theorem (Entropy condition)

Assume existence of entropy η ∈ C2 (S, R) and entropy fluxes ψn ∈ C1 (S, R)
that satisfy
Fundamentals 10
Entropy solutions
ε→0
d d
+ −ε 2
= s(qε ) , x ∈ Rd , t > 0
∂t n=1
∂xn
n=1
∂x n

that satisfy
∂η(q) T ∂fn (q) ∂ψn (q) T
· = , n = 1, . . . , d
∂q ∂q ∂q
Fundamentals 10
Entropy solutions
ε→0
d d
+ −ε 2
= s(qε ) , x ∈ Rd , t > 0
∂t n=1
∂xn
n=1
∂x n

that satisfy
∂η(q) T ∂fn (q) ∂ψn (q) T
· = , n = 1, . . . , d
∂q ∂q ∂q
then lim qε = q almost everywhere in D is weak solution and satisfies
ε→0
d
∂η(q) X ∂ψn (q) ∂η(q) T
+ ≤ · s(q)
∂t n=1
∂xn ∂q
in the sense of distributions. Proof: [Godlewski and Raviart, 1996]
Fundamentals 10
Entropy solutions II
Definition (Entropy solution)
Weak solution q is called an entropy solution if q satisfies
Fundamentals 11
Z∞Z " d
#
∂η(q) T
Z
∂ϕ X ∂ϕ
η(q) + ψn (q) − ϕ · s(q) dx dt + ϕ(x, 0) η(q0 (x)) dx ≥ 0
∂t n=1
∂xn ∂q
0 Rd Rd
for all entropy functions η(q) and all test functions ϕ ∈ C10 (D, R+
0 ), ϕ ≥ 0
Fundamentals 11
Z∞Z " d
#
∂η(q) T
Z
∂ϕ X ∂ϕ
∂t n=1
∂xn ∂q
0 Rd Rd
0 ), ϕ ≥ 0
Theorem (Jump conditions)

An entropy solution q is a classical solution q ∈ C1 (D,S) almost everywhere and
satisfies the Rankine-Hugoniot (RH) jump condition
d
X
(q+ − q− ) σt + fn (q+ ) − fn (q− ) σn = 0

n=1
Fundamentals 11
Z∞Z " d
#
∂η(q) T
Z
∂ϕ X ∂ϕ
∂t n=1
∂xn ∂q
0 Rd Rd
0 ), ϕ ≥ 0
Theorem (Jump conditions)

An entropy solution q is a classical solution q ∈ C1 (D,S) almost everywhere and
satisfies the Rankine-Hugoniot (RH) jump condition
d
X
(q+ − q− ) σt + fn (q+ ) − fn (q− ) σn = 0

n=1
and the jump inequality

d
X
(η(q+ ) − η(q− )) σt + ψn (q+ ) − ψn (q− ) σn ≤ 0

n=1
along discontinuities. Proof: [Godlewski and Raviart, 1996]

Fundamentals 11
Euler equations
∂ρ ∂
+ ρun = 0
∂t ∂xn
∂ ∂
ρuk + ρuk un + δkn p = 0 , k = 1, . . . , d
∂t ∂xn
∂ ∂
ρE + un (ρE + p) = 0
∂t ∂xn
Fundamentals 12
Euler equations
∂ρ ∂
+ ρun = 0
∂t ∂xn
∂ ∂
ρuk + ρuk un + δkn p = 0 , k = 1, . . . , d
∂t ∂xn
∂ ∂
ρE + un (ρE + p) = 0
∂t ∂xn
with polytrope gas equation of state
1
p = (γ − 1) ρE − ρun un
2
Fundamentals 12
Euler equations
∂ρ ∂
+ ρun = 0
∂t ∂xn
∂ ∂
ρuk + ρuk un + δkn p = 0 , k = 1, . . . , d
∂t ∂xn
∂ ∂
ρE + un (ρE + p) = 0
∂t ∂xn
with polytrope gas equation of state
1
p = (γ − 1) ρE − ρun un
2
have structure
∂t q(x, t) + ∇ · f(q(x, t)) = 0
Fundamentals 12

The Jacobian can be written in different forms, using
p p 1 2 a2 1
a2 = γ , h=e+ , H =h+ u ⇒ H= + u2
ρ ρ 2 γ−1 2
For
0 1 0
 
γ−3 2
u (3 − γ)u γ−1 
 
A=

2 
 1 
−uH + (γ − 1) u 3 H − (γ − 1) u 2 γu
2
Fundamentals 13

The Jacobian can be written in different forms, using
p p 1 2 a2 1
a2 = γ , h=e+ , H =h+ u ⇒ H= + u2
ρ ρ 2 γ−1 2
For
0 1 0
 
γ−3 2
u (3 − γ)u γ−1 
 
A=

2 
 1 
−uH + (γ − 1) u 3 H − (γ − 1) u 2 γu
2
The matrices  
1 1 1
R= u−a u u+a 
1 2
H − ua 2
u H + ua
 1
(γ − 1)u 2 + ua (1 − γ) u − a γ−1

2
1 
R−1 2a2 − (γ − 1)u 2

= 2 2(γ − 1) u 2(1 − γ) 
2a  
1
2
(γ − 1)u 2 − ua (1 − γ) u + a γ−1
diagonalize A as
Fundamentals 13
Characteristic form of the Euler equations - II

 
u−a 0 0
−1
R AR = Λ =  0 u 0 
0 0 u+a
Fundamentals 14

 
u−a 0 0
−1
R AR = Λ =  0 u 0 
0 0 u+a
The transformation R−1 dq = R−1 (dρ, d (ρu) , d (ρE ))T into characteristic
variables therefore leads to
∂v− ∂v−
+ (u − a) =0
∂t ∂x
∂v0 ∂v0
+u =0
∂t ∂x
+
∂v ∂v+
+ (u + a) =0
∂t ∂x
Fundamentals 14

 
u−a 0 0
−1
R AR = Λ =  0 u 0 
0 0 u+a
The transformation R−1 dq = R−1 (dρ, d (ρu) , d (ρE ))T into characteristic
variables therefore leads to
∂v− ∂v−
+ (u − a) =0
∂t ∂x
∂v0 ∂v0
+u =0
∂t ∂x
+
∂v ∂v+
+ (u + a) =0
∂t ∂x
with
dp
dv − = du − = 0 for dx = (u − a)dt
ρa
dp
dv0 = dρ − 2 = 0 for dx = u dt
a The crossing of characteristics
+ dp causes a shock wave.
dv = du + = 0 for dx = (u + a)dt
ρa
Fundamentals 14
Rarefaction and shock waves in the x − t plane

Consider the two enclosing characteristics b1 (t) ≤ x ≤ y ≤ b2 (t)
Fundamentals 15

Rarefaction:
u(x, t)±a(x, t) ≤ u(y , t)±a(y , t)
Fundamentals 15

Shocks:
Rarefaction: u(x, t)±a(x, t) ≥ u(y , t)±a(y , t)
which gives for the shock speed

u(x, t)±a(x, t) ≤ u(y , t)±a(y , t)
uL ± aL ≥ S ≥ uR ± aR
Fundamentals 15
∂ρ ∂
+ ρun = 0
∂t ∂xn
∂ ∂
ρuk + k = 1, . . . , d
ρuk un + δkn p − τkn = 0 ,
∂t ∂xn
∂ ∂
ρE + un (ρE + p) + qn − τnj uj = 0
∂t ∂xn
Fundamentals 16
∂ρ ∂
+ ρun = 0
∂t ∂xn
∂ ∂
ρuk + k = 1, . . . , d
∂t ∂xn
∂ ∂
∂t ∂xn
with stress tensor
∂un ∂uk 2 ∂uj
τkn = µ + − µ δkn
∂xk ∂xn 3 ∂xj
and heat conduction
∂T
qn = −λ
∂xn
Fundamentals 16
∂ρ ∂
+ ρun = 0
∂t ∂xn
∂ ∂
ρuk + k = 1, . . . , d
∂t ∂xn
∂ ∂
∂t ∂xn
with stress tensor
∂un ∂uk 2 ∂uj
∂xk ∂xn 3 ∂xj
and heat conduction
∂T
qn = −λ
∂xn
have structure
∂t q(x, t) + ∇ · f(q(x, t)) + ∇ · h(q(x, t), ∇q(x, t)) = 0
Fundamentals 16
∂ρ ∂
+ ρun = 0
∂t ∂xn
∂ ∂
ρuk + k = 1, . . . , d
∂t ∂xn
∂ ∂
∂t ∂xn
with stress tensor
∂un ∂uk 2 ∂uj
∂xk ∂xn 3 ∂xj
and heat conduction
∂T
qn = −λ
∂xn
have structure
∂t q(x, t) + ∇ · f(q(x, t)) + ∇ · h(q(x, t), ∇q(x, t)) = 0
Type can be either hyperbolic or parabolic

Fundamentals 16
Navier-Stokes equations for multiple species

For multiple species with chemical reaction, the Navier-Stokes equations would
be extended to
∂ρi ∂
+ (ρi un + ρνin ) = Wi ω̇i , i = 1, . . . , N
∂t ∂xn
∂ ∂
(ρuk ) + (ρuk un + δkn p − τkn ) = 0 , k = 1, . . . , d
∂t ∂xn
∂ ∂ X
(ρE ) + (un (ρE + p) + qn + ρ hj νjn − τnj uj ) = 0
∂t ∂xn j
with diffusivities
∂Yi
νin = Di
∂xn
of species i into the mixture (note difference to binary diffusivities).
Fundamentals 17

be extended to
∂ρi ∂
+ (ρi un + ρνin ) = Wi ω̇i , i = 1, . . . , N
∂t ∂xn
∂ ∂
∂t ∂xn
∂ ∂ X
∂t ∂xn j
with diffusivities
∂Yi
νin = Di
∂xn
of species i into the mixture (note difference to binary diffusivities). The
equation of state X
p= ρi R i T
i
still contains the temperature, which complicates the analysis.
Fundamentals 17

be extended to
∂ρi ∂
+ (ρi un + ρνin ) = Wi ω̇i , i = 1, . . . , N
∂t ∂xn
∂ ∂
∂t ∂xn
∂ ∂ X
∂t ∂xn j
with diffusivities
∂Yi
νin = Di
∂xn
of species i into the mixture (note difference to binary diffusivities). The
equation of state X
p= ρi R i T
i
still contains the temperature, which complicates the analysis. The structure is
∂t q(x, t) + ∇ · f(q(x, t)) + ∇ · h(q(x, t), ∇q(x, t)) = s(q(x, t))
Fundamentals 17
Outline
Conservation laws

Upwind schemes
Fundamentals 18
Derivation
Assume ∂t q + ∂x f(q) + ∂x h(q(·, ∂x q)) = s(q)
Fundamentals 19
Derivation
Time discretization tn = n∆t, discrete volumes
Ij = [xj − 12 ∆x, xj + 21 ∆x[=: [xj−1/2 , xj+1/2 [
Fundamentals 19
Derivation
Ij = [xj − 12 ∆x, xj + 21 ∆x[=: [xj−1/2 , xj+1/2 [
1 1
Z Z
Using approximations Qj (t) ≈ q(x, t) dx, s(Qj (t)) ≈ s(q(x, t)) dx
|Ij | |Ij |
Ij Ij
and numerical fluxes

F Qj (t), Qj+1 (t) ≈ f(q(xj+1/2 , t)), H Qj (t), Qj+1 (t) ≈ h(q(xj+1/2 , t), ∇q(xj+1/2 , t))
Fundamentals 19
Derivation
Ij = [xj − 12 ∆x, xj + 21 ∆x[=: [xj−1/2 , xj+1/2 [
1 1
Z Z
|Ij | |Ij |
Ij Ij

yields after integration (Gauss theorem)
tn+1
Z
1
Qj (tn+1 ) = Qj (tn ) − [F (Qj (t), Qj+1 (t)) − F (Qj−1 (t), Qj (t))] dt−
∆x
tn
tn+1 tn+1
Z Z
1
[H (Qj (t), Qj+1 (t)) − H (Qj−1 (t), Qj (t))] dt + s(Qj (t)) dt
∆x
tn tn
Fundamentals 19
Derivation
Ij = [xj − 12 ∆x, xj + 21 ∆x[=: [xj−1/2 , xj+1/2 [
1 1
Z Z
|Ij | |Ij |
Ij Ij

yields after integration (Gauss theorem)
tn+1
Z
1
Qj (tn+1 ) = Qj (tn ) − [F (Qj (t), Qj+1 (t)) − F (Qj−1 (t), Qj (t))] dt−
∆x
tn
tn+1 tn+1
Z Z
1
[H (Qj (t), Qj+1 (t)) − H (Qj−1 (t), Qj (t))] dt + s(Qj (t)) dt
∆x
tn tn
For instance:
n+1 n ∆t h n n
n n
i
Qj = Qj − F Qj , Qj+1 − F Qj−1 , Qj −
∆x
∆t h n n
n n
i
n
H Qj , Qj+1 − H Qj−1 , Qj + ∆ts(Qj ) dt
∆x
Fundamentals 19
Some classical definitions

(2s + 1)-point difference scheme of the form
Qn+1
j = H(∆t) (Qnj−s , . . . , Qnj+s )
Fundamentals 20

Qn+1
j = H(∆t) (Qnj−s , . . . , Qnj+s )
Definition (Stability)
For each time τ there is a constant CS and a value n0 ∈ N such that
kH(∆t) (Qn )k ≤ CS for all n∆t ≤ τ , n < n0
Fundamentals 20

Qn+1
j = H(∆t) (Qnj−s , . . . , Qnj+s )
Definition (Consistency)
If the local truncation error
1 h i
L(∆t) (x, t) := q(x, t + ∆t) − H(∆t) (q(·, t))
∆t
satisfies kL(∆t) (·, t)k → 0 as ∆t → 0
Fundamentals 20

Qn+1
j = H(∆t) (Qnj−s , . . . , Qnj+s )
Definition (Consistency)
If the local truncation error
1 h i
L(∆t) (x, t) := q(x, t + ∆t) − H(∆t) (q(·, t))
∆t
satisfies kL(∆t) (·, t)k → 0 as ∆t → 0
Definition (Convergence)
If the global error E (∆t) (x, t) := Q(x, t) − q(x, t) satisfies kE (∆t) (·, t)k → 0 as
∆t → 0 for all admissible initial data q0 (x)
Fundamentals 20
Some classical definitions II

Definition (Order of accuracy)
H(·) is accurate of order o if for all sufficiently smooth initial data q0 (x), there
is a constant CL , such that the local truncation error satisfies
kL(∆t) (·, t)k ≤ CL ∆t o for all ∆t < ∆t0 , t ≤ τ
Fundamentals 21

Definition (Conservative form)

If H(·) can be written in the form
∆t
Qn+1 = Qnj − F(Qnj−s+1 , . . . , Qnj+s ) − F(Qnj−s , . . . , Qnj+s−1 )

j
∆x
Fundamentals 21


∆t

j
∆x
A conservative scheme satisfies

X X
Qn+1
j = Qnj
j ∈Z j ∈Z
Fundamentals 21


∆t

j
∆x
A conservative scheme satisfies

X X
Qn+1
j = Qnj
j ∈Z j ∈Z
Definition (Consistency of a conservative method)

If the numerical flux satisfies F(q, . . . , q) = f(q) for all q ∈ S
Fundamentals 21
Splitting methods
Solve homogeneous PDE and ODE successively!

∆t
H(∆t) : ∂t q + ∇ · f(q) = 0 , IC: Q(tm ) =⇒ Q̃
∆t
S (∆t) : ∂t q = s(q) , IC: Q̃ =⇒ Q(tm + ∆t)
Fundamentals 22
Splitting methods

∆t
H(∆t) : ∂t q + ∇ · f(q) = 0 , IC: Q(tm ) =⇒ Q̃
∆t
S (∆t) : ∂t q = s(q) , IC: Q̃ =⇒ Q(tm + ∆t)
1st-order Godunov splitting: Q(tm + ∆t) = S (∆t) H(∆t) (Q(tm )),
Fundamentals 22
Splitting methods

∆t
H(∆t) : ∂t q + ∇ · f(q) = 0 , IC: Q(tm ) =⇒ Q̃
∆t
S (∆t) : ∂t q = s(q) , IC: Q̃ =⇒ Q(tm + ∆t)

1 1
2nd-order Strang splitting : Q(tm + ∆t) = S ( 2 ∆t) H(∆t) S ( 2 ∆t) (Q(tm ))
Fundamentals 22
Splitting methods

∆t
H(∆t) : ∂t q + ∇ · f(q) = 0 , IC: Q(tm ) =⇒ Q̃
∆t
S (∆t) : ∂t q = s(q) , IC: Q̃ =⇒ Q(tm + ∆t)

1 1
2nd-order Strang splitting : Q(tm + ∆t) = S ( 2 ∆t) H(∆t) S ( 2 ∆t) (Q(tm ))
1st-order dimensional splitting for H(·) :

(∆t) ∆t
X1 : ∂t q + ∂x1 f1 (q) = 0 , IC: Q(tm ) =⇒ Q̃1/2
(∆t) ∆t
X2 : ∂t q + ∂x2 f2 (q) = 0 , IC: Q̃1/2 =⇒ Q̃
[Toro, 1999]
Fundamentals 22
Conservative scheme for diffusion equation

Consider ∂t q − c∆q = 0 with c ∈ R+
Fundamentals 23

Consider ∂t q − c∆q = 0 with c ∈ R+ , which is readily discretized as
n+1 n ∆t n n n
∆t n n n

Qjk = Qjk +c 2
Qj+1,k − 2Qjk + Qj−1,k +c 2
Qj,k+1 − 2Qjk + Qj,k−1
∆x1 ∆x2
Fundamentals 23

n+1 n ∆t n n n
∆t n n n

Qjk = Qjk +c 2
Qj+1,k − 2Qjk + Qj−1,k +c 2
∆x1 ∆x2
or conservatively
∆t ∆t

n+1 n 1 1 2 2
Qjk = Qjk +c Hj+ 1 ,k − Hj− 1 ,k +c Hj,k+ 1 − Hj,k− 1
∆x1 2 2 ∆x2 2 2
Fundamentals 23

n+1 n ∆t n n n
∆t n n n

Qjk = Qjk +c 2
Qj+1,k − 2Qjk + Qj−1,k +c 2
∆x1 ∆x2
or conservatively
∆t ∆t

n+1 n 1 1 2 2
∆x1 2 2 ∆x2 2 2
Von Neumann stability analysis: Insert single eigenmode Q̂(t)e ik1 x1 e ik2 x2 into
discretization
Q̂ n+1 = Q̂ n +C1 Q̂ n e ik1 ∆x1 − 2Q̂ n + Q̂ n e −ik1 ∆x1 +C2 Q̂ n e ik2 ∆x2 − 2Q̂ n + Q̂ n e −ik2 ∆x2
∆t
with Cι = c ∆x 2 , ι = 1, 2,
ι
Fundamentals 23

n+1 n ∆t n n n
∆t n n n

Qjk = Qjk +c 2
Qj+1,k − 2Qjk + Qj−1,k +c 2
∆x1 ∆x2
or conservatively
∆t ∆t

n+1 n 1 1 2 2
∆x1 2 2 ∆x2 2 2
discretization
∆t ikι xι = cos(k x ) + i sin(k x )
with Cι = c ∆x 2 , ι = 1, 2, which gives after inserting e ι ι ι ι
ι
Q̂ n+1 = Q̂ n (1 + 2C1 (cos(k1 ∆x1 ) − 1) + 2C2 (cos(k2 ∆x2 ) − 1))
Fundamentals 23

n+1 n ∆t n n n
∆t n n n

Qjk = Qjk +c 2
Qj+1,k − 2Qjk + Qj−1,k +c 2
∆x1 ∆x2
or conservatively
∆t ∆t

n+1 n 1 1 2 2
∆x1 2 2 ∆x2 2 2
discretization
ι

Stability requires
|1 + 2C1 (cos(k1 ∆x1 ) − 1) + 2C2 (cos(k2 ∆x2 ) − 1)| ≤ 1
Fundamentals 23

n+1 n ∆t n n n
∆t n n n

Qjk = Qjk +c 2
Qj+1,k − 2Qjk + Qj−1,k +c 2
∆x1 ∆x2
or conservatively
∆t ∆t

n+1 n 1 1 2 2
∆x1 2 2 ∆x2 2 2
discretization
ι

Stability requires
|1 + 2C1 (cos(k1 ∆x1 ) − 1) + 2C2 (cos(k2 ∆x2 ) − 1)| ≤ 1
i.e.
|1 − 4C1 − 4C2 | ≤ 1
from which we derive the stability condition
∆t ∆t 1

0≤c 2
+ 2
≤
∆x1 ∆x2 2
Fundamentals 23
Outline
Conservation laws

Upwind schemes
Fundamentals 24
The Riemann problem in the linear case
Consider the linear hyperbolic equation (i.e. A = const. )

∂ ∂
q(x, t) + A q(x, t) = 0
∂t ∂x
Assume (for simplicity) that A has M distinct real eigenvalues λ1 < . . . < λM
with M linear independent right eigenvectors rm .
Fundamentals 25

∂ ∂
q(x, t) + A q(x, t) = 0
∂t ∂x
We can readily apply the characteristic transformation R−1 q = v to obtain
∂v ∂v
+Λ =0
∂t ∂x
or
∂vm ∂vm
+ λm =0 for all m = 1, . . . , M
∂t ∂x
Fundamentals 25

∂ ∂
q(x, t) + A q(x, t) = 0
∂t ∂x
We can readily apply the characteristic transformation R−1 q = v to obtain
∂v ∂v
+Λ =0
∂t ∂x
or
∂vm ∂vm
+ λm =0 for all m = 1, . . . , M
∂t ∂x
Each characteristic variable vm changes only across the characteristic line
associated to λm .
Since the entire problem is linear, we can simply sum up all these jumps ∆vm
successively to connect the RP states vL , vR in characteristic variables.
Fundamentals 25
The Riemann problem in the linear case - II
Example of a linear 3 PDE system:
Introducing the jumps
∆v1 = [vR1 − vL1 , 0, 0]T
∆v2 = [0, vR2 − vL2 , 0]T

∆v3 = [0, 0, vR3 − vL3 ]T
the solution reads
Fundamentals 26
The Riemann problem in the linear case - II
Example of a linear 3 PDE system:
Introducing the jumps
∆v1 = [vR1 − vL1 , 0, 0]T
∆v2 = [0, vR2 − vL2 , 0]T

∆v3 = [0, 0, vR3 − vL3 ]T
the solution reads


 vL = vR − ∆v3 − ∆v2 − ∆v1 x/t < λ3
x 
vL + ∆v3 = vR − ∆v2 − ∆v1 λ3 < x/t < λ2
v =
t  vL + ∆v3 + ∆v2 = vR − ∆v1
 λ2 < x/t < λ1
vL + ∆v3 + ∆v2 + ∆v1 = vR λ1 < x/t

Fundamentals 26
The Riemann problem in the linear case - III

or using the transformation q = Rv and R∆vm = rm ∆vm

 qL = qR − r3 ∆v3 − r2 ∆v2 − r1 ∆v1 x/t < λ3
x  
qL + r3 ∆v3 = qR − r2 ∆v2 − r1 ∆v1 λ3 < x/t < λ2
q =
t 
 qL + r3 ∆v3 + r2 ∆v2 = qR − r1 ∆v1 λ2 < x/t < λ1
qL + r3 ∆v3 + r2 ∆v2 + r1 ∆v1 = qR λ1 < x/t

Fundamentals 27


 qL = qR − r3 ∆v3 − r2 ∆v2 − r1 ∆v1 x/t < λ3
x  
qL + r3 ∆v3 = qR − r2 ∆v2 − r1 ∆v1 λ3 < x/t < λ2
q =
t 
 qL + r3 ∆v3 + r2 ∆v2 = qR − r1 ∆v1 λ2 < x/t < λ1
qL + r3 ∆v3 + r2 ∆v2 + r1 ∆v1 = qR λ1 < x/t

Multiplying with A and using Arm = λm rm gives


 AqL = AqR − r3 λ3 ∆v3 − r2 λ2 ∆v2 − r1 λ1 ∆v1 x/t < λ3
x  
AqL + r3 λ3 ∆v3 = AqR − r2 λ2 ∆v2 − r1 λ1 ∆v1 λ3 < x/t < λ2
Aq =
t 
 AqL + r3 λ3 ∆v3 + r2 λ2 ∆v2 = AqR − r1 λ1 ∆v1 λ2 < x/t < λ1
AqL + r3 λ3 ∆v3 + r2 λ2 ∆v2 + r1 λ1 ∆v1 = AqR λ1 < x/t

Fundamentals 27


 qL = qR − r3 ∆v3 − r2 ∆v2 − r1 ∆v1 x/t < λ3
x  
qL + r3 ∆v3 = qR − r2 ∆v2 − r1 ∆v1 λ3 < x/t < λ2
q =
t 
 qL + r3 ∆v3 + r2 ∆v2 = qR − r1 ∆v1 λ2 < x/t < λ1
qL + r3 ∆v3 + r2 ∆v2 + r1 ∆v1 = qR λ1 < x/t

Multiplying with A and using Arm = λm rm gives


 AqL = AqR − r3 λ3 ∆v3 − r2 λ2 ∆v2 − r1 λ1 ∆v1 x/t < λ3
x  
AqL + r3 λ3 ∆v3 = AqR − r2 λ2 ∆v2 − r1 λ1 ∆v1 λ3 < x/t < λ2
Aq =
t 
 AqL + r3 λ3 ∆v3 + r2 λ2 ∆v2 = AqR − r1 λ1 ∆v1 λ2 < x/t < λ1
AqL + r3 λ3 ∆v3 + r2 λ2 ∆v2 + r1 λ1 ∆v1 = AqR λ1 < x/t

This allows direct evaluation of the flux at x=0.
F(qL , qR ) := f(q(0, t)) = A q(0, t)
Fundamentals 27
Linear upwind schemes
Consider Riemann problem

∂ ∂
q(x, t)+A q(x, t) = 0 , x ∈ R , t > 0
∂t ∂x
Fundamentals 28
t

. . . . .
M
X M−1
X
β 1 r1 + δm r m β m r m + δM r M
m=2 m=1
M
X M
X
∂ ∂ q = δm r m q = β m rm
q(x, t)+A q(x, t) = 0 , x ∈ R , t > 0 L
m=1
R
m=1
∂t ∂x
x
Has exact solution 0
X X X X
q(x, t) = qL + am rm = qR − am r m = δm r m + βm r m
λm <x/t λm ≥x/t λm ≥x/t λm <x/t
Fundamentals 28
t

. . . . .
M
X M−1
X
m=2 m=1
M
X M
X
q(x, t)+A q(x, t) = 0 , x ∈ R , t > 0 L
m=1
R
m=1
∂t ∂x
x
X X X X
Use Riemann problem to evaluate numerical flux F(qL , qR ) := f(q(0, t)) = Aq(0, t) as
X X X X
F(qL , qR ) = AqL + am λm rm = AqR − am λ m r m = δ m λm r m + βm λm rm
λm <0 λm ≥0 λm ≥0 λm <0
Fundamentals 28
t

. . . . .
M
X M−1
X
m=2 m=1
M
X M
X
q(x, t)+A q(x, t) = 0 , x ∈ R , t > 0 L
m=1
R
m=1
∂t ∂x
x
X X X X
X X X X
λm <0 λm ≥0 λm ≥0 λm <0
Use λ+
m = max(λm , 0) , λ−m = min(λm , 0)
to define Λ+ := diag(λ1 , . . . , λ+
+
M) , Λ− := diag(λ− −
1 , . . . , λM )
Fundamentals 28
t

. . . . .
M
X M−1
X
m=2 m=1
M
X M
X
q(x, t)+A q(x, t) = 0 , x ∈ R , t > 0 L
m=1
R
m=1
∂t ∂x
x
X X X X
X X X X
λm <0 λm ≥0 λm ≥0 λm <0
Use λ+
m = max(λm , 0) , λ−m = min(λm , 0)
to define Λ+ := diag(λ1 , . . . , λ+
+
M) , Λ− := diag(λ− −
1 , . . . , λM )
and A+ := +
RΛ R ,−1 A := R Λ R−1
− − which gives
F(qL , qR ) = AqL + A ∆q = AqR − A ∆q = A+ qL + A− qR
− +
with ∆q = qR − qL
Fundamentals 28
Flux difference splitting
Godunov-type scheme with ∆Qnj+1/2 = Qnj+1 − Qnj
∆t −
Qn+1
j = Qnj − A ∆Qnj+1/2 + A+ ∆Qnj−1/2
∆x
Fundamentals 29
∆t −
Qn+1
j = Qnj − A ∆Qnj+1/2 + A+ ∆Qnj−1/2
∆x
Use linearization f̄(q̄) = Â(qL , qR )q̄ and construct scheme for nonlinear
problem as
∆t − n n
Qn+1
j = Qnj − Â (Qj , Qj+1 )∆Qnj+ 1 + Â+ (Qnj−1 , Qnj )∆Qnj− 1
∆x 2 2
Fundamentals 29
∆t −
Qn+1
j = Qnj − A ∆Qnj+1/2 + A+ ∆Qnj−1/2
∆x
Use linearization f̄(q̄) = Â(qL , qR )q̄ and construct scheme for nonlinear
problem as
∆t − n n
Qn+1
j = Qnj − Â (Qj , Qj+1 )∆Qnj+ 1 + Â+ (Qnj−1 , Qnj )∆Qnj− 1
∆x 2 2
stability condition
∆t
max |λ̂m,j+ 12 | ≤1, for all m = 1, . . . , M
j∈Z ∆x
[LeVeque, 1992]
Fundamentals 29
Roe’s approximate Riemann solver
Choosing Â(qL , qR ) [Roe, 1981]: tn+1
ql qr
tn
Fundamentals 30

(i) Â(qL , qR ) has real eigenvalues
ql qr
tn
Fundamentals 30

∂f(q)
(ii) Â(qL , qR ) → as qL , qR → q
∂q ql qr
tn
Fundamentals 30

∂f(q)
(ii) Â(qL , qR ) → as qL , qR → q
∂q ql qr
(iii) Â(qL , qR )∆q = f(qR ) − f(qL ) tn
Fundamentals 30

∂f(q)
(ii) as qL , qR → q
Â(qL , qR ) →
∂q ql qr
(iii) Â(qL , qR )∆q = f(qR ) − f(qL ) tn
X
Wave decomposition: ∆q = qr − ql = am r̂m
m
X X
F(qL , qR ) = f(qL ) + λ̂m am r̂m = f(qR ) − λ̂m am r̂m
λ̂m <0 λ̂!
m ≥0
1 X
= f(qL ) + f(qR ) − |λ̂m | am r̂m
2 m
Fundamentals 30
The Roe solver for Euler equations

For Euler equations, the following non-apparent average defines the Roe method:
The average for u and H read
√ √ √ √
ρL uL + ρR uR ρL HL + ρR HR
û := √ √ , Ĥ := √ √
ρL + ρR ρL + ρR
The average of the density is
√ √
ρL ρR + ρR ρL √
ρ̂ = √ √ = ρL ρR
ρL + ρR
Fundamentals 31

√ √ √ √
û := √ √ , Ĥ := √ √
ρL + ρR ρL + ρR
√ √
ρ̂ = √ √ = ρL ρR
ρL + ρR
and the averaged speed of sound is
1/2
1

â := (γ − 1)(Ĥ − û 2 )
2
The eigenvectors read
1
     
1 1
r̂1 =  û − â  , r̂2 =  û  , r̂3 =  û + â 
Ĥ − ûâ û 2 Ĥ + ûâ
2
Fundamentals 31

√ √ √ √
û := √ √ , Ĥ := √ √
ρL + ρR ρL + ρR
√ √
ρ̂ = √ √ = ρL ρR
ρL + ρR
and the averaged speed of sound is
1/2
1

â := (γ − 1)(Ĥ − û 2 )
2
The eigenvectors read
1
     
1 1
r̂2 =  û  ,
r̂1 =  û − â  , r̂3 =  û + â 
Ĥ − ûâ û 2 Ĥ + ûâ
2
and the characteristic wave strengths are
∆p − ρ̂â∆u ∆p ∆p + ρ̂â∆u
∆v1 = a1 = , ∆v2 = a2 = ∆ρ− , ∆v3 = a3 = .
2â2 â2 2â2
Fundamentals 31
Harten-Lax-Van Leer (HLL) approximate Riemann solver

sL tn+1 sR tn+1
tn+1
q⋆

 qL , x < sL t
q̄(x, t) = q? , sL t ≤ x ≤ sR t
qR , x > sR t

qL qR
tn
Fundamentals 32

sL tn+1 sR tn+1
tn+1
q⋆

 qL , x < sL t
q̄(x, t) = q? , sL t ≤ x ≤ sR t
qR , x > sR t

qL qR
tn


 f(qL ) , 0 < sL ,

 sR f(qL ) − sL f(qR ) + sL sR (qR − qL )
FHLL (qL , qR ) = , sL ≤ 0 ≤ sR ,


 sR − sL
f(qR ) , 0 > sR ,

Fundamentals 32

sL tn+1 sR tn+1
tn+1
q⋆

 qL , x < sL t
q̄(x, t) = q? , sL t ≤ x ≤ sR t
qR , x > sR t

qL qR
tn


 f(qL ) , 0 < sL ,

 sR f(qL ) − sL f(qR ) + sL sR (qR − qL )
FHLL (qL , qR ) = , sL ≤ 0 ≤ sR ,


 sR − sL
f(qR ) , 0 > sR ,

Euler equations:
sL = min(u1,L − cL , u1,R − cR ) , sR = max(u1,L + cl , u1,R + cR )
[Toro, 1999], HLLC: [Toro et al., 1994]

Fundamentals 32
Flux vector splitting
Splitting
F(qL , qR ) = f + (qL ) + f − (qR )
f(q) = f + (q) + f − (q) tl+1
f − (qL ) f + (qL ) f − (qR ) f + (qR )
tl
qL qR
Fundamentals 33
Splitting
F(qL , qR ) = f + (qL ) + f − (qR )
f(q) = f + (q) + f − (q) tl+1
derived under restriction λ̂+ m ≥ 0 and f − (qL ) f + (qL ) f − (qR ) f + (qR )

λ̂−
m ≤ 0 for all m = 1, . . . , M for
∂f + (q) ∂f − (q) tl
Â+ (q) = , Â− (q) = qL qR
∂q ∂q
Fundamentals 33
Splitting
F(qL , qR ) = f + (qL ) + f − (qR )
f(q) = f + (q) + f − (q) tl+1
derived under restriction λ̂+ m ≥ 0 and f − (qL ) f + (qL ) f − (qR ) f + (qR )

λ̂−
m ≤ 0 for all m = 1, . . . , M for
∂f + (q) ∂f − (q) tl
Â+ (q) = , Â− (q) = qL qR
∂q ∂q
plus reproduction of regular upwinding
f + (q) = f(q) , f − (q) = 0 if λm ≥ 0 for all m = 1, . . . , M

f + (q) = 0, f − (q) = f(q) if λm ≤ 0 for all m = 1, . . . , M
Then use
F(qL , qR ) = f + (qL ) + f − (qR )
Fundamentals 33
Steger-Warming
Required f(q) = A(q) q
Fundamentals 34
Steger-Warming

1 1
λ+
m = (λm + |λm |) λ−
m = (λm − |λm |)
2 2
A+ (q) := R(q) Λ+ (q) R−1 (q) , A− (q) := R(q) Λ− (q) R−1 (q)
Fundamentals 34
Steger-Warming

1 1
λ+
m = (λm + |λm |) λ−
m = (λm − |λm |)
2 2
Gives
f(q) = A+ (q) q + A− (q) q
and the numerical flux
F(qL , qR ) = A+ (qL ) qL + A− (qR ) qR
Fundamentals 34
Steger-Warming

1 1
λ+
m = (λm + |λm |) λ−
m = (λm − |λm |)
2 2
Gives
f(q) = A+ (q) q + A− (q) q
and the numerical flux
F(qL , qR ) = A+ (qL ) qL + A− (qR ) qR
Jacobians of the split fluxes are identical to A± (q) only in linear case
∂ A± (q) q

∂f ± (q) ∂A± (q)
= = A± (q) + q
∂q ∂q ∂q
Further methods: Van Leer FVS [Toro, 1999], AUSM [Wada and Liou, 1997]
Fundamentals 34
Steger-Warming FVS for Euler equations

For Euler equations, f(q) = A(q) q holds true. We also know all matrices R−1 AR = Λ.
Fundamentals 35

Approach 1: Introduce λ+ 1
m = 2 (λm + |λm |) , λ− 1
m = 2 (λm − |λm |) and compute A
+
and A− directly.
Fundamentals 35

m = 2 (λm + |λm |) , λ− 1
+
and A− directly.
Approach 2: Analyze sign of eigenvalues:
u < −a : All eigenvalues are negative: u − a < u < u + a < 0
T
f − (q) = ρu, ρu 2 + p, ρuH , f + (q) = 0

Fundamentals 35

m = 2 (λm + |λm |) , λ− 1
+
and A− directly.
T
f − (q) = ρu, ρu 2 + p, ρuH , f + (q) = 0

u > a : All eigenvalues are positive: 0 < u − a < u < u + a

T
f − (q) = 0, f + (q) = ρu, ρu 2 + p, ρuH

Fundamentals 35

m = 2 (λm + |λm |) , λ− 1
+
and A− directly.
T
f − (q) = ρu, ρu 2 + p, ρuH , f + (q) = 0

u > a : All eigenvalues are positive: 0 < u − a < u < u + a

T
f − (q) = 0, f + (q) = ρu, ρu 2 + p, ρuH

−a ≤ u ≤ a : We find u − a ≤ 0 and u + a ≥ 0 are always satisfied. For u < 0, we

need to evaluate
   
u−a 0 0 0 0 0
− +
f (q) = R  0 u 0  R−1 q, f (q) = R  0 0 0  R−1 q
0 0 0 0 0 u+a
For u ≥ 0, we need to evaluate
   
u−a 0 0 0 0 0
− +
f (q) = R  0 0 0  R−1 q, f (q) = R  0 u 0  R−1 q
0 0 0 0 0 u+a
Fundamentals 35
References
References I
[Godlewski and Raviart, 1996] Godlewski, E. and Raviart, P.-A. (1996). Numerical
approximation of hyperbolic systems of conservation laws. Springer Verlag, New
York.
[Kröner, 1997] Kröner, D. (1997). Numerical schemes for conservation laws. John
Wiley & Sons and B. G. Teubner, New York, Leipzig.
[LeVeque, 1992] LeVeque, R. J. (1992). Numerical methods for conservation laws.

Birkhäuser, Basel.
[Majda, 1984] Majda, A. (1984). Compressible fluid flow and systems of conservation
laws in several space variables. Applied Mathematical Sciences Vol. 53.
Springer-Verlag, New York.
[Roe, 1981] Roe, P. L. (1981). Approximate Riemann solvers, parameter vectors and
difference schemes. J. Comput. Phys., 43:357–372.
[Toro, 1999] Toro, E. F. (1999). Riemann solvers and numerical methods for fluid
dynamics. Springer-Verlag, Berlin, Heidelberg, 2nd edition.
[Toro et al., 1994] Toro, E. F., Spruce, M., and Speares, W. (1994). Restoration of
the contact surface in the HLL-Riemann solver. Shock Waves, 4:25–34.
Fundamentals 36
References
References II
[Wada and Liou, 1997] Wada, Y. and Liou, M.-S. (1997). An accurate and robust
flux splitting scheme for shock and contact discontinuities. SIAM J. Sci. Comp.,
18(3):633–657.
Fundamentals 37

Fundamentals: Course Block-Structured Adaptive Finite Volume Methods in C++

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fundamentals: Course Block-Structured Adaptive Finite Volume Methods in C++

Uploaded by

Copyright:

Available Formats

Conservation laws Finite volume methods Upwind schemes References

Finite volume methods

Finite volume methods

Finite volume methods

Hyperbolic Conservation Laws

Hyperbolic Conservation Laws

q = q(x, t) ∈ S ⊂ RM - vector of state, fn (q) ∈ C1 (S, RM ) - flux functions,

Hyperbolic Conservation Laws

q = q(x, t) ∈ S ⊂ RM - vector of state, fn (q) ∈ C1 (S, RM ) - flux functions,

Hyperbolic Conservation Laws

q = q(x, t) ∈ S ⊂ RM - vector of state, fn (q) ∈ C1 (S, RM ) - flux functions,

Hyperbolic Conservation Laws

q = q(x, t) ∈ S ⊂ RM - vector of state, fn (q) ∈ C1 (S, RM ) - flux functions,

If fn (q) is nonlinear, classical solutions

Example: Euler equations

R is the matrix of right eigenvectors (column-wise)

R is the matrix of right eigenvectors (column-wise)

and Λ the diagonal matrix of eigenvalues

Typical wave diagram for vector model problem.

Weak and entropy solutions

Integral form (Gauss’s theorem):

Weak and entropy solutions

Integral form (Gauss’s theorem):

Theorem (Weak solution)

Weak and entropy solutions

Integral form (Gauss’s theorem):

Theorem (Weak solution)

for any test function ϕ ∈ C10 (D, S)

Weak and entropy solutions

Weak and entropy solutions

Weak and entropy solutions

State on the left of discontinuity is index

(qL − qR ) dx = − [f(qR ) − f(qL )] dt

Weak and entropy solutions

State on the left of discontinuity is index

(qL − qR ) dx = − [f(qR ) − f(qL )] dt

Or using the above speed definition

Weak and entropy solutions

State on the left of discontinuity is index

(qL − qR ) dx = − [f(qR ) − f(qL )] dt

Or using the above speed definition

Weak and entropy solutions

Weak and entropy solutions

Theorem (Entropy condition)

Weak and entropy solutions

Theorem (Entropy condition)

Weak and entropy solutions

Theorem (Entropy condition)

in the sense of distributions. Proof: [Godlewski and Raviart, 1996]

Weak and entropy solutions

Weak and entropy solutions

Weak and entropy solutions

Theorem (Jump conditions)

Weak and entropy solutions

Theorem (Jump conditions)

and the jump inequality

along discontinuities. Proof: [Godlewski and Raviart, 1996]

Characteristic form of the Euler equations

Characteristic form of the Euler equations

Characteristic form of the Euler equations

Characteristic form of the Euler equations

Characteristic form of the Euler equations

Characteristic form of the Euler equations