A collection of well-balanced finite volume schemes
Christophe BERTHON jointed work with • Ana¨ıs CRESTETTO, University of Nantes (France) • Vivien DESVEAUX, University of Nantes (France) • Fran¸coise FOUCHER, University of Nantes (France) • Christian KLINGENBERG, University of W¨ urzburg (Germany) • Victor MICHEL-DANSAC, University of Nantes (France) • Markus ZENK, University of W¨ urzburg (Germany) SHARK-FV 2017, May 2017
Main motivations □ Hyperbolic systems of conservation law with source term ∂t w + ∂x f (w) = S(w)
w∈Ω
• Steady states ∂x f (w) = S(w)
⇐⇒
Manifold given by M = {w ∈ Ω; g(w) = 0}
□ Finite volume schemes • Numerical approximations of the weak solutions • Approximation / Exact capture of (a part of) M • Robustness • Entropy stability
Outline □ Ripa model - Euler with gravity Non-explicit steady state capturing scheme • Model and main properties • Incomplete relaxation technique • Numerical scheme and main properties □ A chemotaxis model - Shallow-water model with Manning friction Godunov type well-balanced scheme • Godunov type scheme • Characterization of the approximate Riemann solver • Robustness and well-balanced properties □ Shallow-water model with Manning friction: Well-balanced scheme • Godunov type scheme • Characterization of the approximate Riemann solver • Robustness and well-balanced properties
Ripa model (Chertock et al 2013) □ Shallow-water with potential temperature filed ∂t h + ∂x hu = 0 ( ) 2 h { } 3 ∂t hu + ∂x hu2 + gΘ = −gΘh∂x z Ω = w ∈ R ; h > 0, u ∈ R, Θ > 0 2 ∂t hΘ + ∂x hΘu = 0 □ Steady states
Moving steady states u ̸= 0 ∂x hu = 0 hu = cst ) ( h2 2 Θ = cst ∂x hu + gΘ = −gΘh∂x z 2 2 u ∂x hΘu = 0 + gΘ(h + z) = cst 2 With u ̸= 0, we recover the general steady states with a gravity constant gΘ • Steady states at rest u = 0 2 ∂ Θ h + hΘ∂ z = 0 x x 2
unsolvable
• Particular steady states u=0 u=0 z = cst Θ = cst h2 Θ = cst h + z = cst
u=0 h = cst z + h ln Θ = cst 2
□ Objectives • Robust finite volume method • Exact capture of the 3 particular steady states at rest • Exact/Approximated solutions of the unsolvable PDE for steady states
Entropic reformulation □ Main properties The Ripa model is hyperbolic The smooth solutions satisfy the additional conservation law ∂t (η(w) + gΘhz) + ∂x (G(w) + gΘhzu) = 0 where we have set u2 h2 η(w) = h + gΘ 2 2 ( 2 ) u G(w) = h + gΘh2 u 2
Partial entropy Partial entropy flux
Unfortunately, w 7→ η(w) is never convex nor concave. It seems the model does not admit entropy pair
□ Equivalent reformulation for weak solutions: Θ = φ(θ) ∂t h + ∂x hu = 0 ) ( 2 h ∂t hu + ∂x hu2 + gφ(θ) = −gφ(θ)h∂x z 2 ∂t hθ + ∂x hθu = 0 □ Entropy inequalities: The smooth solutions of the new reformulation satisfy ) ( ˜ ∂t (˜ η (w) + gφ(θ)hz) + ∂x G(w) + gφ(θ)hzu = 0 u2 h2 η˜(w) = h + gφ(θ) 2 2 ) ( 2 u ˜ G(w) = h + gφ(θ)h2 u 2 The partial entropy function w 7→ η˜(w) is convex iff 1 ′ 2 φ (θ)φ(θ) − φ (θ) > 0 2 ′′
Example
φ(θ) = eθ
and
θ2 ′′ φ(θ) − θφ (θ) + φ (θ) > 0 2 ′
Incomplete relaxation model □ Suliciu model (Bouchut) ∂t h + ∂x hu = 0 ( 2 ) ∂t hu + ∂x hu + π + gφ(θ)h∂x z = 0 ∂ hθ + ∂ hθu = 0 t x ( ) 2 h h 2 ∂ hπ + ∂ (hπ + a )u = gφ(θ) − π t x ε 2 ∂t z = 0
z := z(x) ⇒ ∂t z = 0 Equilibrium ε → 0 h2 π = gφ(θ) 2
□ Suliciu type model (Cargo-LeRoux 94, Chalons et al 10) ∂t h + ∂x hu = 0 ) ( 2 Equilibrium ε → 0 ∂t hu + ∂x hu + π + gφ(θ)h∂x Z = 0 ∂t hθ + ∂x hθu = 0, h2 π = gφ(θ) ( ) 2 h h2 2 ∂t hπ + ∂x (hπ + a )u = gφ(θ) − π Z=z ε 2 ∂t hZ + ∂x hZu = h (z − Z) ε
□ Main properties The Suliciu type relaxation model is hyperbolic with eigenvalues u ± a/h (simple) and u (triple). All fields are linearly degenerated. The eigenvalue u admits only one Riemann invariant (instead of 2). The Riemann problem is thus illposed. □ Additional arbitrary relation To mimic the unsolvable PDE h2 ∂x gΘ = −ghΘ∂x z 2
⇐⇒ equilibrium
uL −
a hL
uR + a hR ⋆ wL
wL
∂x π = −ghΘ∂x z
We propose ⋆ ⋆ ¯ L , WR ))h(W ¯ L , WR )(Z ⋆ − Z ⋆ ) πR − πL = −gφ(θ(W R L
¯ and θ¯ are suitable averages where h
u⋆
⋆ wR
wR
□ Suliciu approximate Riemann solver W = t (h, hu, hθ, hπ, hZ) ( ( ) ) 2 h W eq (w) = t h, hu, hθ, h gφ(θ) , hz 2 ) (x ; WL , WR exact Riemann solution for the relaxation model WR t Theorem With the additional relation, there exists a unique Riemann solution. Moreover ( ) (h,hu,hθ) x WR ; W eq (wL ), W eq (wR ) t defines an approximate Riemann solver (in the sense of Harten, Lax and van Leer) for the Ripa model
Local well-balanced property • Assume wL and wR be given by the steady state relation (general) uL = uR = 0 2 2 h h φ(θ ) R − φ(θ ) L + φ(θ(w ¯ L , wR ))h(w ¯ L , wR )(zr − zL ) = 0 R L 2 2 then the approximate Riemann solver stays at rest { (x ) wL (h,hu,hθ) eq eq ; W (wL ), W (wR ) = WR t wR
if x < 0 if x > 0
• Assume wL and wR be given by the steady state relation (particular 1) uL = uR = 0 1 ¯ h(wL , wR ) = (hL + hR ) 2 θL = θR where free from θ¯ hL + zL = hR + zR then the approximate Riemann solver stays at rest
• Assume wL and wR be given by the steady state relation (particular 2) uL = uR = 0 1 ¯ h(wL , wR ) = (hL + hR ) zL = zR 2 where free from θ¯ φ(θ )h2 = φ(θ )h2 L
L
R
R
then the approximate Riemann solver stays at rest • Assume wL and wR be given by the steady state relation (particular 3) uL = uR = 0 hL = hR zL + hL ln φ(θL ) = zR + hR ln φ(θR ) 2 2
where
¯ L , wR ) = 1 (hL + hR ) h(w 2 ( ) φ(θ ) − φ(θ ) R L φ−1 ¯ L , wR ) = ln(φ(θR )) − ln(φ(θL )) θ(w θ L
then the approximate Riemann solver stays at rest
if θL ̸= θR , if θL = θR
Relaxation scheme □ Numerical approximation: (win )i∈Z known at time tn (h,hu,hθ) x n , wn ) W ( ; wi−1 i R t
CFL restriction ai±1/2 1 ∆t n max ui ± ≤ n ∆x i∈Z hi 2 win+1 =
1 ∆x
∫
n wi−1
(h,hu,hθ)
xi− 1 2
1 ∆x
WR ∫
xi+ 1 2
xi
x − xi− 12 ∆t (
(h,hu,hθ)
WR
(h,hu,hθ) x n , wn ) ( ; wi i+1 R t
n wi x i− 1 2
(
xi
W
xi
n wi+1 x i+ 1 2
) n ; W eq (wi−1 ), W eq (win ) +
x − xi+ 21 ∆t
) n ) ; W eq (win ), W eq (wi+1
Theorem • The scheme is robust: win+1 ∈ Ω • The scheme exactly preserves the particular steady states: i.e. if wi0 satisfies 0 0 0 ui = 0 u = 0 u = 0 i i 0 hi = H zi = Z θi0 = θ or or 0 0 (h0 )2 φ(θ0 ) = P h hi + zi = H i i zi + i ln(φ(θi0 )) = P 2 then win = win+1 • If the initial data wi0 is an approximation of the unsolvable steady state PDE as follows: ( ) 0 2 0 2 (hi+1 ) 1 1 0 (hi ) 0 0 0 0 0 ¯ ¯ − φ(θi ) (zi+1 −zi ) = 0 φ(θi+1 ) +φ(θ(wi , wi+1 ))h(wi , wi+1 ) ∆x 2 2 ∆x then the scheme exactly captures this approximation win+1 = win
Numerical experiments (Chertock, Kurganov, Liu 13) □ Perturbation of a nonlinear steady state z(x) = 6 − 2 exp(x)
and
(hs , us , Θs )(x) = (exp(x), 0, exp(2x))
0.04
Initial perturbation t=0.2 t=0.4
0.1
t=0.2 t=0.4
0.03 0.02
0.08
0.01 u−us
h−hs
0.06
0.04
0 −0.01 −0.02
0.02
−0.03 −0.04
0 −1
−0.5
0
0.5
1
−0.05 −1
−0.5
0
0.5
1
Figure 1: Perturbation of the free surface h − hs (left) and perturbation of the velocity u − us (right)
□ Dam break over a non-flat bottom 5 5 4.5
4
4
3 θ
h+z
3.5
2
3 2.5 2
1
1.5 1
0 −1
−0.5
12 10
p
8 6 4 2 0
0
0.5
1
−1
−0.5
0
0.5
1
Free surface h + z (top left) Temperature Θ (top right) Pressure p (bottom) at time t = 0.2
Euler with gravity (Xu et al 2010-2011, K¨ appeli and Mishra 2014) ∂t ρ + ∂x ρu = 0 ∂t ρu + ∂x (ρu2 + p(ρ, e)) = −ρ∂x Φ ∂ E + ∂ (E + p(ρ, e))u = −ρu∂ Φ t
x
u2 E = ρe + ρ 2 Φ gravity field Ω = {w ∈ R3 ; ρ > 0, u ∈ R, e > 0}
x
□ Entropy inequalities ∂t ρF(s) + ∂x ρF(s)u ≤ 0 □ Steady states at rest { u=0 ∂x p = −ρ∂x Φ
(Gibbs law)
Isothermal steady states u=0
ρ = αe
−βΦ
α −βΦ p= e β
□ Objectives • Robust and stable finite volume scheme • Exact capture of the isothermal steady states • Exact/Approximated solutions of the unsolvable PDE for steady states
Incomplete relaxation model □ Suliciu type model (Cargo-LeRoux 94, Chalons et al 10) ∂t ρ + ∂x ρu = 0 ( 2 ) Equilibrium ε → 0 ∂t ρu + ∂x ρu + π = −ρ∂x Z ∂t E + ∂x (E + π) u = −ρu∂x Z π = p(ρ, e) ρ 2 Z=Φ ∂ ρπ + ∂ (ρπ + a )u = (p(ρ, e) − π) t x ε ∂ ρZ + ∂ ρZu = ρ (Φ − Z) t x ε □ Additional closure relation ⋆ ⋆ ⋆ πR − πL = −¯ ρ(ρL , ρR )(ZR − ZL⋆ )
to mimic ∂x p = −ρ∂x Φ
Definition of an approximate Riemann solver ) ( (h,hu,hθ) x eq eq WR ; W (wL ), W (wR ) t with good properties: Robustness - Local well-balance - Stability
Relaxation scheme
win+1 =
1 ∆x
∫
(
xi
(h,hu,hθ)
xi− 1 2
1 ∆x
WR ∫
xi+ 1 2
xi
x − xi− 12 ∆t (
(h,hu,hθ)
WR
) n ; W eq (wi−1 ), W eq (win ) +
x − xi+ 21 ∆t
) n ; W eq (win ), W eq (wi+1 )
Theorem With a relaxation parameter ai+1/2 >> 1 and under a relevant CFL restriction: • Robustness: win+1 ∈ Ω • Entropy preserving: n+1 ρn+1 F(s ) i i
≤
ρni F(sni )
) ∆t ( n n − {ρF(s)u}i+1/2 − {ρF(s)u}i+1/2 ∆x
• Steady state preserving: With an initial data given by 1 0 Φi+1 − Φi (pi+1 − p0i ) + ρ(ρ0i , ρ0i+1 ) =0 ∆x ∆x Then win+1 = win
• Hydrostatic steady state preserving: Assume ρ=
ρR − ρL ln(ρR ) − ln(ρR )
and an initial data given by the hydrostatic atmosphere u0i Then win+1 = win
=0
ρ0i
= αe
−βΦi
p0i
α −βΦi = e β
Numerical Results: Small perturbation of an hydrostatic atmosphere • Computational domain: [0, 1] until time T = 0.25 • Φ(x) = x • Initial condition:
−x ρ (x) = e 0 u0 (x) = 0 −x 100(x−0.5)2 p0 (x) = e + 0.01e
• Pressure perturbation: δp = p − p0
0.01 Initial perturbation Reference solution Solution with 100 cells
pressure perturbation
0.008
0.006
0.004
0.002
0
Numerical Results: Non-hydrostatic steady state • Computational domain: [0, 1] with periodic boundary conditions • Final time T = 1 • Φ(x) = sin(2πx)
ρ0 (x) = 3 + 2 sin(2πx) • Initial condition: u0 (x) = 0 p0 (x) = 3 + 3 sin(2πx) − To have ∂x p0 + ρ0 ∂x Φ = 0 N
• L1 error:
Density
1 2
cos(4πx)
Velocity
100
4.46E-05
–
2.03E-05
–
200
7.11E-06
2.65
5.29E-06
1.94
400
1.23E-06
2.53
1.34E-06
1.98
800
2.35E-07
2.39
3.37E-07
1.99
1600
5.02E-08
2.23
8.44E-08
2.00
3200
1.15E-08
2.13
2.11E-08
2.00
A model of Chemotaxis (Ribot et al 2012 - 2014) ∂t ρ + ∂x ρu = 0 ∂t ρu + ∂x (ρu2 + p(ρ)) = −χρ∂x Φ − αρu ∂ Φ − D∂ Φ = aρ − bΦ t
χ, α, D, a, b given parameters Ω = {w ∈ R3 ; ρ ≥ 0, u ∈ R, Φ ≥ 0} p(ρ) = δργ
xx
□ Steady states at rest u = 0 e(ρ) − χΦ = cste
e(ρ) = δ
γ ργ−1 + cste γ−1
solutions of D∂xx Φ − bΦ = aρ if ρ = 0 : if ρ > 0, C < 0 : if ρ > 0, C > 0 :
( √ ) ( √ ) ϕ (x) = A cosh x b + B sinh x b , ( √ ) ( √ ) ϕ (x) = A cos x |C| + B sin x |C| − ϕp , ( √ ) ( √ ) ϕ (x) = A cosh x C + B sinh x C − ϕp ,
where A and B are some constants, C =
1 D
( b−
aχ 2ε
)
ρ (x) = ρ (x) =
and ϕp =
χ 2ε χ 2ε
(ϕ (x) − K) , (ϕ (x) − K) ,
Kaχ 2εb−aχ .
□ Asymptotic behavior Rescaling: t → t/ε (long time) and α → α/ε (dominant friction) Limit ε → 0 to get a diffusive regime { ∂t ρ = ∂x (∂x p − χρ∂x Φ) D∂xx Φ = bΦ − aρ □ Objectives • Robustness (ρ ≥ 0 and Φ ≥ 0) • Steady state preserving (well-balanced) • Asymptotic preserving Godunov type strategy (CB and Chalons 16)
Godunov type scheme {
∂t ρ + ∂x ρu = 0 ∂t ρu + ∂x (ρu2 + p(ρ)) = −χρ∂x Φ − αρu
Φ given
□ Approximate Riemann solver: w ˜ λL
0 ⋆ wL
λR ⋆ wR
wL
λL < 0 < λR HLL type solver Source term → stationary contact wave
wR
x
• Harten-Lax-van Leer consistency condition ∫ ∆x/2 ∫ ∆x/2 1 1 w ˜ (x, ∆t; wL , wR ) dx = wR (x, ∆t; wL , wR ) dx ∆x −∆x/2 ∆x −∆x/2 1 ∆x
∫
∆x/2
1 ∆t ⋆ ⋆ w ˜ (x, ∆t; wL , wR ) dx = (wL +wR )+ (λL (wL −wL )+λR (wR −wR )) 2 ∆x −∆x/2
Because tof the source term, wR stays unknown wR (x, t; wL , wR ) wR (t)
wL (t) wL
1 ∆x
wR
∫
x
constant is not a natural solution ∫ ∆x/2 ∫ ∆t ( ) ∂t w + ∂x f (w) = S(w) dxdt −∆x/2
0
∫ ∆x/2 ∫ ∆t 1 1 wR (x, ∆t; wL , wR ) dx = (wL + wR ) + S(w)dxdt 2 ∆x −∆x/2 −∆x/2 0 ∫ ∆t ∫ ∆t 1 1 − f (wR (∆x/2, t))dt + f (wR (−∆x/2, t))dt ∆x 0 ∆x 0 ∆x/2
Approximation wR (−∆x/2, t) ≃ wL wR (∆x/2, t) ≃ wR
As a consequence ∫ ∆x/2 ∆t 1 1 ρR (x, ∆t; wL , wR ) dx ≃ (ρL + ρR ) − (ρR uR − ρL uL ) ∆x −∆x/2 2 ∆x ∫ ∆x/2 1 1 ∆t (ρu)R (x, ∆t; wL , wR ) dx ≃ (ρL uL + ρR uR ) − (ρR u2R + pR − ρL u2L − pL ) ∆x −∆x/2 2 ∆x ∫ ∆t ∫ ∆x/2 1 + ∆tSR − α (ρu)R (x, t; wL , wR ) dxdt ∆x 0 −∆x/2 ∫ ∆t ∫ ∆x/2 1 SR = χρR ∂x Φdxdt ∆t∆x 0 −∆x/2 • Approximation SR ≃ S ⋆ to be defined (independently from ∆t) • Approximation ∫ ∆x/2 1 (ρu)R (x, ∆t; wL , wR ) dx ≃ F (∆t) ∆x −∆x/2 solution of the following integral equation 1 ∆t F(∆t) = (ρL uL + ρR uR ) − (ρR u2R + pR − ρL u2L − pL ) + ∆tS ⋆ − α 2 ∆x
∫
∆t
F(t)dt 0
□ Consistency conditions λL (ρL − ρ⋆L ) + λR (ρ⋆R − ρR ) = ρL uL − ρR uR λL (ρL uL − ρ⋆L u⋆L ) + λR (ρ⋆R u⋆R − ρR uR ) = ) ) (α 1 ( −α∆t e −1 (ρL uL + ρR uR )∆x − (ρR u2R + pR − ρL u2L − pL ) + ∆xS ⋆ α∆t 2 □ Flux continuity: ρ⋆L u⋆L = ρ⋆R u⋆R □ Well-balanced conditions χ pR − pL (ΦR − ΦL ) S = ∆x eR − eL ρ⋆L ρ⋆R eL − χϕL = eR − χϕR ρL ρR ⋆
□ Approximate Riemann solver (Munz et al 09): w ˜ λL
0 Φ⋆ L
λR
{
Φ⋆ R
ΦL
ΦR
∂t Φ + ∂x Ψ = aρ − bΦ Ψ = ∂x Φ
ρ given
x
□ Harten-Lax-van Leer consistency condition ∫ ∆x/2 ∫ ∆x/2 1 ˜ (x, ∆t; wL , wR ) dx = 1 Φ ΦR (x, ∆t; wL , wR ) dx ∆x −∆x/2 ∆x −∆x/2 Main difficulty: Evaluate ΦR ∫ ∆x/2 ∫ ∆t ∫ (∂t Φ − D∂x Ψ) dxdt = a −∆x/2
0
∆x/2 −∆x/2
∫
∫ ∫
∆t
ρR (x, t)dxdt−b 0
Approximations ∫ ∆x/2 t 1 1 (ρR uR − ρL uL ) ρR (x, t)dx ≃ (ρL + ρR ) − ∆x −∆x/2 2 ∆x ∫ ∆x/2 ∫ ∆t 1 ∆t ∂x Ψ(x, t)dx ≃ (ΨR − ΨL ) ∆x −∆x/2 0 ∆x
ΦR (x, t)dxdt
Godunov type scheme □ Approximate Riemann solver: w ˜ λL
0 Φ⋆ L
λR
{
Φ⋆ R
ΦL
ΦR
∂t Φ + ∂x Ψ = aρ − bΦ Ψ = ∂x Φ
ρ given
x
□ Harten-Lax-van Leer consistency condition ∫ ∆x/2 1 ΦR (x, t)dx ≃ G(∆t) ∆x −∆x/2 G(∆t) solution of the following integral equation ( ) 2 ∆t ∆t ∆t 1 (ΨR − ΨL ) + a (ρL + ρR ) − (ρR uR − ρL uL ) G(∆t) = (ΦL + ΦR ) + D 2 ∆x 2 2∆x ∫ ∆t G(t)dt −b 0
□ Godunov type scheme ( ( ) ( )) n+1 ∆t ⋆ ⋆ w = win − ∆x λi− 21 ,R win − wi− − λi+ 12 ,L win − wi+ 1 1 i ,R 2 2 ,L)) ( ) ( ( n ⋆ n ⋆ Φn+1 = Φn − ∆t λ 1 1 Φ − λ Φ − Φ − Φ 1 i− ,R i+ ,L i i i i ∆x i− ,R i+ 1 ,L 2 2 2
2
□ Definition of Ψni 1 (Φni+1 − Φni−1 ) × E(∆x) 2∆x E(∆x) is consitent with 1 lim E(∆x) = 1 Ψni =
∆x→0
We impose
b ∆x2 ( ) √ 2 cos b∆x − 1 ∆x2 C ( ) √ E(∆x) = 2 cos |C|∆x − 1 2 ∆x C ( ) √ 2 cosh C∆x − 1
to recover the steady states
if ρ = 0
if ρ > 0, C < 0
if ρ > 0, C > 0
Theorem • Robustness (adopting a local cut-off, Chalons 2014) ρ⋆L,R = min(max(0, ρ⋆L,R ), 2ρHLL ) Φ⋆L,R = min(max(0, Φ⋆L,R ), 2ΦHLL ) • Well-balance property • Asymptotic preserving
⇒
ρn+1 ≥0 i
and
Φn+1 ≥0 i
Numerical results Influence of γ at χ = 10, L = 3, ∆x = 0.01. 8
30
=3
=2 7 25
6 20
5
4
15
3 10
2 5
1
0
0 0
0.5
1
1.5
2
2.5
3
0
0.5
1
1.5
2
2.5
3
x
x
1.8
3.5
=5
=4 1.6 3
1.4 2.5
1.2 2
1
0.8
1.5
0.6 1
0.4 0.5
0.2
0
0 0
0.5
1
1.5
x
2
2.5
3
0
0.5
1
1.5
x
2
2.5
3
Shallow-water model with Manning friction (Vazquez-Cend´ on) ∂t h + ∂x hu = 0 ( ) 2 h gκ 2 = − η |hu|hu − gh∂x z ∂t hu + ∂x hu + g 2 h ∂t w + ∂x f (w) = S(w) Ω = {w ∈ R2 ; h > 0} To simplify, we fix z = cst ⇒ □ Steady states
gκ S = − η |hu|hu h
∂x hu = 0 ) ( 2 gκ h = − η |hu|hu ∂x hu2 + g 2 h
With hu = 0, we recover the usual shallow-water steady states
η>2
□ Steady states hu = cst 2 g (h u ) 0 0 η+2 (hη+2 − h0 ) − (hη−1 − h0η−1 ) = −gκ|h0 u0 |h0 u0 (x − x0 ) η+2 η−1 where h0 = h(x0 ) and u0 = u(x0 ) Remark • Steady states are not globally defined (∀x ∈ R) □ Objectives: Numerical scheme preserving the main properties • Steady states preserving (well-balanced) • Water height positive preserving
Approximate Riemann solver □ Two intermediate constant states λL
0 ⋆ wL
wL
λR ⋆ wR wR
x
λL < 0 < λR HLLC type solver Source term → stationary contact wave 4 unknowns: h⋆L,R and u⋆L,R
• Consistency condition ∫ ∆x ∫ ∆x 2 2 1 x 1 x w( ˜ ; wL , wR ) dx = wR ( ; wL , wR ) dx ∆x − ∆x t ∆x − ∆x t 2 2
Approximate Riemann solver □ Two intermediate constant states λL
0 ⋆ wL
λR ⋆ wR
wL
wR
x
λL < 0 < λR HLLC type solver Source term → stationary contact wave 4 unknowns: h⋆L,R and u⋆L,R
• Consistency condition ⋆ ⋆ h u − h u = λ (h − h ) + λ (h L L R R L L R L R − hR ) ) ( ) ( 2 2 h h hL u2L + g L − hR u2R + g R + ∆x S¯LR = 2 2 λL (hL uL − h⋆L u⋆L ) + λR (h⋆R u⋆R − hR uR ) where We approximate S¯LR
1 S¯LR = ∆t∆x ,→
∫
∆x 2
− ∆x 2
∫
∆t
− 0
gκ η |hR uR |hR uR dx dt hR
We get an additional unknown
• Contact preserving: h⋆L u⋆L = h⋆R u⋆R = q ⋆ • Steady state preserving: S¯LR is defined by enforcing
hL uL = hR uR = q0 [ ] 2 g q0 η−1 η+2 h − h − gκ|q0 |q0 ∆x = η+2 η−1 LR to get
[ S¯LR = −gκ|q ⋆ |q ⋆ [
(q ⋆ )2 h
g η+2 h η+2
{ and
2
+ g h2
−
h⋆L = hL
u⋆L = uL
h⋆R = hR
u⋆R = uR
] LR
(q ⋆ )2 η−1 η−1 h
] LR
,→ It defines a consistent average ,→ It is a necessary condition to recover all steady states • Additional relation (contact preserving): ) ( ) ( ⋆ 2 ⋆ 2 (q ) g (q ) g (h⋆R )η+2 − (h⋆R )η−1 − (h⋆L )η+2 − (h⋆L )η−1 −gκ|q ⋆ |q ⋆ ∆x = η+2 η−1 η+2 η−1 Remark Nonlinear equations are solved by a Newton method
Numerical experiments □ Perturbation of a nonlinear steady state
Figure 2: Perturbation of the free surface h
Numerical experiments □ Dam break
Thanks for your attention