kinetic methods for cfd - ossanworld.com
TRANSCRIPT
Kinetic Methods for CFD
Li-Shi Luo
Department of Mathematics and StatisticsCenter for Computational Sciences
Old Dominion University, Norfolk, Virginia 23529, USAEmail: [email protected]
URL: http://www.lions.odu.edu/~lluo
National Institute of Aerospace, Hampton, USATuesday, January 17, 2012
Collaborators: Wei Liao, Yan Peng, Manfred Krafczyk, et al.Supports from: NSF, DoD/AFOSR, NASA, ODU-RF
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 1 / 53
Outline
1 MotivationsThe Role of Kinetic TheoryScales and Related Methods
2 Mathematicsal BackgroundsKinetic Theory – A Multiscale FormalismLBE: Finite Difference Scheme for Incompressible FlowsD3Q19 LBE ModelGKS: Finite Volume Scheme for Compressible Flows
3 Numerical ResultsDNS of Homogeneous Isotropic TurbulenceCompressible and Nonequilibrium FlowsSome Animations
4 Conclusions
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 2 / 53
Outline
1 MotivationsThe Role of Kinetic TheoryScales and Related Methods
2 Mathematicsal BackgroundsKinetic Theory – A Multiscale FormalismLBE: Finite Difference Scheme for Incompressible FlowsD3Q19 LBE ModelGKS: Finite Volume Scheme for Compressible Flows
3 Numerical ResultsDNS of Homogeneous Isotropic TurbulenceCompressible and Nonequilibrium FlowsSome Animations
4 Conclusions
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 3 / 53
Navier-Stokes Equations: Continuum Theory
Conservation laws of mass, momentum, and energy:
∂tρ+ ∇·ρu = 0 (1a)
ρ∂u+ ρu·∇u = −∇·P (1b)
ρ∂te+ ρu·∇e = −P : ∇u−∇·q (1c)
Pαβ := pδαβ − µ(∂αuβ + ∂βuα −
2
3δαβ∇·u
)− ζδαβ∇·u
q = −κ∇T , e = e(T ), p = p(ρ, T )
Dimensionless Navier-Stokes equations (similarity law):
ρ∂u+ ρu·∇u = − 1
γMa2∇p+
1
Re∇·S (2a)
ρ∂te+ ρu·∇e = − 1
γMa2p∇·u+
1
α∇2T +
1
ReS : ∇u (2b)
α = (γ − 1)PrMaRe
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 4 / 53
Hierarchy of Scales and PDEs
Microscopic Scale
qk =∂H
∂pk, pk = −∂H
∂qk
H =∑(D+K)N
k=1 p2k + V
i~ψ = Hψ
H = − ~2
2m
∑Nj=1∇
2j+V
h ≈ 6.62 · 10−34(J · s)c ≈ 2.99 · 108(m/s)
a ≈ 5·10−11(m)
ta ≈ 2.41·10−17(s)
m ≈ 10−27(kg)
N = 1, 2, . . . , N0
Mesoscopic Scale
∂tf + ξ ·∇f =1
εQ(f, f)
f = f(x, ξ, t)
ε = Kn =`
L, Ma =
U
cskB ≈ 1.38·10−23(J/K)
` ≈ 102 – 103 (A)
≈ 10 – 100 (nm)
τ ≈ 10−10(s)
cs ≈ 300 (m/s)
N 1
Macroscopic Scale
ρDtu = −∇p+1
Re
∇·σ
Dt = ∂t + u·∇σ =
ρν
2[(∇u)+(∇u)†]
+2ρζ
DI(∇·u)
Reδ =UL
ν∼ Ma
Knν ≈ 10−6 – 10−4 (m2/s)
Kn ≈ 0 Ma < 103
L ≥ 10−5(m) = 10(µm)
T ≥ 10−4(s)
N ≥ NA ≈ 6.02 · 1023
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 5 / 53
Micro-, Meso-, Macro-Descriptions of Fluids
Knudsen Number Kn :=`
L=
Mean Free Path
Characteristic Hydrodynamic Length
Microscopic
Theory
DeterministicNewton’s Law
MolecularDynamics
Mesoscopic
Theory
StatisticalMechanics
LiouvilleEquation
BBGKYHierachy
BoltzmannEquation
xi, tn ci
Direct SimulationMonte Carlo
DiscreteVelocity Models
DiscreteOrdinance Method
MomentEquations
Macroscopic (Continuum) Theory (Kn ≈ 0)
PDEs of Conservation Laws
Navier-StokesEquations
EulerEquations
Turbulence ModelsRANS, LES, · · ·
Super-Burnett
Equation
BurnettEquation
Kinetic Theory
Hilbert and Chapman-Enskog analysis
Lattice GasAutomata
Lattice BoltzmannEquation
Gas-KineticScheme
Linear Relaxation ModelsSystem of Finite Moments
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 6 / 53
Outline
1 MotivationsThe Role of Kinetic TheoryScales and Related Methods
2 Mathematicsal BackgroundsKinetic Theory – A Multiscale FormalismLBE: Finite Difference Scheme for Incompressible FlowsD3Q19 LBE ModelGKS: Finite Volume Scheme for Compressible Flows
3 Numerical ResultsDNS of Homogeneous Isotropic TurbulenceCompressible and Nonequilibrium FlowsSome Animations
4 Conclusions
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 7 / 53
Kinetic Equation
The kinetic equation for the single particle distribution function f inphase space Γ := (x, ξ):
∂tf + ∇ · (ξf) = Q(f, f), f := f(x, ξ, t), (3)
The Uehling-Uhlenbeck collision model:
Q =
∫dξ2
∫dΩK
[(1 + ηf1) (1 + ηf2) f ′1f
′2 −
(1 + ηf ′1
) (1 + ηf ′2
)f1f2
],
(4)where K := K(ξ1, ξ2,Ω) is the collision kernel, depending on the solidangle Ω, and
η =
+1, Bose-Einstein,
0, Maxwell-Boltzmann,−1, Fermi-Dirac.
(5)
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 8 / 53
From KE to Hydrodynamics Equations
Expansion of f in terms of the Knudsen number ε = Kn:
f = f (0) + εf (1) + ε2f (2) + · · ·, (6)
f (0) =ρ
(2πRT )D/2e−(ξ−u)2/2RT , (7)
ρ =
∫fdξ =
∫f (0)dξ, (8a)
ρu =
∫fξdξ =
∫f (0)ξdξ, (8b)
ρe =D
2ρkBT =
∫1
2c2cf dξ =
∫1
2c2cf (0) dξ, c := ξ − u, (8c)
P =
∫ξξf dξ, q =
∫1
2c2cf dξ, (8d)
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 9 / 53
Multiscale Formalism
∂tρ+ ∇ · ρu = 0 (9a)
∂tρu+ ∇ (ρuu+ P) = 0 (9b)
∂tρE + ∇ · (ρuE + u · P + q) = 0, E := ρe+1
2ρu2 (9c)
P = P(0) + P(1) + P(2) + · · ·, (9d)
q = q(0) + q(1) + q(2) + · · ·, (9e)
f = f (0) =⇒
P(0) = 32ρkBTδij
q(0) = 0
=⇒ Euler Eqns
f = f (0) + f (1) =⇒
P(1) = −1
2µ[(∇u) + (∇u)†]
q(1) = κ∇T
=⇒ NS Eqns
f = f (0) + f (1) + f (2) + · · ·+ f (n) =⇒ ?
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 10 / 53
Linearize Collision Operator
The collision operator for the classical (Maxwell-Boltzmann) particle:
Q(f, f)=
∫dµ2
(f ′1f′2 − f1f2
)=
∫dµ2f
′1f′2 − f1
∫dµ2f2
≈ f (0)
1
∫dµ2f
(0)
2 − f1
∫dµ2f2 = − 1
λ
[f1 − f (0)
1
](10)
The above renowned Welander-Bhatnagar-Gross-Krook model reducesthe Boltzmann equation to:
∂tf + ∇ · (ξf) = − 1
λ
[f − f (0)
], f (0) =
ρ
(2πRT )3/2e−(ξ−u)2/2RT (11)
and its moment equations forms a hierarchy of first-order PDEs:
∂tm(n) + ∇ ·m(n+1) = − 1
λ
[m(n) −m
(n)0
](12a)
m(n) := 〈ξ · · · ξ︸ ︷︷ ︸n
〉 :=
∫dξf ξ · · · ξ︸ ︷︷ ︸
n
, m(n)0 :=
∫dξf (0) ξ · · · ξ︸ ︷︷ ︸
n
(12b)
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 11 / 53
A Priori Derivation of Lattice Boltzmann Equation
The Boltzmann Equation for f := f(x, ξ, t) with BGK approximation:
∂tf + ξ ·∇f =
∫[f ′1f
′2 − f1f2]dµ ≈ L(f, f) ≈ − 1
λ[f − f (0)] (13)
The Boltzmann-Maxwellian equilibrium distribution function:
f (0) = ρ (2πθ)−D/2 exp
[−(ξ − u)2
2θ
], θ := RT (14)
The macroscopic variables are the first few moments of f and f (0):
ρ =
∫fdξ =
∫f (0)dξ , (15a)
ρu =
∫ξfdξ =
∫ξf (0)dξ , (15b)
ρε =1
2
∫(ξ − u)2fdξ =
1
2
∫(ξ − u)2f (0)dξ . (15c)
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 12 / 53
Integral Solution of Continuous Boltzmann Equation
Rewrite the Boltzmann BGK Equation in the form of ODE:
Dtf +1
λf =
1
λf (0) , Dt := ∂t + ξ ·∇ . (16)
Integrate Eq. (35) over a time step δt along characteristics:
f(x+ ξδt, ξ, t+ δt) = e−δt/λ f(x, ξ, t) (17)
+1
λe−δt/λ
∫ δt
0et′/λ f (0)(x+ ξt′, ξ, t+ t′) dt′ .
By Taylor expansion, and with τ := λ/δt, we obtain:
f(x+ ξδt, ξ, t+ δt)− f(x, ξ, t) = −1
τ[f(x, ξ, t)− f (0)(x, ξ, t)] +O(δ2
t ) .
(18)Note that a finite-volume scheme or higher-order schemes can also beformulated based upon the integral solution.
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 13 / 53
Passage to Lattice Boltzmann EquationThree necessary steps to derive LBE:1,2,3
1 Low Mach number expansion of the distribution functions;
2 Discretize ξ-space with necessary and min. number of ξα;
3 Discretization of x space according to ξα.Low Mach Number (u ≈ 0) Expansion of the distribution functions f (0)
and f up to O(u2) is sufficient to derive the Navier-Stokes equations:
f (eq) =ρ
(2πθ)D/2exp
[−ξ
2
2θ
]1 +
ξ · uθ
+(ξ · u)2
2θ2− u
2
2θ
+O(u3) .(19a)
f =ρ
(2πθ)D/2exp
[−ξ
2
2θ
] 2∑n=0
1
n!a(n)(x, t) : H(n)(ξ) , (19b)
where a(0) = 1, a(1) = u, a(2) = uu− (θ − 1)I, and H(n)(ξ) aregeneralized Hermite polynomials.
1X. He and L.-S. Luo, Phys. Rev. E 55:R6333 (1997); ibid 56:6811 (1997).
2T. Abe, J. Comput. Phys. 131(1):241–246(1997).
3X. Shan and X. He, Phys. Rev. Lett. 80:65 (1998).
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 14 / 53
Discretization and Conservation Laws
To compute conserved moments (ρ, ρu, and ρε) and their fluxesexactly, one must evalute:
I =
∫ξmf (eq)dξ =
∫exp(−ξ2/2θ)ψ(ξ)dξ, (20)
where 0 ≤ m ≤ 3, and ψ(ξ) is a polynomial in ξ. The above integralcan be evaluated by quadrature:
I =
∫exp(−ξ2/2θ)ψ(ξ)dξ=
∑j
Wj exp(−ξ2j /2θ)ψ(ξj) (21)
where ξj and Wj are the abscissas and the weights. Then
ρ=∑α
f (eq)α =
∑α
fα, ρu=∑α
ξαf(eq)α =
∑α
ξαfα, (22)
where fα := fα(x, t) := Wαf(x, ξα, t), and f (eq)α := Wαf
(eq)(x, ξα, t).
The quadrature must preserve the conservation laws exactly!
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 15 / 53
Example: 9-bit LBE Model with Square Lattice
In two-dimensional Cartesian (momentum) space, set
ψ(ξ) = ξmx ξny ,
the integral of the moments can be given by
I = (√
2θ)(m+n+2)ImIn, Im =
∫ +∞
−∞e−ζ
2ζmdζ, (23)
where ζ = ξx/√
2θ or ξy/√
2θ.The second-order Hermite formula (k = 2) is the optimal choice toevaluate Im for the purpose of deriving the 9-bit model, i.e.,
Im =
3∑j=1
ωjζmj .
Note that the above quadrature is exact up to m = 5 = (2k + 1).Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 16 / 53
Discretization of Velocity ξ-Space (9-bit Model)
The three abscissas in momentum space (ζj) and the correspondingweights (ωj) are:
ζ1 = −√
3/2 , ζ2 = 0 , ζ3 =√
3/2 ,ω1 =
√π/6 , ω2 = 2
√π/3 , ω3 =
√π/6 .
(24)
Then, the integral of moments becomes:
I = 2θ
[ω2
2ψ(0) +4∑
α=1
ω1ω2ψ(ξα) +8∑
α=5
ω21ψ(ξα)
], (25)
where
ξα =
(0, 0) α = 0,
(±1, 0)√
3θ, (0, ±1)√
3θ, α = 1 – 4,
(±1, ±1)√
3θ, α = 5 – 8.
(26)
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 17 / 53
Discretization of Velocity ξ-Space (9-bit Model)
IdentifyingWα = (2π θ) exp(ξ2
α/2θ)wα , (27)
with c := δx/δt =√
3θ, or c2s = θ = c2/3, δx is the lattice constant,
then:
f (eq)α (x, t) = Wα f
(eq)(x, ξα, t)
= wα ρ
1 +
3(cα · u)
c2+
9(cα · u)2
2c4− 3u2
2c2
, (28)
where weight coefficient wα and discrete velocity cα are:
wα =
4/9,1/9,1/36,
cα = ξα =
(0, 0), α = 0 ,(±1, 0) c, (0, ±1) c, α = 1 – 4,(±1, ±1) c, α = 5 – 8.
(29)
With cα|α = 0, 1, . . . , 8, a square lattice structure is constructed inthe physical space.
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 18 / 53
Discretized 2D Velocity Space (9-bit)
With Cartesian coordinates in ξ-space, a 2D square lattice is obtained:
cα =
(0, 0), α = 0 ,(±1, 0) c, (0, ±1) c, α = 1 – 4,(±1, ±1) c, α = 5 – 8.
If spherical coordinates are used, a 2D triangular lattice is obtained.
cα =
(0, 0), α = 0 ,(cos((α− 1)π/3), sin((α− 1)π/3)) c, α = 1 – 6.
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 19 / 53
D3Q19 MRT-LBE Model
f(xj + cδt, tn + δt) = f(xj , tn)−M−1 · S ·[m−m(eq)
], (30)
e(eq) = −11δρ+19
ρ0j · j, ε(eq) = 3δρ− 11
2ρ0j · j, (31a)
(q(eq)x , q(eq)
y , q(eq)z ) = −2
3(jx, jy, jz), (31b)
p(eq)xx =
1
3ρ0
[2j2x − (j2
y + j2z )], p(eq)
ww =1
ρ0
[j2y − j2
z
], (31c)
p(eq)xy =
1
ρ0jxjy, p(eq)
yz =1
ρ0jyjz, p(eq)
xz =1
ρ0jxjz, (31d)
π(eq)xx = −1
2p(eq)xx , π(eq)
ww = −1
2p(eq)ww , (31e)
m(eq)x = m(eq)
y = m(eq)z = 0, (31f)
where δρ is the density fluctuation, ρ = ρ0 + δρ and ρ0 = 1.Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 20 / 53
D3Q19 MRT-LBE Model (cont.)
Conserved quantities:
δρ =
Q−1∑i=0
fi, j = ρ0u =
Q−1∑i=0
fici,
Transport coefficients and the speed of sound:
ν =1
3
(1
sν− 1
2
), ζ =
(5− 9c3s)
9
(1
se− 1
2
), c2
s =1
3c2,
where c := δx/δt.The transform between the discrete distribution functions f ∈ V = RQand the moments m ∈M = RQ:
m = M · f , f = M−1 ·m.
Note that M−1 is related M†.
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 21 / 53
The BGK Model Equation
The Boltzmann Equation with BGK approximation:
∂tf + ξ ·∇f =
∫[f ′1f
′2 − f1f2]dµ ≈ − 1
λ[f − f (0)] , f ≡ f(x, ξ, ζ, t)
(32)The Boltzmann-Maxwellian equilibrium distribution function:
f (0) = ρ (2π/β)−(D+K)/2 exp
[−β (c · c+ ζ · ζ)
2
], β := 1/kBT (33)
The hydrodynamic variables are moments of f and f (0):
ρ =
∫fdΞ =
∫f (0)dΞ , dΞ := dξdζ, (34a)
ρu =
∫ξfdΞ =
∫ξf (0)dΞ , (34b)
ρε =1
2
∫(c · c+ ζ · ζ)fdΞ =
1
2
∫(c · c+ ζ · ζ)f (0)dΞ . (34c)
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 22 / 53
Integral Solution of Continuous Boltzmann Equation
Rewrite the Boltzmann BGK Equation in the form of ODE:
Dtf +1
λf =
1
λf (0) , Dt ≡ ∂t + ξ ·∇ . (35)
Integrate Eq. (35) over a time step δt along characteristics:
f(x+ ξδt, ξ, t+ δt) = e−δt/λ f(x, ξ, t)
+1
λe−δt/λ
∫ δt
0et′/λ f (0)(x+ ξt′, ξ, t+ t′) dt′ . (36)
From Eq. (35), formally we have:
f = f (0) − λDtf
For the Navier-Stokes equations, f ≈ f (0) − λDtf(0).
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 23 / 53
Some Convenient Notations
We use the bold-face font for (D + 2)-tuple vectors:
Ψ := (1, ξ1 , ξ2, ξ3, (c2 + ζ2)/2)T, (37a)
W := (ρ, ρu1, ρu2, ux, ρε)T =
∫fΨdΞ =
∫f (0)ΨdΞ, (37b)
Fα :=
∫fΨξαdΞ, α ∈ x, y, z, (37c)
h := (ρ, u1, u2, u3, T )T, (37d)
h′ := (ρ−1, βu1, βu2, u3,1
2[β(c2 + ζ2)− (D +K)]T−1)T, (37e)
K = (5− 3γ)/(γ − 1) + 1, c := (ξ − u), Ξ := (ξ, ζ).
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 24 / 53
Initial and Equilibrium Values
The initial value f0 is approximated by its 1st-order CE expansion
f0 ≈ [1− λ(∂t + ξ ·∇)]f (0) =[1− λh′ · (∂t + ξ ·∇)h
]f (0) (38)
The equilibrium f (0) by its 1st-order Taylor expansion at x = 0:
f (0)(x, t) ≈ [1 + x·∇]f (0)(0, t) = f (0)(0, t)[1 + (x·∇h) · h′
](39)
f0(x, t) ≈ f (0)(0, t)[1 + (x·∇h) · h′
] [1− λh′ · (∂t + ξ ·∇)h
]≈ f (0)(0, t)[1 + a·x− λ(a·ξ +A)] (40)
a ≡∇ ln f (0) = ∇ ln ρ+
[(c2 + ζ2)
2RT− 3 + Z
2
]∇ lnT +
(∇u) · cRT
(41)
A ≡ ∂t ln f (0) = −a·ξ +
[(c2 + ζ2)
2RT− 5 + Z
2
]c·∇ lnT
+
[cc
RT− (c2 + ζ2)
5RTI
]: ∇u (42)
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 25 / 53
Initial and Equilibrium Values (2)
Coefficients a and A are related by the compatibility condition of f :∫f (n)ΨdΞ = 0, ∀n > 0 =⇒
∫Af (0)dΞ = −
∫a·ξ f (0)dΞ
because
∫f (1)dΞ = −λ
∫Dtf
(0)dΞ = −λ∫
(A+ a·ξ)f (0)dΞ = 0
Thus the equilibrium f (0)(x, t) in the solution (36) is approximated by:
f (0)(x, t) ≈ (1 + t∂t + x·∇)f (0)(0, 0)
=[1 + h′ · (t∂t + x·∇)h
]f (0)(0, 0)
= f (0)(0, 0)(1 + a · x+ At) (43)
where a and A are similar to a and A, respectively: in a and A, bothhydrodynamic varibles (ρ, u, T ) and their gradients may bediscontinuous, while in a and A, only the gradients may be so.
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 26 / 53
Finite Volume Formulation
Assuming the hydrodynamic variables h can be discontinuous at thecell boundary of xi+1/2 = 0, f (0) in both sides of xi+1/2 have to beevaluated differently:
On the left, h are interpolated to x−i+1/2 from three points xi−1, xi
and xi+1, and f (0)L is computed from h at x−i+1/2;
Similarly, f (0)R are evaluated from h, which are interpolated to the
x+i+1/2 from three points xi, xi+1 and xi+2.
Therefore:
f0(x, t) = [1 + aR(x− λξ1)− λAR]H(x)f (0)R (0, t)
+[1 + aL(x− λξ1)− λAL]H(−x)f (0)L (0, t), (44a)
f (0)(x, t) = [1 +H(−x)aLx+H(x)aRx+ At]f (0)(0, 0), (44b)
where H(x) is the Heaviside function.
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 27 / 53
Finite Volume Formulation (2)
The value of f at a cell boundary can be obtained by using f0(x, t)and f (0)(x, t):
f(xi+1/2, t) =[
(1− Aλ)(1− e−t/λ) + At]
+[(t+ λ)e−t/λ − λ
][aLH(ξ1) + aRH(−ξ1)] ξ1
f (0)
0 (0, 0)
+e−t/λ
[(1− ξ1(t+ λ)aL)− λAL]H(ξ1)f (0)0L (0, 0)
+ [(1− ξ1(t+ λ)aR)− λAR]H(−ξ1)f (0)0R (0, 0)
(45)
where f (0)
0 , f (0)0L and f (0)
0R are initial values of f (0), f (0)L and f (0)
R evaluatedat cell boundary xi+1/2.
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 28 / 53
Calculation of Fluxes
With f given at cell boundaries, the time-dependent fluxes are
Fi+1/2,jα =
∫ξαΨf(xi+1/2, t)dΞ (46)
By integrating the above equation over each time step, we obtain thetotal fluxes:
Fi±1/2,jx =
∫ ∆t
0Fi±1/2,jx dt, Fi,j±1/2
y =
∫ ∆t
0Fi,j±1/2y dt. (47)
The FV formulation for the governing equations are:
Wn+1ij = Wn
ij−1
∆x(Fi+1/2,j
x − Fi−1/2,jx )− 1
∆y(Fi,j+1/2
y − Fi,j−1/2y ), (48)
which is used to update the flow field.
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 29 / 53
GKS for Near Continuum Flows
In the near-continuum or nonequilibrium flows, the viscosity is not aconstant (e.g., within a shock). Therefore, λ is no longer a constant.Assume:
f = f (0) − λDtf =⇒ f ≈ f (0) − λ∗Dtf(0)
where λ∗ is the modified relaxation time. Also
f ≈ f (0) − λDtf(1) = f (0) − λDtf
(0) + λλ∗D2t f
(0) (49)
thenλ∗Dtf
(0) ≈ λ(Dtf(0) − λ∗D2
t f(0)) (50)
λ∗ =λ
1 + λ⟨D2t g⟩/ 〈Dtg〉
(51)
where 〈D2t g〉 and 〈Dtg〉 can be computed as functions of hydrodynamic
variables and their derivatives up to second order.
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 30 / 53
Outline
1 MotivationsThe Role of Kinetic TheoryScales and Related Methods
2 Mathematicsal BackgroundsKinetic Theory – A Multiscale FormalismLBE: Finite Difference Scheme for Incompressible FlowsD3Q19 LBE ModelGKS: Finite Volume Scheme for Compressible Flows
3 Numerical ResultsDNS of Homogeneous Isotropic TurbulenceCompressible and Nonequilibrium FlowsSome Animations
4 Conclusions
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 31 / 53
In the Course of Hydrodynamic Events ...
Ma (Mach) and Kn (Knudsen) characterize nonequilibrium
Van Karmen relation based on Navier-Stokes equation (Kn=O(ε)): Ma=Re·Kn
Re1 Re≈1 Re1
Stokes Flows Incompressible Navier-Stokes FlowsMa1
Ma=O(ε2), Re=O(ε) Ma=O(ε), Re=O(1) Ma=O(ε1−α), Re=O(ε−α)
Sub/Transonic FlowsMa≈1
Ma=O(1), Re=O(ε−1)
Super/Hypersonic FlowsMa1
Ma=O(ε−1), Re=O(ε−2)
With the framework of kinetic theory (Boltzmann equation)
Hydrodynamics Slip Flow Transitional Free Molecular
Kn<10−3 10−3<Kn<10−1 10−1<Kn<10 10<Kn
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 32 / 53
Decaying Homogeneous Isotropic Turbulence
The decaying homogeneous isotropic turbulence is the solution of theincompressible Navier-Stokes equation
∂tu+ u ·∇u = −∇p+ ν∇2u, ∇ · u = 0, x ∈ [0, 2π]3, (52)
with periodic boundary conditions. The initial velocity satisfies a giveninitial energy spectrum E0(k)
E0(k) := E(k, t = 0) = Ak4e−0.14k4, k ∈ [ka, kb]. (53)
The initial velocity u0 can be given by Rogallo procedure:
u0(k) =αkk2 + βk1k3
k√k2
1 + k22
k1 +βk2k3 − αk1k
k√k2
1 + k22
k2 −β√k2
1 + k22
kk3, (54)
where α =√E0(k)/4πk2eıθ1 cosφ, β =
√E0(k)/4πk2eıθ2 sinφ,
ı :=√−1, and θ1, θ2, φ ∈ [0, 2π] are uniform random variables.
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 33 / 53
Low-Order Turbulence Statistics
The energy and the compensated spectra:
E(k, t) :=1
2u(k, t) · u†(k, t), Ψ(k) := ε(k)−2/3k5/3E(k), (55)
And the pressure spectrum P (k, t). Low-order moments of E(k, t):
K(t) :=
∫dkE(k, t), Ω(t) :=
∫dk k2E(k, t) (56a)
ε(t) := 2νΩ(t), η := 4√ν3/ε (56b)
Sui(t) =〈(∂iui)3〉〈(∂iui)2〉3/2
, Su(t) =1
3
∑i
Sui (56c)
Fui(t) =〈(∂iui)4〉〈(∂iui)2〉2
, Fu(t) =1
3
∑i
Fui (56d)
We will also compare instantaneous flows fields u(x, t) and ω(x, t).
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 34 / 53
Pseudo-Spectral Method
The pseudo-spectral (PS) method solve the Navier-Stokes equation inthe Fourier space k, i.e.,
u(x, t) =∑ku(k, t)eık·x, −N/2 + 1 ≤ kα ≤ N/2.
The nonlinear term u ·∇u computed in physical space x byinverse Fourier-transform u and ku to x for form the nonlinearterm; and it is transformed back to k space;
De-aliasing: u(k, t) = 0 ∀‖k‖ ≥ N/6;
Time matching: second-order Adams-Bashforth scheme:
u(t+ δt)− u(t)
δt= −3
2T (t) +
1
2T (t− δt)e−νk2δt,
where T := F [ω × u]− (F [ω × u] · k)k.
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 35 / 53
Parameters in DNS
Use N3 = 1283 and [ka, kb] = [3, 8].In LBE: ν = 1/600 (cδx), c := δx/δt = 1, Mamax = ‖u0‖max/cs ≤ 0.15,A = 1.4293 · 10−4 in E0(k), and K0 ≈ 1.0130 · 10−2, u′0 ≈ 8.2181 · 10−2.The time t is normalized by the turbulence turn-over time t0 = K0/ε0.In SP method, K0 = 1 and u′0 =
√2/3.
Method L δx u′0 δt ν δt′
LBE 2π 2π/N√
2K0/3 2π/N ν 2π/Nt0PS 2π 2π/N
√2/3 2π
√K0/N ν/
√K0 2π/Nt0
The Taylor microscale Reynolds number:
Reλ :=u′λ
ν, λ :=
√15
2Ωu′ :=
√15ν
εu′ (57)
The resolution criterion:
SP: N ∼ 0.4Re3/2λ , η/δx ≥ 1/2.1, N = 128→ Reλ = 46.78
LBE: ηkmax = η/δx ≥ 1, N = 128→ Reλ = 24.35.
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 36 / 53
Initial Conditions
For the psuedo-spectral method:
Generate u0(k) in k-space with a given E0(k) (Rogallo’sprocedure) with K0 = 1 and u′ =
√3/2;
The initial pressure p0 is obtained by solving the Poisson equationin k-space.
For the LBE method:4
Use the initial velocity u0 as in PS method except a scaling factorso that Mamax = 0.15;
The pressure p0 is obtained by an iterative procedure with a givenu0.
4R. Mei, L.-S. Luo, P. Lallemand, and D. d’Humieres. Computers & Fluids 35(8/9):855–862 (2006).
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 37 / 53
‖u(t′)/u′‖ and ‖ω(t′)/u′‖ at Reλ = 24.37, t′ = 4.048
LBE vs. PS1 (equal δt) and PS2 (δt/3), PS1 vs. PS2ve
loci
tyu/u′ 0
vort
icit
yω/u′ 0
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 38 / 53
‖u(t′)/u′‖ and ‖ω(t′)/u′‖ at Reλ = 24.37, t′ = 29.949
LBE vs. PS1 (equal δt) and PS2 (δt/3), PS1 vs. PS2ve
loci
tyu/u′ 0
vort
icit
yω/u′ 0
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 39 / 53
L2 ‖δu(t′)‖ and ‖δω(t′)‖
LBE vs. PS1 (equal δt) and PS2 (δt/3), PS1 vs. PS2.
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 40 / 53
Flow past 25-55 Double-Cone Ma = 9.59 (Run 28)
The free stream parameters for Run 28:Ma∞ = 9.59, p∞ = 36.06 (Pa), T∞ = 185.6 (K), Tw = 293.3 (K)Re∞ = 12, 838 (based on L = 3.625′′), Kn∞ ≈ 1.1× 10−3
The mesh size is 481× 201, minimum distant to wall ≈ 2.81 · 10−6 (m).Convergence residual: ≥ 10−7.
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 41 / 53
25-55 Double-Cone: Ma, p, T
The contours of Ma/Ma∞ (L), p/p∞ (C), and T/T∞ (R).
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 42 / 53
25-55 Double-Cone: Interaction Zone
The contours of ρ/ρ∞ (recirculation) and p/p∞ (shock interactions).
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 43 / 53
25-55 Double-Cone: Density schliere
ϕ(x, y) = exp
(−3000
‖∇ρ(x, y)‖2
‖∇ρ‖2max
)Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 44 / 53
25-55 Double-Cone: Cp and St
The pressure coefficient Cp and Stanton number St, experiment,5 NS(1024× 256),6 and GKS (481× 201).
5M. Holden and T. P. Wadhams. AIAA Paper 2001-1031 (2001).
6P. Gnoffo. AIAA Paper 2001-1025 (2001); G. Candler. AIAA Paper 2001-1024 (2001); I. Nompelis. PhD
Thesis (2004).
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 45 / 53
25-55 Double-Cone: Separation/Reattachment
Abscissa x/L LAURA NS GKS
Separation 0.5555 0.5594 0.5755
Reattachment 1.291 1.287 1.263
Length 0.7355 0.7276 0.6875
mesh size 1024× 256 1024× 256 481× 201
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 46 / 53
Shock Structures: ρ and T at Ma=8.0 and 9.0
The density and temperature profiles of monoatomic gas inside a shockof Ma= 8.0 and 9.0. GKS results vs. experimental7 and DSMC8 data.
δρ
∆ρ:=
(ρ− ρ1)
(ρ2 − ρ1)
δT
∆T:=
(T − T1)
(T2 − T1)λ1 :=
2µ(7− 2ω)(5− 2ω)
15ρ1
√2πRT1
7Schmidt, J. Fluid Mech. 39:361 (1969); Alsmeyer, J. Fluid Mech. 74:497 (1976).
8Bird, Phys. Fluids 13:1172 (1970).
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 47 / 53
Shock Structures: τxx and qx at Ma=8.0
The stress τxx and heat-flux qx profiles of monoatomic gas inside theshock of Ma= 8.0. GKS results vs. DSMC data.9
9Bird, Phys. Fluids 13:1172 (1970).
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 48 / 53
Other Applications (Animations)
Ladd, Univ. Florida: Cluster of 1812 particles, Re = 0.3
Flow through Porous Media (Krafczyk et al., TUB):
Air through fluids in a packed-sphere bio-filterMulti-component flow past porous media: imbibe, drain, re-imbibe
Free-Surface Flow (Krafczyk et al., TUB):
Flow past a bridgeWater-Tunnel: 384× 64× 48, δx = 10(cm), LES, Re = 3 · 106
Tsumani over S. Manhatten: 512× 512× 80, δx = 2.35(m)
Krafczyk et al.:Digital Wind Tunnel, 64× 64× 320, δx = 10(cm), Re = 4 · 103
Droplet collision (Frohn et al., Univ. Stuttgart):
Merge, Separate, and one extra; LBE vs. Experiment.
More (Thury et al., Univ. Erlangen):
Bubbles rising in water tankMetal foams
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 49 / 53
Outline
1 MotivationsThe Role of Kinetic TheoryScales and Related Methods
2 Mathematicsal BackgroundsKinetic Theory – A Multiscale FormalismLBE: Finite Difference Scheme for Incompressible FlowsD3Q19 LBE ModelGKS: Finite Volume Scheme for Compressible Flows
3 Numerical ResultsDNS of Homogeneous Isotropic TurbulenceCompressible and Nonequilibrium FlowsSome Animations
4 Conclusions
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 50 / 53
Why Kinetic Methods?
The Boltzmann equation (BE) is 1st-order PDE with a linearadvection term, the Navier-Stokes equation is 2nd-order and nonlinear.It is more costly to solve the BE in phase space, kinetic schemes10
Requires the smallest possible stencil for accurate discretization,thus least need for inter-nodal data communication;
Nonlinearity is in local collision term, stiffness of which can beovercome by local techniques.
Discretized 1st-order systems may be easier to converge thanequivalent higher-order systems.
1st-order PDE’s yield the highest potential discretization accuracyon non-smooth, adaptively refined grids.
The systems of 1st-order PDE’s are better suited for functionaldecomposition, thus easier to parallelize.
The Boltzmann equation is valid for nonequilibrium flows whichcannot be modeled by the Navier-Stokes equations.
10B. van Leer, AIAA Paper 2001-2520 (2001).
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 51 / 53
Questions?
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 52 / 53
Food for Thought
For an n-th order scheme, how to guarantee that the operators ∇and ∇2 are isotropic to the order consistent/comparable with n?
How to construct physics-based CFD schemes beyond upwind,(approximated) multi-dimensional Riemann solver, artificialdissipation, and limiters?
Luo (Math Dept, ODU) Kinetic Methods for CFD NIA, 01/17/’12 53 / 53