A multiphase lattice Boltzmann model with sharp interfaces
0 likes127 views
The document presents a multiphase lattice Boltzmann model focused on simulating immiscible fluid flows with sharp interfaces, utilizing a colour-based approach to represent different fluids. It introduces a phase field to manage interface dynamics and includes a random projection method to address numerical pinning issues. The model is validated against analytical solutions and demonstrates capabilities in maintaining sharp interfaces and accurately representing surface tension effects.
![A multiphase lattice Boltzmann
model with sharp interfaces
Tim Reis
Oxford Centre for Collaborative Applied Mathematics (OCCAM)
http://www.maths.ox.ac.uk/occam/
This poster is based on work supported by Award No. KUK-C1-013-04 made by
King Abdullah University of Science and Technology (KAUST)
Introduction
Multiphase lattice Boltzmann models usually fall into one of three cate-
gories: Colour gradient, or Rothman-Keller, models; Shan-Chen models;
or free-energy models. The Free-Energy and Shan-Chen models are in-
tended to simulate flows where the kinematics of phase separation or
interactions at the molecular level are of interest. These models do not
allow for independent adjustment of either the surface tension or the
viscosity ratio and the resolved interface is spread over several lattice
nodes. Here we are concerned with macroscopic, immiscible, flows and
therefore pursue colour-based models since they offer the most attrac-
tive features to the modeller who wishes to simulate binary flows on the
length scale addressed by continuum mechanics.
We introduce a phase field (or colour field) φ to represent the differ-
ent fluids. Regions where the phase field φ ≈ 0 might represent one
phase (e.g oil) and regions where φ ≈ 1 the other phase (e.g water).
In principle, φ is advected by the fluid velocity, hence φt + u · ∇φ = 0.
However, to make the most of limited spacial resolution it is common
to add a “sharpening term” to counteract the inevitable numerical dif-
fusion and preserve relatively thin boundaries between different phases.
For example, the sharpening term might drive all points where φ > 1
2
back towards φ = 1, and all points where φ < 1
2 back towards φ = 0.
The width of the phase boundary is then controlled by a balance be-
tween diffusion and the sharpening term. However, overly narrow phase
boundaries can fail to propagate correctly, becoming fixed or “pinned”
to the grid [1].
Phase field sharpening and pinning
A useful model for studying hyperbolic equations with ”stiff” source
terms is [2]:
∂φ
∂t
+ u
∂φ
∂x
= S(φ) = −
1
T
φ 1 − φ
φc − φ
,
The problem is called stiff when the timescale T of the source term is
much shorter than a typical advective timescale.
Using an over-long timestep may cause plausible but spurious numerical
behaviour:
“The numerical results presented above indicate a disturbing feature of
this problem – it is possible to compute perfectly reasonable results that
are stable and free from oscillations and yet are completely incorrect.
Needless to say, this can be misleading.” [2]
This sounds exactly like pinning.
Random projection method
A deterministic value of φc causes the phase field to become “pinned”
to the lattice. However, taking φc to be a quasi-random variable cho-
sen from a van der Corput sampling sequence [3] greatly reduces this
spurious phenomenon [4].
Fig. 1: Interface propagation speeds as a function of τ/T.
Multiphase LBM
We introduce two discrete Boltzmann equations with distribution func-
tions fi and gi for the fluid flow and phase field, respectively:
∂fi
∂t
+ ci · ∇fi = −
1
τf
fi − f
(e)
i
;
∂gi
∂t
+ ci · ∇gi = −
1
τg
gi − g
(e)
i
+ Ri,
where
f
(e)
i = Wiρ (1 + 3ci · u+
9
2
uu + σI
1
3
(|ci|2 + |n|2) − (ci · n)2
: (cici −
1
3
I)
,
g
(e)
i = Wiφ (1 + 3ci · u) ,
Ri = −WiS(φ) (1 + 3ci · u) .
Here, ρ =
P
i fi, ρu =
P
i fici, φ =
P
i gi, S(φ) =
P
Ri, σ is the
surface tension and n = ∇φ/|∇φ| is the interface unit normal, which is
approximated using central differences. The surface tension is included
directly into the equilibrium distribution.
Macroscopic equations
Applying the Chapman-Enskog expansion to the fluid equation yields
∂ρ
∂t
+ ∇ · ρu = 0,
∂ρu
∂t
+ ∇ · (PI + ρuu) = ∇ ·
µ
∇u + (∇u)†
+σ∇ · (I − nn) δs + O(Ma3),
where µ = τfρ/3 is the dynamic viscosity and δs is the surface delta
function. In the small Mach number limit, these are the continuous
surface force equations of Brackbill et al. [5]. Note that the surface
tension coefficient in the macroscopic equations is precisely σ from the
discrete Boltzmann equation.
Applying the Chapman-Enskog expansion to the phase field equation
yields
∂φ
∂t
+ ∇ · (φu) = S(φ) + τg
1
3
I − uu
: ∇∇φ + O(τ2
g).
To leading order this is the sharpening equation with an additional O(τg)
diffusive term, which sets the width of the interface.
Fully discrete form
The previous two discrete Boltzmann equations can be discretised in
space and time by integrating along their characteristics to obtain a
second-order algorithm:
fi(x + ci∆t, t + ∆t) − fi(x, t) = −
∆t
τf + ∆t/2
fi(x, t) − f
(e)
i (x, t)
,
gi(x + ci∆t, t + ∆t) − gi(x, t) = −
∆t
τg + ∆t/2
gi(x, t) − g
(e)
i (x, t)
+
τg∆t
τ + ∆t/2
Ri(x, t),
where the following change of variable has been used to obtain an explicit
second order system at time t + ∆t:
fi(x, t) = fi(x, t) +
∆t
2τf
fi(x, t) − f
(e)
i (x, t)
,
gi(x, t) = gi(x, t) +
∆t
2τg
gi(x, t) − g
(e)
i (x, t)
−
∆t
2
Ri(x, t).
However, we need to find φ in order to compute f
(0)
i and Ri.
Reconstruction of φ
The source term does not conserve φ:
φ̄ =
X
i
gi =
X
i
gi −
∆t
2τ
gi − g
(e)
i
−
∆t
2
Ri
= φ +
∆t
2T
φ 1 − φ
φc − φ
= N(φ).
We can reconstruct φ by solving the nonlinear equation N(φ) = φ̄ using
Newton’s method:
φ → φ −
N(φ) − φ̄
N0(φ)
.
For the random projection case, we must convert from gi to the gi using
the threshold from the previous timestep, φ
(n−1)
c , and then from gi to
the gi using the current threshold φ
(n)
c .
Numerical Results
Fig. 2: Numerical validation of the proposed model.
Left: Numerical validation of layered Poiseuille flow. The symbols are the
numerical predictions and the solid line is the analytic solution.
Right: Laplace’s law. The symbols are the numerical predictions. The
slope of the curve is σ.
Fig. 3: Numerical simulation of viscous fingering. Here the capillary num-
ber is Ca = 0.1 and the viscosity ratio is 20. The interface thickness is
less than 3 grid cells.
Conclusions
We have presented a continuum-based multiphase lattice Boltzmann
method where the surface tension is included directly into the momen-
tum flux tensor, eliminating the need for an additional “body force”
obtained though spatial derivatives of a source term. This removes the
viscosity-dependence from the surface tension which would otherwise
appear due to the second-order moments in the Chapman-Enskog ex-
pansion. Separation of phases is maintained by an additional equation
for the advection and sharpening of a phase field [4].
The numerical scheme is validated against the analytical solution of
layered Poiseuille flow. Preliminary investigations of viscous fingering
confirm the method’s ability to maintain sharp interfaces, free from
lattice pinning. Moreover, verification of Laplace’s law confirms that
the parameter σ is precisely the value of the surface tension.
References
[1] M. Latva-Kokko and D. H. Rothman, Diffusion properties of gradient-
based lattice Boltzmann models of immiscible fluids, Phys. Rev. E, 71
(2005), 056702
[2] R. J. LeVeque and H. C. Yee, A study of numerical methods for
hyperbolic conservation laws with stiff source terms, J. Comput. Phys.,
86 (1990), 187-210
[3] W. Bao and S. Jin, The random projection method for hyperbolic
conservation laws with stiff reaction terms, J. Comput. Phys., 163
(2000), 216-248
[4] T. Reis and P. J. Dellar, A volume-preserving sharpening approach for
the propagation of sharp phase boundaries in multiphase lattice Boltz-
mann simulations, Comp. Fluids, 46 (2011), 417-421
[5] J. U. Brackbill, D. B. Kothe and C. Zemach, A continuum method
for modeling surface tension, J. Comput. Phys. 100 (1992), 335-354](https://image.slidesharecdn.com/dsfdposterreisv2-210201031028/85/A-multiphase-lattice-Boltzmann-model-with-sharp-interfaces-1-320.jpg)
A multiphase lattice Boltzmann model with sharp interfaces
- 1. A multiphase lattice Boltzmann model with sharp interfaces Tim Reis Oxford Centre for Collaborative Applied Mathematics (OCCAM) http://www.maths.ox.ac.uk/occam/ This poster is based on work supported by Award No. KUK-C1-013-04 made by King Abdullah University of Science and Technology (KAUST) Introduction Multiphase lattice Boltzmann models usually fall into one of three cate- gories: Colour gradient, or Rothman-Keller, models; Shan-Chen models; or free-energy models. The Free-Energy and Shan-Chen models are in- tended to simulate flows where the kinematics of phase separation or interactions at the molecular level are of interest. These models do not allow for independent adjustment of either the surface tension or the viscosity ratio and the resolved interface is spread over several lattice nodes. Here we are concerned with macroscopic, immiscible, flows and therefore pursue colour-based models since they offer the most attrac- tive features to the modeller who wishes to simulate binary flows on the length scale addressed by continuum mechanics. We introduce a phase field (or colour field) φ to represent the differ- ent fluids. Regions where the phase field φ ≈ 0 might represent one phase (e.g oil) and regions where φ ≈ 1 the other phase (e.g water). In principle, φ is advected by the fluid velocity, hence φt + u · ∇φ = 0. However, to make the most of limited spacial resolution it is common to add a “sharpening term” to counteract the inevitable numerical dif- fusion and preserve relatively thin boundaries between different phases. For example, the sharpening term might drive all points where φ > 1 2 back towards φ = 1, and all points where φ < 1 2 back towards φ = 0. The width of the phase boundary is then controlled by a balance be- tween diffusion and the sharpening term. However, overly narrow phase boundaries can fail to propagate correctly, becoming fixed or “pinned” to the grid [1]. Phase field sharpening and pinning A useful model for studying hyperbolic equations with ”stiff” source terms is [2]: ∂φ ∂t + u ∂φ ∂x = S(φ) = − 1 T φ 1 − φ φc − φ , The problem is called stiff when the timescale T of the source term is much shorter than a typical advective timescale. Using an over-long timestep may cause plausible but spurious numerical behaviour: “The numerical results presented above indicate a disturbing feature of this problem – it is possible to compute perfectly reasonable results that are stable and free from oscillations and yet are completely incorrect. Needless to say, this can be misleading.” [2] This sounds exactly like pinning. Random projection method A deterministic value of φc causes the phase field to become “pinned” to the lattice. However, taking φc to be a quasi-random variable cho- sen from a van der Corput sampling sequence [3] greatly reduces this spurious phenomenon [4]. Fig. 1: Interface propagation speeds as a function of τ/T. Multiphase LBM We introduce two discrete Boltzmann equations with distribution func- tions fi and gi for the fluid flow and phase field, respectively: ∂fi ∂t + ci · ∇fi = − 1 τf fi − f (e) i ; ∂gi ∂t + ci · ∇gi = − 1 τg gi − g (e) i + Ri, where f (e) i = Wiρ (1 + 3ci · u+ 9 2 uu + σI 1 3 (|ci|2 + |n|2) − (ci · n)2 : (cici − 1 3 I) , g (e) i = Wiφ (1 + 3ci · u) , Ri = −WiS(φ) (1 + 3ci · u) . Here, ρ = P i fi, ρu = P i fici, φ = P i gi, S(φ) = P Ri, σ is the surface tension and n = ∇φ/|∇φ| is the interface unit normal, which is approximated using central differences. The surface tension is included directly into the equilibrium distribution. Macroscopic equations Applying the Chapman-Enskog expansion to the fluid equation yields ∂ρ ∂t + ∇ · ρu = 0, ∂ρu ∂t + ∇ · (PI + ρuu) = ∇ · µ ∇u + (∇u)† +σ∇ · (I − nn) δs + O(Ma3), where µ = τfρ/3 is the dynamic viscosity and δs is the surface delta function. In the small Mach number limit, these are the continuous surface force equations of Brackbill et al. [5]. Note that the surface tension coefficient in the macroscopic equations is precisely σ from the discrete Boltzmann equation. Applying the Chapman-Enskog expansion to the phase field equation yields ∂φ ∂t + ∇ · (φu) = S(φ) + τg 1 3 I − uu : ∇∇φ + O(τ2 g). To leading order this is the sharpening equation with an additional O(τg) diffusive term, which sets the width of the interface. Fully discrete form The previous two discrete Boltzmann equations can be discretised in space and time by integrating along their characteristics to obtain a second-order algorithm: fi(x + ci∆t, t + ∆t) − fi(x, t) = − ∆t τf + ∆t/2 fi(x, t) − f (e) i (x, t) , gi(x + ci∆t, t + ∆t) − gi(x, t) = − ∆t τg + ∆t/2 gi(x, t) − g (e) i (x, t) + τg∆t τ + ∆t/2 Ri(x, t), where the following change of variable has been used to obtain an explicit second order system at time t + ∆t: fi(x, t) = fi(x, t) + ∆t 2τf fi(x, t) − f (e) i (x, t) , gi(x, t) = gi(x, t) + ∆t 2τg gi(x, t) − g (e) i (x, t) − ∆t 2 Ri(x, t). However, we need to find φ in order to compute f (0) i and Ri. Reconstruction of φ The source term does not conserve φ: φ̄ = X i gi = X i gi − ∆t 2τ gi − g (e) i − ∆t 2 Ri = φ + ∆t 2T φ 1 − φ φc − φ = N(φ). We can reconstruct φ by solving the nonlinear equation N(φ) = φ̄ using Newton’s method: φ → φ − N(φ) − φ̄ N0(φ) . For the random projection case, we must convert from gi to the gi using the threshold from the previous timestep, φ (n−1) c , and then from gi to the gi using the current threshold φ (n) c . Numerical Results Fig. 2: Numerical validation of the proposed model. Left: Numerical validation of layered Poiseuille flow. The symbols are the numerical predictions and the solid line is the analytic solution. Right: Laplace’s law. The symbols are the numerical predictions. The slope of the curve is σ. Fig. 3: Numerical simulation of viscous fingering. Here the capillary num- ber is Ca = 0.1 and the viscosity ratio is 20. The interface thickness is less than 3 grid cells. Conclusions We have presented a continuum-based multiphase lattice Boltzmann method where the surface tension is included directly into the momen- tum flux tensor, eliminating the need for an additional “body force” obtained though spatial derivatives of a source term. This removes the viscosity-dependence from the surface tension which would otherwise appear due to the second-order moments in the Chapman-Enskog ex- pansion. Separation of phases is maintained by an additional equation for the advection and sharpening of a phase field [4]. The numerical scheme is validated against the analytical solution of layered Poiseuille flow. Preliminary investigations of viscous fingering confirm the method’s ability to maintain sharp interfaces, free from lattice pinning. Moreover, verification of Laplace’s law confirms that the parameter σ is precisely the value of the surface tension. References [1] M. Latva-Kokko and D. H. Rothman, Diffusion properties of gradient- based lattice Boltzmann models of immiscible fluids, Phys. Rev. E, 71 (2005), 056702 [2] R. J. LeVeque and H. C. Yee, A study of numerical methods for hyperbolic conservation laws with stiff source terms, J. Comput. Phys., 86 (1990), 187-210 [3] W. Bao and S. Jin, The random projection method for hyperbolic conservation laws with stiff reaction terms, J. Comput. Phys., 163 (2000), 216-248 [4] T. Reis and P. J. Dellar, A volume-preserving sharpening approach for the propagation of sharp phase boundaries in multiphase lattice Boltz- mann simulations, Comp. Fluids, 46 (2011), 417-421 [5] J. U. Brackbill, D. B. Kothe and C. Zemach, A continuum method for modeling surface tension, J. Comput. Phys. 100 (1992), 335-354