/ A | Lue | AIT A lattice Boltzmann model for multiphase fluid flows Daryl Grunau, Shiyi Chen, and Kenneth Eggert Citation: Physics of Fluids A: Fluid Dynamics (1989-1993) 8, 2567 (1993); doi: 10.1063/1.858769 jew online: hitp://dx.doi.org/10.1063/1 858769 View Table of Contents: http://scitation ip. org/content/aip/journal/pofa/S/10?ver=pdfcov Published by the AIP Publishing Articles you may be interested in Simulations of viscous flows of complex fluids with a Bhatnagar, Gross, and Krook lattice gas Phys. Fluids 8, 3200 (1996); 10.1063/1.869093, ‘Anonslip boundary condition for lattice Boltzmann simulations Phys. Fluids 7, 2928 (1995); 10.1063/1.868766 Prandtl number of lattice Bhatnagar-Gross-Krook fluid Phys. Fluids 7, 2280 (1995); 10.1063/1.868771 A consistent hydrodynamic boundary condition for the lattice Boltzmann method Phys. Fluids 7, 203 (1995); 10.1063/1.868767 Lattice Boltzmann magnetohydrodynamics Phys. Plasmas 1, 1850 (1994); 10.1063/1.870640A lattice Boltzmann model for multiphase fluid flows Daryl Grunau,*””” Shiyi Chen,®° and Kenneth Eggert” Los Alamos National Laboratory, Los Alamos, New Mexico 87545 (Received 27 January 1993; accepted 24 May 1993) A lattice Boltzmann equation method for simulating multiphase immiscible fluid flows with variation of density and viscosity, based on the model proposed by Gunstensen et al. for two-component immiscible fuids [Phys, Rev. A 43, 4320 (1991) is developed. The numerical measurements of surface tension and viscosity agree well with theoretical predictions. Several basic numerical tests, including spinodal decomposition, two-phase fluid flows in ‘two-dimensional channels, and two-phase viscous fingering, are shown in agreement of experiments and analytical solutions. |. INTRODUCTION It has recently been shown that lattice gas computa- tional fluid dynamics, including lattice gas automata (LGA)! and the lattice Boltzmann (LB)? equation meth- ods, provide alternative numerical techniques for solving the Navier-Stokes equations, multiphase fui flows,” and ‘a variety of other fiuid systems.” The parallel nature of ‘these newly developed schemes, adapted from cellular au- tomata, affords an casy implementation of fast, efficient, ‘and accurate simulations on parallel machines. ‘Ta solve incompressible Aid Awe hy traditional nie merical methods such as finite differences or finite ele- ‘ments, one must deal with a Poisson equation for the pres- ssure term that is induced by the continuum condition and ‘the momentum equation. Solution of the Poisson equation can be avoided by relaxing the incompressibiity require- ‘ment, and penalizing errors in velocity divergence." Lattice ges methods are most closely compared Ww Uiese ‘pseudocompressible algorithms for solving incompressible fluid flows. However, penalty of nonvanishing velocity gra dients appear in the macroscopic equations as a result of microscopic particle dynamics. A lattice gas solver is only required to solve the kinetic equation of the particle distri- bution function. All other quantities, including density, ve- locity, and energy can be obtained by a macroscopic aver- aging process through the distribution function. Thus, the [pressure effects on the momentum equation are controlled by an equation of state. Since lattice gas algorithms proceed at a microscopic scale, compressible effects on the macroscopic fiuid behav- jar ean haceme significant Nonetheless, the incampreasible limit theory of lattice gas methods? and their numerical simulations,’ predict and demonstrate that such a lattice g28 solution at low Mach number produces results compa- rable with those of the incompressible Navier-Stokes equa- tions for a wide variety of problems and with very small numerical errors. Also, two-phase surface dynamic bound- aries and wall boundaries arc far more casily implemented with a lattice gas method compared to direct simulations of the incompressible Navier-Stokes equations? ‘Center for Nonlinear Stade, 2557 Phys. Fhids A 5 (10), Ootober 1993 (0899-8219/98/5(10)/2557/6/$6.00 Rothman and Keller’ were the first to extend the single-phase lattice gas model proposed by Frisch, Has- slacher, and Pomeau (sHP)' to simulate multiphase tuid problems. Colored particles were introduced to distinguish Vetwecu plaors, aud a ucarcotucightos partite intcroction ‘was used to facilitate interfacial dynamics, such as La- place's formula for surface tension. Later, Somers and Rem* and Chen et al? extended the original colored par- ticle scheme by introducing colored holes. It has been shown’ that the colored-hole lattice gas method extends the original nearest-neighbor particle interaction to several lattice lengths, leading to a Yukawa potential. Moreover, the colored-hole scheme carties purely local information in its particle collision step, reducing the size ot the lookup table in the algorithm, and consequently speeding up the siuuulation. Although two-phase lattice gas algorithms are able to produce inter ting curface phenomena, they are dificult to compare quantitatively with experiments and other nu- merical simulations due to their noisy characteristics in- duced by particle fluctuations. The lattice Boltzmann ‘model proposed by McNamara and Zanetti” however, solves the kinetic equation for the particle distribution in- stead of tracking each particle's motion, as is done in lat= tice gas methods. Nonetheless, lattice gas algorithms will bbe superior to lattice Boltzmann models for simulating un- derlying physies and modeling microscopic dynamics, suck as correlation effects and phase transitions, since they con- tain-more information shout the micrascapic behavior of particles. On the other hand, to provide a numerical ‘method for solving partial differential equations governing macroscopic behavior, lattice Boltzmann schemes will be at least as good as lattice gas models. The finite-dfference nature of lattice Boltzmann methods not only can simulate the macroscopic equations more efficiently and accurately, Dut also preserve some of dhe advantages of Taltive gas ‘models, such as their parallel computing nature and their ease of boundary implementation. ‘Combining the lattice Boltzmann ideology of McNa- mara and Zanetti? with the linearized collision operator proposed by Higuera and Jimenez," Gunstensen et al" proposed a lattice Boltzmann method for solving two- phase fluid flows based on the Rothman and Keller lattice gas model.’ An important contribution of this mode is the © 1999 American Insitute of Physics 2557introduction of a perturbation step (shown in detail below) so that Laplace's formula at an interface can be approxi- ‘mately recovered. This lattice Boltzmann method has been used in several applications,!” however, it has a few funda- menial problems. First, the model does not solve the exact two-phase fiuid equations—although Galilean invariance is recovered by the proper assignment of rest particles, the equation of state remains to be velocity dependent."? Sec- cond, the model uses a fully linearized collision operator, which involves a 24X24 matrix multiplication at each time step and position in three-dimensional space. reducing computational efficiency. Finally, the model does not in- clude two-phase fluid flows that have variable densities and viscosities. In this paper, we extend their model by using a single-time relaxation lattice Boltzmann model. With the proper choice of the particle equitibriuin distribution fun tion, we are able to recover the incompressible Navier— Stokes equations for the color-blind fluid. In addition, our new model makes provision for variable densities and vis- cosities by use of the freedom of the rest particle equilib- rium distribution, and a space-dependent relaxation pro- cess. For simplicity, we present a two-phase, two- dimensional fluid model on a hexagonal lattice in this, paper. The extension of the current model to multiphase fluids and other iattices, inciuding a two-dimensional square lattice, a three-climensional face-centered hypercu- bic lattice,” and a body-centered-cubie" lattice, will be pre- sented in detail in another paper."* Ul, NUMERICAL MODEL Denote F(x.1), f1?(x, and f(x) as the particle fle D Fey. a ie Here p, and py are densities of the red and blue fuids, respectively, p=p,+ ps is the total density, and v is the local velocity Following the pressure-corrected Galilean-invariant single-phase lattice Boltzmann model proposed by Ref. 13, wwe assume the following equilibrium distribution for both red and blue fluids: pace 0 gaat y or 45 eon) —G 0) fee-olsta8} ne. — J ean az tern? =o Ses ty ONG ee 1 Sm, f-n(ss-*) Here we have introduced two real-valued parameters, m, and my. They can be understood as the ensemble-averaged number of degenerate rest particle states for the red and blue fuids, respectively. To achieve a stable interface, we Furthermore assume that te moving particle distributions for both red and blue fluids are the same when v=0:, ie, d= po/(6-1,)— py/(6-+7,)- This implies the following density ratio: b,_S+m, Ye omy ‘The second part of the collision operator is similar 10 that given in the Gunstensen eta. paper: (Of = (4/2) /F | [Cer FY°/|EPP—H, where F is the local color gradient, defined as F(x) = Lelpdx+e)—pr(x+en] Note that in a single-phase region of our incompressible fluid model, F=0. Thus the second term of the collision operator, (#7, only has a contribution at two-phase in- terfaces. The parameter 4; is a free parameter, controlling the surface tension (shown below). For simplicity, we choose 4,=4,=< in this paper. Gru 2558 Chen, and EggertTo maintain surfaces between fluids, we follow Roth- man’s scheme’ to redistribute colored particles at two- phase interfaces (without changing the total particle dis- tribution, /)) by enforcing the red color momentum §'=3,fle,, to align with the direction of the local color gradient. In other words, we redistribute the red density at an interface to maximize the following quantity: —(j"-F). ‘The blue particle distribution can then be recovered using Simfi-Si. ‘To derive hydrodynamics, we use a long-wavelength, lowefreqiency approximation and a multiscaling analysis Here f and & represent fast and slow time scales, respec- tively, and eis assumed to be a small expansion parameter. ‘A Taylor series expansion of Eq. (1) to second order in the lattice spacing and time step gives the following continuum kinetic equation: aft 1 a a Liew ihazeevv rt renv Sf 109, +37 @ Taking the zeroth- and first-order moments of e, over Eq. (2), we readily obtain the continuum equations for the red and blue fluids, and the following momentum equation for the color-blind (total) fluid: 2 2 B cedalede| %+(i-ge\ A] 2, where ff ig the nextorder perturbation for ft and Cian) A is isibuted io the yeonnlondes amp nents of (2). Note that since 2 ah*= E (ah*e=0 and FalV(p—py) to ~e+ oe), the second term in the collision operator does not contnib- lute to the continuum and momentum equations of the first-order apprusiaation. Tt, however, will eoutiibute to the pressure term at an interface: P=P)-+e|¥F|, where Py ‘comes from the equation of state. Our two-phase immiscible fluid model contains three fluid regions: a red or blue homogeneous region and a thin region, where the two fluids mix at the interfaces. In a homogeneous region, the evolution of our model will re- cover the Navier-Stokes equations with kinematic viscos- ity of y=(r,-1)/8 and a sound speed of 2559 Phys. Fuids A, Vol 5, No. 10, October 1998 ex = \3/(6-+m4) according to the Chapman-Enskog ex- pansion shown in Ref. 14. A variation of viscosities for the two fluids can be obtained by choosing different 7. ‘The last issue remaining to be addressed is the interfa- cial dynamics that take place in a region where red and blue fluids are adjacent. Usually the thickness of such an interface will depend on an averaged relaxation time and the rest particle distribution. There are several ways to ‘construct a relaxation parameter so that the red and blue fluids have a smooth change of viscosity at their interfaces. To do this, we define an order parameter, y, depending on red and blue densities as follows: J=(p,—ps)/(p,+ Ps) Note that, in general, [| <1, however, in purely red iid region, Y=1, and in a purely blue fluid region, y= —1. To ‘continuously connect relaxation parameters 7, and 7, at an interface, we employ the following simple formula: tm $8 Je), S>¥>0, gH), O>d>—6, tm <8, where g,() and gy() are second-order functions of y= 8) = a+ BYP aud go =atmb+ee Assuming that g,(5)=7,5 g(—-3)=T5 91/46=0 at [Wl-8 and“ g,(0)—g9(0)—=(7), wo have QA Tt» B=2U7,—a)/b, B/(28), (a—7,)/6, and E=7/(28). Here 5<1 is a free pa- rameter controlling the interface thickness and (r) is the averaged relaxation time step across the interface, (1) =2r ir e+ 79) Using the defini siven in Ref. 19, mn of mechanical surface tension J ceemroax, @) and, after some algcbra,'® we obtain a theoretical formula for the surface tension on a two-dimensional hexagonal lattice: @=94(r)d(12-+m,-+m,). Here the integration is over whole space along the direction perpendicular to a given interface, and P, and P, are, respectively, the normal and tangential stress tensor at the two-phase interface. In Fig. 1, we show the theoretical prediction (—) and numer- teal measurements (() and + ) O1 ¢ asa runction ot particie density d with a mass ratio, 7=}. The () symbols represent 1 measurements of surface tension using the above definition, and the + signs represent the surface tension obtained by the Laplace formuly AP==2a/R. The mechane ical surface tension was obtained by numerical integration of (3) across an interface, whereas surface tension from Laplace's law was inferred from the slope of AP plotted against 1/R for radii ranging from 20 to 40. As can be seen, ‘the theoretical prediction and numerical measurements agree very well in both cases. ‘Grunau, Chen, and Eggert 25690.08 0.08 IG. 1. The theoretical prediction (—) and nomeceal measurements (0 and +) of surface tension o a function of particle density d with a mass I, NUMERICAL SIMULATIONS AND DISCUSSION To demonstrate the application of the presented lattice Dolizauauns wel, we show duce numesical exaunples be- low. The first numerical test isthe study of the phase seg- regation procese by twe dimensional epinadal decomaposi tion when the fluids have differing densities. Figure 2 is a snapshot (¢=8300) of the two-phase area distribution, when the density ratio, y=p,/ps, is 2.3. One can sce that the red fuid (shown as black) represents a high-density region and the blue uid (shown as white) represents a low-density region. The initial condition for this simulation is constant density and random color distsibution with pe- riodie boundaries. Here the surface tension parameter is set to 4—001. The current model preserves the basic two- phase segregation processes and, in simulations not shown haere, we have seen the preservation of stable two-phase interfaces for much higher density ratios (~200). This capability will allow us to approach the simulation of gas and oil mixing flows. ‘The second numerical testis two-dimensional channel flow with two Muids Inving diferent visenilin. A lattice length of 65 in the y direction (W=65) and 128 in the x spinodal decomposition simulation, where the two fuids have a 23:1 tase rato, y. The boundaries are periodic, and the inital red and blue SSnaty i sired « cuinion viata valve we avalon Ct ee ‘riution, The surface tension parameter Is asigned to 4=001 2560 Phys. Fluids A, Vol. 5, No. 10, October 1083 0.96 — vanory re a FIG. 3. The analytical prediction (—) and direct aumericalsimelaion (0) of velocity s a fonction of channel width. direction (Z=128) is used, with a nonslip condition on ‘both the upper and bottom boundaries. Initially, the lower half space (from lattice 1-32) is filled with red fuid, and the top half space (from lattice 34-65) is filled with blue fluid. The middle line is initialized to half red and half blue, and the relaxation parameters for red and blue fluids are assigned values of r,=2 and ry=I. Using the viscosity formula discussed above, this translates into a kinematic viscosity ratio of 3:1 for the red and blue fiuids, respec- tively. A forcing technique, as described in Ref. 13, is used to establish a small pressure gradient across the length of the channel. Assuming a thin interface in the middle, an analytical solution of the velocity distribution as a function of the channel width, y, can be derived, centered on the middle of the channel:"* (Po= Pru? ( ur BL Pr Ho Br Boe i aap ( iy te Hty Y Daag Miata oaks where (Pj—P;)/L. is the pressure gradient across the channel, 1, and p15 are shear viscosities for red (0.7875) and blue (0.2625) fluids, respectively. Here, also, w=V3W/4 is half the channel width, where the factor 5/2 is « hexagonal lattice effect. In Fig. 3 we present the analytical prediction (—) and direct numerical simulation (©) for velocity as a function of channel width y. As seen, Use axgiceanent is excellent. The last numerical test is a two-dimensional Hele~ Shaw viecous fingering experiment. The upper and lower walls are assigned a no-slip boundary condition, and the fluids are given a viscosity ratio of 1:10 by assianing 7,—1 and 7y=5.5. To develop a Hele-Shaw pattern, an ini perturbation is introduced at the interface,” which then forced by maintaining a high pressure at the inlet and low pressure at the outlet. The shape of the finger can be altered by a change in the surface tension pataneter A, given as 0.0065 in this experiment. Figure 4 shows the evolution of n lest viscous fluid penetrating one having « higher viscosity at times=0, 2000, 4000, 6000, 8000, and Grunau, Chen, and Eggert 2560ra FIG. 4, The formation of stable finger ina Hele-Shaw cela times, 2000, 4000, 6000, 000, and 10.000. A viscosity ratio of 1:10 is established by acting rym} ond ry5.5 The low vzcoety Mud ie paneteting the Highersviscosty region 10.000. We see that a stable finger develops and is main- tained throughout the simulation, although we expect a tip-splitting instability to occur for smaller values of , as demonstrated by Ket. 22. This phenomenon will be dis- cussed in more detail in a subsequent paper. The fluid paltesns shown here are similar to results obtained by other ‘numerical methods? ‘The lattice Roltymann model in this paper is the ex- tension of the model proposed by Gunstensen ef al. The current model gives an exact Navier-Stokes solution, in the incompressible limit, for each fuid individually. The sur- face tension at an interface satisfies the Laplace formula, and the model has the capability to simulate two-phase fluid flows with different densities and viscosities. The sim- ulation tesults for several typical tworphase Suid flows are shown, in good agreement, with analytical solutions and some experimental observations Several issues, however, remain to be addressed. First, to achieve a density varia- tion, the current model uses the freedom of the rest particle equilibrium distribution. Although the scheme is simple, it is dificult to obtain high velocity flows, since the assign- ment of a high particle mass ratio wall accumulate a large fnumber of particles in the rest state, reducing the fluid speed. Ia addition, the equation of state of the current ‘model is that of an ideal gas similar to the original FHP model. To simulate Aid Rawe with a large density varia- tion, a lattice Boltzmann model with an equation of state for nonideal gases (liquids) may provide some alternatives. Second, the current model uses two parabolic curves (0 match the interface relaxation characteristic time in the mixing region. Kor a very thin mntertace (tor example, one lattice width), the effects of the parameter 5 will be negli- 2581 Phys. Fhids A, Vol 5, No. 10, October 1998, sible, On the other hand, for flows with slow relaxation, the interface will be much wider. The relaxation form and the choice of 6 may affect interfacial dynamics. Other fune- tional forms, such as a binormal distribution, may give better results. ACKNOWLEDGMENTS We thank R. Vaidya, G. D. Dovten useful suggestions. Eugene Loh isa contributor of the orig inal oM43 code This work is performed under the auspices of the United States Department of Energy. by Los Alamos Na- tional Laboratory, under Contract No. W-7405-ENG-36. ‘Numerical simulations were carried out using the compu- tational resources at the Advanced Computing Laboratory at Los Alamos. Partial funding for this work was provided by the Ull Kecovery Lechnology rartnersmp under the US. Department of Energy Office of Fossil Energy, admin- istered by the Dartesvile Project Offic. "U, enh, i Hasslacher, and Y. Pomeau,“Lardce gas automata forthe [Navier-Stokes equation,” Phys. Rev. Let 86, 1505 (1986) 2G.G. MeNamara and G, Zanetti, “Use ofthe Boltzmann equation to ‘moulat latie-gas automata,” Phy, Rev. Lett. 61, 2332 (1988); FJ. Higuera, S. Sues, and R. Beni, “Lattice gs dynamics with enanced collisions." Europhvs, Let. 9.345 (1989) °D. HL Rothman and J. M. Keller, "Tmmiscible cellular automaton fi is" J. Stat. Phys, 82, 119 (1988). STA. Somers and B. Rem, “Analyst of erin tension in tener lattice gases," Physica D 47, 39 (1991). 45. Chen, G. D. Doole, K. Eggert, D. Grunau, and E. ¥. Loh, “Local lattice gus meds for immiseibe ids," Phy. Rev. A 4, 7053 (1901) “G. D. Doolen, in Lattice Gas Methods for PDE’ (Addison-Wesley, Reading, MA, 1989); G. D. Dooln, “Lattice ane methods for PDE'S Thoweysapplcaion sal Daidwaie Physica D 47, No. 1 & 2 (1990), $. Chen, G. D. Doolen, and W. H. Maithaeus, “Lattice gas automata for simple and complex uids.” J. Stat. Phys. 64, 1138 (1991); R. ‘Rape, A. Lavinezak, and ¥. assiar, “Useulaions and waves in a reactive atice-ga8sulomaton,” Phys. Rey Let, 66, 2539 (1991). SS. D.. Ramshaw and GT. Mesina, “A” hybrid penalty. ‘pscudccompressblty method for transient incompressible fui No Comput. Fluids 20, 168 (1991). °G, Frisch. D, d'Homires, B. Hasslacher,P. Lallemand, Y. Pomeau, and J.P. Rivet, “Lattice gas hydrodynamics in two and three dimen sions," Complex Syst. 1, 649 (1987), \WB. Higuera and. Tintner,“Roltymann appenach to Inioe gx sim lations," Burophys. Let. 9,663 (1989). ACK. Gunstersen, D. H. Rothms S. Zaleski, and G. Zaneti, “Lattice Dattzmann model of imenissible tui.” Phys Rav. A 48,1320 (1001) A" K. Gunstensen, PhD. dissertation, 1992 HL Chen, 8. Chen, and W. HL Matthaeus, “Recovery of the Navier= ‘Stokes ejvations ing laltice gan Doltanann wor” Phys. Ress A 45, R5339 (1991). MS Chen, J. Wang, X. Shan, and G. Doolen, “Lattice Boltzmann eom- ‘tational Auta dypamice ares dimensions” J. Stl. Phys. 68 379 (498). "8D. Gronaa, S. Chen, K. Eggert and G. Doslen (in preparation). HL. D. Chen, D. Martinez, and W. Matthoeas, “Lattice Doltzmann model for simulation of megnetohydrodynamies," Phys Rev. Lett. 67. 3776 11991) "8Y"'HL Qian and P. Lallemand, “Lattice BGK models for Navier-Stokes arophys. Lett. 17, 479 (1992) Grunau, Chen, and Eggert 2501"R.J, Alexander, &. Chen, and J.D. Steding, “Lattice Boltzmann ther- mokydrodynamics,” Phys. Rev. E. $7, R2279 (1993) S"Rowiinom and B. Walom, Moiculor Theory of Copllanty (laren on, Oxford, 1982) R. B, Bind, W. E- Stewart, and E. N. Lightfoot, Transpo Phenomena ‘vies, New York, 1960) MD, Burgess and F. Hayot,"Safiman-Taylor-ype instability in lattice 2562 Phys. Fluids A, Vol 5, No. 10, October 1003 257" Phys. Rev. A 40, S187 (1989); B. Meilburg and G. M, Homsy, “Nonlinear unstable viscous fingers in Hele-Shaw Hows. 1. Numerical simulation,” Phys. Fluids 3, 429 (1988), A. 5. DeGreporia and I. W. Schwaets, “A boundary-integral method for two-phase displacement in Hele-Shaw cll" J. Kua Meet, 10%, 383 (1986), Grunau, Chen, and Eggert 2562