1: \documentstyle[aps,preprint,epsfig]{revtex}

2: \tightenlines

3: \begin{document}

4: \draft

5: \title{Phase separation of incompressible binary fluids with 

6: Lattice Boltzmann Methods}

7: \author{ Aiguo Xu$^{1}$, G.  Gonnella$^{1,2}$ and  A. Lamura$^{3}$}

8: \address{

9: $^{1}$ Istituto Nazionale per la  Fisica della Materia, Unit\`a di Bari,

10: {\rm and} Dipartimento di Fisica, Universit\`a di Bari, {\rm and}

11: TIRES, Center of Innovative Technologies for Signal Detection

12: and Processing,

13:  via Amendola 173, 70126 Bari, Italy\\

14: $^{2}$ INFN, Sezione di Bari,  via Amendola 173, 70126 Bari\\

15: $^{3}$ Istituto Applicazioni Calcolo, CNR, Sezione di Bari,\\

16: Via Amendola 122/I, 70126 Bari, Italy }

17: \date{\today}

18: \maketitle

19: \begin{abstract}

20: 

21: We introduce new versions of 

22: lattice Boltzmann methods (LBM) for incompressible binary mixtures

23: where fluctuations of total density are inhibited. 

24: As a test for the improved algorithms we consider the  problem of

25: phase separation of two-dimensional binary mixtures 

26: quenched from a disordered state into the coexistence region.

27: We find that the stability properties of LBM

28: are greatly improved. The control of density fluctuations

29: and the possibility of running longer simulations

30: allow a more precise evaluation  of the growth exponent. 

31: 

32: \end{abstract}

33: \vskip 0.5cm

34: \pacs{KEY WORDS: Lattice Boltzmann methods; binary mixtures; phase

35: separation.}

36: 

37: \section{Introduction}

38: 

39: Lattice Boltzmann methods (LBM) are computational schemes for

40: solving the macroscopic equations of fluid systems \cite{Chen98,succi}.

41: In the case of binary  mixtures they have been largely 

42: used to study  the dynamics of  systems 

43: described by  the Navier-Stokes and convection-diffusion equations \cite{Yeomans}. 

44: The advantages 

45: of LBM compared with standard  methods for partial differential equations

46: are particularly relevant in problems involving complex boundaries such as

47: multiphase flow in porous media \cite{olson,Chen98,succi}. 

48: The large number of applications of LBM in

49: studies on  complex fluids \cite{gonnella,gompper,lamepl}, 

50: polymer mixtures \cite{Malevanets99}, liquid crystals \cite{yeom},

51: etc. demonstrates the versatility of the method.

52: In studies of spinodal decomposition of binary mixtures

53: a version of LBM based on a free-energy approach,

54: where the equilibrium distributions can be consistently chosen 

55: with the thermodynamics of the system, has been largely used \cite{Orlandini,Swift,catesjfm}.

56: 

57: The stability properties of LBM are always crucial,

58: especially in phase separation studies where long lasting 

59: simulations are needed to establish the growth properties.

60: Indeed, the average size of domains of the two phases 

61: generally grows  with power law behavior and one wants to determine

62: the exponent \cite{Furukawa85,Bray94}.

63: However, previous studies showed the existence of  a stability problem:

64: total density fluctuations can suddenly grow in an uncontrolled

65: way interrupting simulations and

66: making problematic the determination of the growth exponent \cite{catesjfm}.

67: 

68: In this paper we consider new versions of LBM 

69: where  the stability  of the algorithm is improved.

70: We have checked the stability properties by studying 

71: the problem of phase separation of two-dimensional binary mixtures.

72: This is a severe test for stability since

73: density fluctuations are enhanced by the presence of many 

74: interfaces in the system though they should be negligible in incompressible

75: fluids. We find that,

76: with the new algorithms, density fluctuations are largely suppressed and 

77: a  more reliable determination of the growth exponent

78: can  be  obtained.

79: 

80: The  incompressible Navier-Stokes equation can be obtained

81: from lattice Boltzmann equations if density

82: fluctuations are negligible \cite{Chen98}. 

83: Compressibility effects, intrinsic to LBM, 

84: can produce some serious errors in numerical 

85: simulations of incompressible fluids.    

86: Due to  these reasons improved versions of LBM

87: have been  introduced for the case of a simple fluid \cite{zou95,dembo}.

88: New schemes are based on the 

89: explicit elimination in the equilibrium distributions of

90:  higher order terms in the Mach number \cite{heluo} or on the

91: introduction of  a feedback mechanism \cite{fang}.

92: With these improvements it has been shown that 

93: analytical solutions known for particular flows  

94: are better reproduced than with old LBM versions.

95: 

96: The above methods are here

97: developed and introduced for the first time 

98: in the case of binary mixtures.

99: The algorithms are presented in Section II.

100: In Section III we study the

101: problem of phase separation comparing the results

102: obtained with different LBM versions.

103: The role  of boundary conditions on the problem of stability

104: is examined in Section IV. Finally, in  Section V,

105: we discuss our results and draw our conclusions.

106: 

107: \section{The model}

108: 

109: Our simulations are based on the free-energy approach 

110: developed by Orlandini {\it et al.} \cite{Orlandini} and Swift {\it et al.} 

111: \cite{Swift}.

112: In this scheme a suitable free energy is introduced to control the

113: equilibrium properties of the system. 

114: 

115: \subsection{The lattice Boltzmann scheme}

116: 

117: The free-energy functional used in the present study is 

118: \begin{equation}

119: \begin{cal} F \end{cal}= \int d {\bf r}\left[ \frac{1}{3} n \ln n 

120: + \frac{a}{2}\varphi^2

121: +\frac{b}{4}\varphi^4+\frac{\kappa}{2} (\nabla \varphi)^{2}\right]

122: \label{fren}

123: \end{equation}

124: where $n$ is the total density and $\varphi$ is the density difference

125: between the two fluids.

126: The term in $n$ gives rise to a positive background pressure. 

127: The terms in $\varphi$ in the free-energy density $f(n, \varphi)$ correspond 

128: to the usual 

129: Ginzburg-Landau expression typically used in coarse-grained models

130: of phase separation \cite{Bray94}.

131: The polynomial terms relate to the bulk

132: properties of the fluid.  

133: While the parameter $b$ is always positive,

134: $a$ allows to distinguish between a disordered ($a>0$)

135: and a segregated mixture ($a<0$) where two pure phases

136: with $\varphi = \pm \sqrt{-a/b}$ coexist. 

137: We will consider quenches into the coexistence region 

138: with $a<0$ and $b=-a$ so that the equilibrium values for the order 

139: parameter are $\varphi = \pm 1$.

140: The gradient term is related to the surface tension.

141: The other thermodynamic properties of the fluid 

142: follow from the free energy (\ref{fren}).

143: The chemical potential difference between the two fluids is given by

144: $\Delta \mu =  \delta{\cal F}/\delta\varphi=

145: a \varphi + b \varphi^3 - \kappa \nabla^2 \varphi$.   

146: The thermodynamic pressure tensor $P_{\alpha\beta}^{th}$ is given by

147: \begin{eqnarray}

148: P_{\alpha\beta}^{th}&=&

149: \Big ( n \; \frac{\delta{\cal F}}{\delta n}

150: +\varphi \; \frac{\delta{\cal F}}{\delta\varphi} - f(n, \varphi) \Big )

151: \delta_{\alpha\beta}+\kappa \partial_{\alpha} \varphi 

152: \partial_{\beta} \varphi

153: \nonumber \\

154: &=&\frac{1}{3} n \delta_{\alpha\beta}

155: + P_{\alpha\beta}^{chem}

156: \label{pres}

157: \end{eqnarray}

158: where the chemical part is 

159: $\displaystyle P_{\alpha\beta}^{chem} = \Big (\frac{a}{2} \varphi^2 +

160: \frac{3}{4} b

161: \varphi^4 -\kappa \varphi \left( \nabla^2 \varphi \right) -\frac{\kappa}{2}

162: \left( \nabla \varphi \right)^2 \Big ) \delta_{\alpha\beta} + 

163: \kappa \partial_{\alpha} \varphi 

164: \partial_{\beta} \varphi$.

165: 

166: The lattice Boltzmann model is defined on the two-dimensional square lattice

167: with the following nine velocity vectors:

168: ${\bf e}_{0}=(0,0)$, ${\bf e}_{i} = (\cos [(i-1)\pi/2],

169: \sin [(i-1)\pi/2 ]) c$,

170: $i=1,2,3,4$, 

171: ${\bf e}_{i} = (\cos[(i-5)\pi/2+\pi/4],\sin[(i-5)\pi/2+\pi/4]) \sqrt{2} c$,

172: $i=5,6,7,8$, where $c=\Delta x/\Delta t$, and $\Delta x$ and $\Delta t$ are

173: the lattice constant and the time step, respectively.

174: Two sets of

175: distribution functions $f_{i}({\bf r}, t)$ and $g_{i}({\bf r}, t)$

176: evolve according to   

177: a single relaxation-time Boltzmann equation that is discrete in both time and 

178: space \cite{bhatnagar,chen1}:

179: \begin{eqnarray}   

180: f_{i}({\bf r}+{\bf e}_{i}\Delta t, t+\Delta t)-f_{i}({\bf r}, t)&=&   

181: -\frac{1}{\tau}[f_{i}({\bf r}, t)-f_{i}^{eq}({\bf r}, t)], \label{dist1}\\   

182: g_{i}({\bf r}+{\bf e}_{i}\Delta t, t+\Delta t)-g_{i}({\bf r}, t)&=&   

183: -\frac{1}{\tau_{\varphi}}[g_{i}({\bf r}, t)-g_{i}^{eq}({\bf r},   

184: t)], \label{dist2} 

185: \end{eqnarray}   

186: where $\tau$ and  $\tau_{\varphi}$ are independent   

187: relaxation parameters. 

188: The distribution functions are related to the total density $n$,  

189: to the fluid momentum $n {\bf v}$ 

190: and to the density difference $\varphi$ through

191: \begin{equation}   

192: n=\sum_{i}f_{i} , \hspace{1.3cm} n{\bf v}=\sum_{i}f_{i}{\bf e}_{i} ,

193: \hspace{1.3cm}   

194: \varphi=\sum_{i}g_{i} .

195: \label{phys}   

196: \end{equation}  

197: These quantities are locally conserved in any collision process and, 

198: therefore, we

199:  require that the local equilibrium distribution functions 

200: $f_{i}^{eq}({\bf r}, t)$ and 

201: $g_{i}^{eq}({\bf r}, t)$ 

202: fulfill the Eqs.~(\ref{phys}).

203: The higher momenta of the local 

204: equilibrium distribution functions are defined so that the 

205: continuum equations pertinent to a binary fluid mixture can be obtained 

206: \cite{Orlandini,Swift}

207: \begin{equation}   

208: \sum_{i}f_{i}^{eq}e_{i\alpha}e_{i\beta}=c^2 P_{\alpha\beta}^{th}

209: +n v_{\alpha} v_{\beta} \;,   

210: \label{eqn0}

211: \end{equation}  

212: \begin{equation}  

213:  \sum_{i}g_{i}^{eq}e_{i\alpha}=\varphi v_{\alpha} \; ,

214: \label{eqn} 

215: \end{equation}

216: \begin{equation}

217: \sum_{i}g_{i}^{eq}e_{i\alpha}   

218: e_{i\beta}=c^2 \Gamma \Delta\mu\delta_{\alpha\beta}+\varphi   

219: v_{\alpha}v_{\beta} \;.

220: \label{eqn6}  

221: \end{equation}   

222: where $\Gamma$ is a coefficient related to the mobility of the fluid.

223: The local 

224: equilibrium distribution functions can be expressed as 

225: an expansion at the second order in the 

226: velocity ${\bf v}$ \cite{Orlandini,Swift}. 

227: \begin{eqnarray}

228: f_0^{eq}&=& A_0+C_0 v^2 \nonumber\\

229: f_i^{eq}&=& A_I+B_I v_\alpha e_{i\alpha}+C_I v^2+D_I v_\alpha v_\beta

230: e_{i\alpha} e_{i\beta}+ G_{I,\alpha\beta}e_{i\alpha} e_{i\beta}

231: \;\;\;\; i=1,2,3,4 \label{svil1}\\

232: f_i^{eq}&=& A_{II}+B_{II} v_\alpha e_{i\alpha}+C_{II} v^2

233: +D_{II} v_\alpha v_\beta

234: e_{i\alpha} e_{i\beta}+ G_{II,\alpha\beta}e_{i\alpha} e_{i\beta}

235: \;\;\;\; i=5,6,7,8 \nonumber

236: \end{eqnarray}

237: and similarly for the $g_i^{eq}$, $i =0, ..., 8$.

238: The coefficients in the previous expansions are explicitly written

239: for convenience to compare the original scheme with 

240: the alternative ones introduced in the following subsections.

241: A suitable choice is \cite{aiguo}

242: \begin{equation}

243: A_0=\frac{4}{9} n - \frac{5}{6} P_{\alpha\beta}^{chem}\delta_{\alpha\beta}, 

244: \hspace{0.5cm} A_I=\frac{1}{9} n + \frac{1}{6} P_{\alpha\beta}^{chem}

245: \delta_{\alpha\beta}, \hspace{0.5cm} 

246: A_{II}=\frac{A_I}{4}

247: \label{as}

248: \end{equation}

249: \begin{equation}

250: B_I=\frac{n}{3 c^2}, \hspace{0.5cm} 

251: B_{II}=\frac{B_I}{4}

252: \label{bs}

253: \end{equation}

254: \begin{equation}

255: C_0=-\frac{2 n}{3 c^2}, 

256: \hspace{0.5cm} C_I=-\frac{n}{6 c^2}, \hspace{0.5cm} 

257: C_{II}=\frac{C_I}{4}

258: \label{cs}

259: \end{equation}

260: \begin{equation}

261: D_I=\frac{n}{2 c^4}, \hspace{0.5cm} 

262: D_{II}=\frac{D_I}{4}

263: \label{ds}

264: \end{equation}

265: \begin{equation}

266: \hspace{-0.2cm} G_{I,\alpha\beta}\!=\!\frac{P_{\alpha \beta}^{th} 

267: - \frac{1}{2} P_{\sigma \sigma}^{th} \delta_{\alpha \beta}}{2 c^2}, 

268: \hspace{0.5cm}

269:  G_{II, \alpha \beta}\!=\!\frac{G_{I,\alpha\beta}}{4}

270: \label{gs}

271: \end{equation}

272: The expansion coefficients for the $g_i^{eq}$ can be obtained

273:  from the previous ones

274: with the formal substitutions $n \rightarrow \varphi$ and 

275: $P_{\alpha \beta}^{th} \rightarrow \Gamma \Delta \mu \delta_{\alpha \beta}$.

276: The quantities

277:  $P_{\alpha \beta}^{th}$ and $\Delta \mu$, which appear in the coefficients

278: of the equilibrium distribution functions, 

279: can be calculated from (\ref{fren}).

280: 

281: It has been shown in Refs.~\cite{Orlandini,Swift}, 

282: using a Chapman-Enskog expansion \cite{chapman},

283: that the above 

284: described lattice Boltzmann scheme simulates at second order in $\Delta t$

285: the continuity equation

286: \begin{equation}

287: \partial_t n + \partial_{\alpha} (n v_{\alpha}) = 0 + o(\Delta t^2),

288: \label{cont0}

289: \end{equation}

290: and the 

291: Navier-Stokes equation

292: \begin{eqnarray}

293: && \partial_t (n v_{\beta}) + \partial_{\alpha} (n v_{\alpha} v_{\beta}) =

294: - \partial_{\beta} p - \partial_{\alpha} c^2 P_{\alpha\beta}^{chem} 

295: + \nu \partial_{\gamma} \Big ( n (\partial_{\beta} v_{\gamma}+

296: \partial_{\gamma} v_{\beta}) \Big) + \nonumber \\

297: && +3 \nu \partial_{\alpha} \Big( \partial_t P_{\alpha\beta}^{chem}

298: - \partial_{\gamma} (n v_{\alpha} v_{\beta} v_{\gamma})

299: - (v_{\beta} \partial_{\gamma} P_{\alpha\gamma}^{chem}

300: + v_{\alpha} \partial_{\gamma} P_{\beta\gamma}^{chem}) \Big) + o(\Delta t^2)

301: \label{ns0}

302: \end{eqnarray}

303: where the isotropic pressure contribution is 

304: \begin{equation}

305: p=c_s^2 n ,

306: \label{presscal}

307: \end{equation}

308: $c_s=c/\sqrt{3}$ and $\nu=(\tau - 1/2) c_s^2 \Delta t$ being the speed of sound

309: and the kinematic viscosity, respectively.

310: The first line of Eq.~(\ref{ns0}) corresponds to the standard

311: Navier-Stokes equation. The second line contains spurious terms.

312: Nonetheless, these terms, being gradient terms with higher order derivatives, 

313: are negligible compared to the ones in the first

314: line when all hydrodynamic fields vary smoothly on the lattice scale \cite{Swift,catesjfm}.

315: Moreover, a convection-diffusion equation is also recovered:

316: \begin{eqnarray}

317: \partial_t \varphi + \partial_{\alpha} (\varphi v_{\alpha}) &=&

318: \Theta \nabla^2 \Delta \mu + \nonumber \\

319: &&-(\tau_{\varphi} -\frac{1}{2}) \Delta t

320: \partial_{\alpha} \Big(

321: \frac{\varphi}{n} \partial_{\alpha} p

322: +\frac{\varphi}{n} \partial_{\beta} c^2 P_{\alpha\beta}^{chem}

323: \Big) + o(\Delta t^2)

324: \label{conv0}

325: \end{eqnarray}

326: where the macroscopic mobility is $\Theta=\Gamma (\tau_{\varphi} -1/2)

327: c^2 \Delta t$.

328: As before, the spurious terms in the second line of Eq.~(\ref{conv0}) 

329: are also expected to be small.

330: 

331: If the density fluctuations are negligible, the continuity and 

332: Navier-Stokes equations (\ref{cont0})-(\ref{ns0})  

333: describe the behavior of an incompressible fluid. 

334: However, this is not always the case in LBM where

335: the compressibility effects can produce errors in numerical 

336: simulations.

337: In the next two subsections we will modify the described LBM in order to

338: reduce unphysical compressibility effects.  

339: 

340: \subsection{The limit of small Mach number}

341: 

342: The first scheme that we consider and adapt to our LBM is the one introduced 

343: in Ref.~\cite{heluo}.

344: Previous studies for incompressible fluids have shown that 

345: the density fluctuations, $\delta n$, are expected to be of order $o(M^2)$ in the limit

346: $M \rightarrow 0$, $M=v/c_s$ being the Mach number \cite{zou95,martinez}.

347: We enforce this condition in the method.

348: By substituting $n=n_0+\delta n$ into the equilibrium distribution functions (\ref{svil1}), 

349: $n_0$ being the specified constant value of density,

350: and neglecting the terms proportional to $\delta n ({\bf v}/c)$,

351: and $\delta n ({\bf v}/c)^2$, which are of the order $o(M^3)$ or higher, 

352: we find that the 

353: net effect is to replace $n$ with $n_0$ into

354: the expansion coefficients (\ref{bs})-(\ref{ds}) keeping unchanged the rest

355: of the scheme.

356: In this way the terms in the equilibrium distribution functions (\ref{svil1}) 

357: are of the order $o(M^2)$ or lower. The condition

358: $M \ll 1$ is satisfied in our runs where it is always $M < 0.01$.

359: 

360: In this case the Chapman-Enskog expansion gives 

361: the continuity equation

362: \begin{equation}

363: \partial_{\alpha} v_{\alpha} = 0 + o(\Delta t^2) + o(M^2),

364: \label{cont1}

365: \end{equation}

366: the incompressible Navier-Stokes equation

367: \begin{equation}

368: \partial_t v_{\beta} + v_{\alpha} \partial_{\alpha} v_{\beta} =

369: -\frac{1}{n_0} \partial_{\beta} p  

370: -\frac{1}{n_0} \partial_{\alpha} c^2 P_{\alpha\beta}^{chem} +

371: \nu \nabla^2 v_{\beta} + o(\Delta t^2)+o(M^3),

372: \label{ns1}

373: \end{equation}

374: and the convection-diffusion equation

375: \begin{equation}

376: \partial_t \varphi + \partial_{\alpha} (\varphi v_{\alpha}) =

377: \Theta \nabla^2 \Delta \mu + o(\Delta t^2)+ o(M^3),

378: \label{conv1}

379: \end{equation}

380: where we have not written again the spurious terms of Eqs.~(\ref{ns0})-(\ref{conv0}).

381: 

382: \subsection{A feedback mechanism}

383: 

384: In Ref.~\cite{fang} a different algorithm based on a feedback 

385: mechanism is proposed to keep the density nearly 

386: constant in LBM for a single fluid. 

387: Here we adapt this scheme to our model combining it with 

388: the one previously described.

389: We use for the expansion

390: coefficients (\ref{as}) the following form:

391: \begin{eqnarray}

392: A_0 &=& a_0 n - \frac{5}{6} P_{\alpha\beta}^{chem} \delta_{\alpha\beta} 

393: \nonumber \\

394: A_I &=& a_I n + \frac{1}{6} P_{\alpha\beta}^{chem} \delta_{\alpha\beta}

395: \label{newa0} \\

396: A_{II} &=& \frac{A_I}{4} \nonumber

397: \end{eqnarray}

398: Let us note that these expressions coincide with (\ref{as}) when $a_0=4/9$ 

399: and 

400: $a_I=1/9$. Here we only ask that $a_0+5 a_I =1$ to guarantee the total density 

401: conservation. 

402: We consider at each node and at each time step before collision 

403: the density $n$ fixing the value

404: of $a_I$ as

405: \begin{equation}

406: a_I=a_I^0 - b (1-\frac{n}{n_0})

407: \label{ai}

408: \end{equation}

409: where $a_I^0=1/9$ and $b$ is a positive constant given by experience.

410: In this paper we generally use $b=0.01$. The value of $b$ is flexible, 

411: but if it is too large the simulation will be unstable.

412: The value of $a_0$ follows from the relation between $a_0$ and $a_I$.

413: These values are then used in (\ref{newa0}).

414: This choice forces, in the next streaming step, more (less) particles

415: to leave the node with $n > n_0$ ($n < n_0$), resulting in 

416: $n \rightarrow n_0$.

417: 

418: By using a Chapman-Enskog expansion and neglecting terms of order 

419: $o(M^3)$ or higher we find 

420: the continuity equation (\ref{cont1}),

421: the incompressible Navier-Stokes equation (\ref{ns1})

422: where now the isotropic pressure contribution is 

423: \begin{equation}

424: p^{'}=3 a_I c^2 n = c_s^2 n - 3 c^2 b (n_0 - n) + o(\delta n^2) 

425: \label{presscal2}

426: \end{equation}

427: and the convection-diffusion equation (\ref{conv1}).

428: For the convenience of description, we label the old scheme as A, the scheme

429: described in subsection B as B, the scheme combining scheme B and the feedback

430: mechanism as C. 

431: 

432: \section{Phase separation with periodic boundary conditions}

433: 

434: After a binary mixture is quenched into the coexistence 

435: region and  domains of the two phases are well established,

436: the typical size of domains generally grows as

437: $R(t) \sim t^{\alpha}$. 

438: A simple scaling analysis of the Navier-Stokes and 

439: of the convection-diffusion 

440: equations shows that three regimes can be found

441: depending on  the role played by  hydrodynamic degrees of freedom.

442: At high viscosity the domain growth is governed 

443: by a diffusive mechanism and  

444: $\alpha = 1/3$ \cite{Lifshitz61}.

445: When  hydrodynamics becomes relevant,

446: the  laws $R(t) \sim t$ or  $R(t) \sim t^{2/3}$

447: are expected depending on whether viscous forces or inertial 

448: effects dominate, respectively \cite{Furukawa85,Bray94}. 

449: However, in real systems the situation

450: is more complex. 

451: The physical mechanism responsible for  viscous growth

452: is not operating in the two-dimensional case \cite{Siggia79}

453: and, indeed, this regime

454: has never been observed in simulations \cite{Yeomans}.

455: In this paper, since we address the question of stability

456: which becomes  more critical at low viscosities,

457: we will focus on cases corresponding to the inertial regime.

458: 

459: The average size of domains $R(t)$ can

460: be calculated as 

461: the first momentum of the structure factor, that is 

462: \begin{equation}

463: R(t) = \frac{ \int d \vec k \;\;C(\vec k,t)}{\int d \vec k \;\; |k|

464: \;\;C(\vec k,t)}

465: \end{equation}

466: where 

467: \begin{equation}

468: C(\vec k,t)= \langle \varphi(\vec k, t)\varphi(-\vec k, t)\rangle   \qquad 

469: \end{equation}

470: and $ \langle \cdot \rangle $ is the average over different histories.

471: 

472: In Fig.~1 we show the behavior of $R(t)$ for the sets of parameters

473: listed in Table I. Similar behaviors have been also observed in many other cases 

474: here not reported.

475: We have also not reported the 

476:  worst situations occurring  at very low $\tau$, when

477: the scheme A gets very soon unstable ($t \simeq 100$) and  

478: the new schemes are not significantly better.

479: All the cases of Table I refer to situations where phase separation

480: has already generated well-formed macroscopic domains

481: and a significant evolution of $R(t)$ can be observed.

482: For each set, results obtained with the schemes A,

483: B and C are compared.

484: While  the scheme A can be used only before

485: density fluctuations start to grow indefinitely,

486: the simulations with the new schemes remain stable at all times for most of

487: the cases considered.

488: We have found only few cases (e.g., set 6 at the lowest

489: considered viscosity in Table I)

490: where the schemes B and C

491: are affected by instability; however, the instability

492: with the scheme C occurs when the system is almost completely phase separated,

493: later than with the scheme B and

494: much later than with the scheme A.

495: 

496: All the cases shown correspond to the inertial regime and for

497: reference we have plotted the line with slope $2/3$. 

498: The sets 1 and 2 only differ in the value of the mobility $\Gamma$.

499: The main difference is that, before reaching the hydrodynamic $2/3$ regime,

500: the growth is faster when  the  mobility is higher

501: as it can be observed in Fig.~1. 

502: The slowing down of growth at large times is due to

503: finite size effects appearing when the size of domains 

504: becomes comparable with the lattice size.

505: 

506: In order to understand the effects of density fluctuations,

507: we show in Fig.~2  pictures of the system 

508: for the parameter set  2 at the time $t=5895$ when 

509:  the scheme A becomes unstable.

510: Simulations  obtained with the old and the new schemes can be compared

511: using  the same initial conditions.

512: Differences in the order parameter 

513: field can be observed due to the different treatment

514: of density fluctuations. When the scheme A is used

515: the density plot shows the appearing of

516: a peak which will grow indefinitely.

517: The presence of the instability can be also observed in the 

518: middle low region of the concentration field as a white stripe. 

519: In the case with the 

520: scheme B fluctuations remain  limited around the average value $n=1$

521: and are larger  in correspondence of interfaces. The velocity field also

522: shows an instability with the scheme A while with the new scheme

523: regular local flows can be observed.

524: 

525: >From sets 3 and 6 of Fig.~1 we observe that $R(t)$  behaves 

526: more regularly when calculated with the scheme C. 

527: This allows 

528: a better determination of the growth exponent.

529: The effects of

530: using a bigger lattice size can be seen

531: in Fig.~1 for cases 4 and 5.  

532: 

533: In recent simulations of three-dimensional 

534: systems based on the original algorithm the growth exponent

535: was obtained by fitting the results of $R(t)$ with the function 

536: $f(t) = c(t-d)^{\alpha}$ in a time interval from

537:  the inflexion point to the onset of the instability \cite{catesjfm}.

538: This procedure, as admitted by the same authors of Ref.~\cite{catesjfm} and

539: discussed also in Ref.~\cite{coveney},

540: is very sensitive to small variations 

541: of the initial time considered, so that the evaluation 

542: of the growth exponent is problematic.

543: We have applied a similar procedure to our results

544: confirming the sensitivity to small variations 

545: of the initial time considered and finding different values for the exponent

546: when the old or the new more stable schemes are applied.

547: For example, for the set 2 with the old method,

548: we find $\alpha = 0.58$ if we start from the inflexion point located at $t=3000$

549: and $\alpha = 0.61$ if data are fitted from $t = 4000$ (the run becomes 

550: unstable at t = 5985). If we use the longer series of 

551: data obtained with the scheme C

552: we get $\alpha = 0.64$ starting from the inflexion point or at a later time.

553: For the set 4 the fitting procedure for the scheme A

554: gives values of $\alpha $ in the range $0.50 - 0.55$ 

555: changing the initial time for the fitting procedure, 

556: while the scheme B

557: gives values in the range $\alpha = 0.64 - 0.67$.

558: Other cases are also worst because sometimes, like for the set 3, 

559: a reliable determination of the exponent is not possible with the old scheme.

560: 

561: \section{Stability with walls}

562: 

563: In this Section we discuss how the presence of boundary conditions

564: with walls affects the stability of simulations of phase separation.

565: Boundary conditions with walls are present  in problems where 

566: a flow is applied to the system, e. g. shear flow,

567: or in situations with geometrically complex boundaries.

568: 

569: Fluctuations of density are generally larger close to the walls

570: due to the fact that the propagation and collision steps involve

571: a reflection on the boundary.

572: Then, if the density is larger than the average value in a point close 

573: to a wall,

574: the reflection tends to enhance this fluctuation.

575: An example of the behavior of density just before the 

576: system becomes unstable is given in Fig.~3 which shows clearly

577: that the instability appears close to the walls.

578: 

579: Also for the case with walls we have  applied

580: the schemes described in Sec. II for controlling

581: the density fluctuations. 

582: In all the considered cases  instabilities in

583: simulations with walls occur  much before 

584: than in simulations with periodic boundary conditions.

585: In Table II we have reported the stability results for the 

586: original and the new scheme C. 

587: The comparison shows that in most of the cases

588: the new algorithm can be very useful for running longer

589: simulations of phase separation.

590: 

591: \section{Conclusion}

592: 

593: In this paper we have considered two-dimensional 

594: Lattice Boltzmann methods for incompressible

595: binary mixtures 

596: based on a  free-energy functional approach.

597: We have introduced mechanisms for controlling the total density

598: fluctuations. These mechanisms were already shown useful 

599: for reproducing more accurately examples of analytical solutions 

600: for simple fluid flows.

601: 

602: In the case of binary mixtures,  density fluctuations

603: can induce numerical instabilities that  

604: make impossible to  run long enough simulations as it is needed

605: in problems like spinodal decomposition.

606: We have used the new algorithms 

607: to study the phase separation of binary mixtures

608: in the inertial regime at low viscosities when the effects of

609: density fluctuations are worst.

610: We have found  that the stability of the simulations 

611: can be largely improved. For  most of the parameters considered,

612: differently from what happens with the original algorithm,

613: it is possible to take under control density fluctuations eliminating

614: the occurring of instabilities. This allows to study  the properties

615: of phase separating binary mixtures more reliably; in particular,

616: a more accurate evaluation of the growth exponent

617: can be given. 

618: As a future research in this direction, it would be interesting to 

619: apply  the  LBM with improved stability to

620:  the case of a three-dimensional mixture.

621: 

622: \acknowledgments

623: A.L. acknowledges INFM for partial support.

624: 

625: \begin{references}

626: 

627: \bibitem{Chen98}

628: S. Chen and G. D. Doolen, Annu. Rev. Fluid Mech. {\bf 30}, 329 (1998).

629: 

630: \bibitem{succi}

631: S. Succi, {\it The Lattice Boltzmann Equation} (Oxford University Press,

632: New York, 2001).

633: 

634: \bibitem{Yeomans}

635: J. M. Yeomans, Ann. Rev. Comp. Physics, VII (2000).

636: 

637: \bibitem{olson}

638: J. Olson and D. Rothman, J. Fluid. Mech. {\bf 341}, 343 (1997).

639: 

640: \bibitem{gonnella}

641: G. Gonnella, E. Orlandini, and J. M. Yeomans, Phys. Rev. Lett. {\bf 78}, 

642: 1695 (1997).

643: 

644: \bibitem{gompper}

645: O. Theissen, G. Gompper, and D. M. Kroll, 

646: Europhys. Lett.  {\bf 42}, 419 (1998).

647: 

648: \bibitem{lamepl}

649: A. Lamura, G. Gonnella, and J. M. Yeomans, 

650: Europhys. Lett.  {\bf 45}, 314 (1999).

651: 

652: \bibitem{Malevanets99}

653: A. Malevanets and J. M. Yeomans, Faraday Discussions {\bf 112}, 237 (1999).

654: 

655: \bibitem{yeom}

656: C. Denniston, E. Orlandini, and J. M. Yeomans, 

657: Europhys. Lett.  {\bf 52}, 481 (2000).

658: 

659: \bibitem{Orlandini}

660: E. Orlandini, M. R. Swift, and J. M.  Yeomans, 

661: Europhys. Lett. {\bf 32}, 463 (1995).

662:  

663: \bibitem{Swift}

664: M. R. Swift, E. Orlandini, W. R. Osborn, and J. M. Yeomans, 

665: Phys. Rev. E {\bf 54}, 5041 (1996).

666: 

667: \bibitem{catesjfm}

668: V. M. Kendon, J. C. Desplat, P. Bladon, and M. E. Cates,  

669: Phys. Rev. Lett. {\bf 83}, 576 (1999);

670: V. M. Kendon, M. E. Cates, J. C. Desplat, I. Pagonabarraga, and P. Bladon,  

671: J. Fluid. Mech. {\bf 440}, 147 (2001).

672: 

673: \bibitem{Furukawa85}

674: H. Furukawa, Adv. Phys. {\bf 34}, 703 (1985).

675: 

676: \bibitem{Bray94}

677: A. J. Bray, Adv. Phys. {\bf 43}, 357 (1994).

678: 

679: \bibitem{zou95}

680: Q. Zou, S. Hou, S. Chen, and G. D. Doolen, J. Stat. Phys. {\bf 81}, 35 (1995).

681: 

682: \bibitem{dembo}

683: H. He, L. S. Luo and M. Dembo, Physica A {\bf 239}, 276 (1997).

684: 

685: \bibitem{heluo}

686: H. He and L. S. Luo, J. Stat. Phys. {\bf 88}, 927 (1997).

687: 

688: \bibitem{fang} H. Fang, R. Wan, and Z. Lin, Phys. Rev. E {\bf 66}, 036314 (2002).

689: 

690: \bibitem{bhatnagar}

691: P. Bhatnagar, E. P. Gross, and M. K. Krook, Phys. Rev. {\bf 94}, 511 (1954).

692: 

693: \bibitem{chen1}

694: H. Chen, S. Chen, and W. Matthaeus, Phys. Rev. A {\bf 45}, R5339 (1992).

695: 

696: \bibitem{aiguo}

697: Aiguo Xu, G. Gonnella, and A. Lamura, Phys. Rev. E  {\bf 67}, 056105 (2003).

698: \bibitem{chapman}

699: S. Chapman and T. Cowling, {\it The Mathematical Theory of

700: Non-uniform Gases} (Cambridge University Press, Cambridge, 1970).

701: 

702: \bibitem{martinez} D. O. Martinez, W. H. Matthaeus, S. Chen, and D. C.

703: Montgomery, Phys. Fluids {\bf 6}, 1285 (1994).

704: 

705: \bibitem{Lifshitz61}

706: I. M. Lifshitz and V. V. Slyozov, J. Phys. Chem. Solids {\bf 19}, 35 (1961). 

707: 

708: \bibitem{Siggia79}

709: E. D. Siggia, Phys. Rev. A {\bf 20}, 595 (1979).

710: 

711: \bibitem{coveney} 

712: N. Gonzalez-Segredo, M. Nekovee, and P. V. Coveney,

713: preprint, cond-mat/0301046, in press in Phys. Rev. E.

714: 

715: \end{references}

716: 

717: \newpage

718: 

719: \begin{table}

720: \begin{center}

721: \centerline{\epsfig{file=table1.eps,bbllx=184pt,bblly=599pt,bburx=447pt,bbury=710pt,

722: width=0.8\textwidth,clip=}}

723: \end{center}

724: \caption{The sets of parameter used in Figure 1.}

725: \label{table1}

726: \end{table}

727: 

728: \newpage

729: 

730: \begin{table}

731: \begin{center}

732: \centerline{\epsfig{file=table2.eps,bbllx=139pt,bblly=568pt,bburx=494pt,bbury=711pt,

733: width=0.8\textwidth,clip=}}

734: \end{center}

735: \caption{Safe running times for the schemes A and C for 8 

736: sets of parameters. $t_{max}$, where the upper label refers to the scheme used, is the longest simulated time.

737: The symbol $=$ means that the simulation becomes unstable after that time; the symbol $>$ means

738: that the simulation is still stable up to that time.}

739: \label{table2}

740: \end{table}

741: 

742: \newpage

743: 

744: \begin{figure}

745: \begin{center}

746: \centerline{\epsfig{file=Fig1.eps,bbllx=70pt,bblly=192pt,bburx=527pt,bbury=754pt,

747: width=0.8\textwidth,clip=}}

748: \caption{$R(t)$ behavior for the sets of parameters listed in Table I. The solid, dashed, and dotted lines

749: correspond to scheme A, B, and C, respectively.

750: Runs with the schemes A and B always start from random initial configurations. The scheme C is applied either

751: to simulations from random initial conditions (sets 5 and 6) or starting from a configuration slightly before 

752: the scheme A fails to work. Runs with the scheme C were eventually stopped, though still stable, once the system

753: was phase separated.}

754: \end{center}

755: \label{fig_1}

756: \end{figure}

757: 

758: \newpage

759: 

760: %\begin{figure}

761: \begin{center}

762: Fig.2 and Fig3 are in JPG format. Captions of Fig.2 and Fig. 3 are as follows,

763: \end{center}

764: %\centerline{\epsfig{file=Fig2.eps,bbllx=106pt,bblly=155pt,bburx=410pt,bbury=710pt,

765: %width=0.6\textwidth,clip=}}

766: %\caption{

767: 

768: Fig. 2 Configurations of the order parameter $\varphi$ (a)-(b), the total density $n$ (c)-(d), 

769: and velocity fields (e)-(f) for the parameter set 2 at the time

770: $t=5895$. Figures (a), (c), (e) are for the scheme A, while (b), (d), (f) are for the scheme B. 

771: In picture (a) it is $-2.13 \le \varphi \le 1.77$ and in (b) it is $-1.06 \le \varphi \le 1.05$.

772: %}

773: 

774: %\end{center}

775: %\label{fig_2}

776: %\end{figure}

777: 

778: %\newpage

779: 

780: %\begin{figure}

781: %\begin{center}

782: %\centerline{\epsfig{file=Fig3.eps,bbllx=109pt,bblly=375pt,bburx=493pt,bbury=819pt,

783: %width=0.6\textwidth,clip=}}

784: %\caption{

785: 

786: Fig.3 Three-dimensional plot of the total density $n$ in a case with walls 

787: at time $t=1315$ using the scheme B.

788: The parameters are those of set 4 in Table I.

789: %}

790: %\end{center}

791: %\label{fig_3}

792: %\end{figure}

793: 

794: 

795: \end{document}

796: 

797: 

798: 

799:             

800: