New hydrodynamic mechanism for drop coarsening ∗ Vadim S. Nikolayev, Daniel Beysens, and Patrick Guenoun Service de Physique de l’Etat Condens´e, CE Saclay, F-91191 Gif-sur-Yvette Cedex, France (Dated: 9 March 1995) Wediscussanewmechanismofdropcoarseningduetocoalescenceonly,whichdescribesthelate stages of phase separation in fluids. Depending on the volume fraction of the minority phase, we identifytwodifferentregimesofgrowth,wherethedropsareinterconnectedandtheircharacteristic sizegrowslinearlywithtimeandwherethesphericaldropsaredisconnectedandthegrowthfollows (time)1/3. The transition between the two regimes is sharp and occurs at a well defined volume 6 fraction of order 30%. 1 0 2 InthisLetterweconcentrateonthekineticsofthelate coefficient of the drops of the same viscosity n stagesofthephaseseparation. Thissubjecthasreceived a considerable attention recently [1]–[4]. Most of the ex- D = kBT . (2) J 5πηR periments on growth kinetics have been performed near 6 thecriticalpointofbinaryliquidmixtures(orsimpleflu- 2 The factor f represents the correction which takes into ids) because there the critical slowing down allows the accountthehydrodynamicinteractionbetweenthedrops. ] phenomenon to be observed during a reasonable time. n It depends on the ratio of the viscosities of the liquid After a thermalquenchfromthe one-phaseregionto the y inside and outside the drops and the average distance d two-phase region of the phase diagram (Fig. 1) the do- betweenthedropsand,therefore,onthe volumefraction - mains of the new phases nucleate andgrow. It turns out φ. This correctionhasbeencalculatedinthedilute limit u that two alternative regimes of coarsening are possible. l (φ 0)byZhangandDavis[10]. Intheproximityofthe f The first can be observed when the volume fraction φ → . criticalpoint we can assume the viscosities of the phases s of the minority phase is lower than some threshold [5] c to be equal and and the domains of the characteristic size R grow ac- i s cording to the law R t1/3 (t is the time elapsed after f(0)=0.56. (3) y ∝ h the quench) as sphericaldrops. The secondregime man- p ifestsitselfwhenthequenchisperformedathighvolume According to the model, the drops coalesce immediately [ fraction, the coarsening law is R t1 and the domains after the collision. Coalescenceis the only reasonfor the ∝ grow as a complicated interconnected structure. Recent decrease of the total number of the drops with the rate 1 v experiments [2] show that when 0.1 < φ < 0.3 the t1/3 dn 4 growth can be explained by a mechanism of Brownian = N . (4) B 0 drop motion and coalescence rather than the Lifshitz- dt − 0 Slyozov mechanism [6] which holds for φ 0 and which With the relation 7 → we will not discuss here. We are interested in the late 0 4 . stagesofgrowthwhenphase boundariesarealreadywell φ= πR3n=const, (5) 1 3 developed and the concentrations of the phases are very 0 6 closetotheequilibriumvaluesatgiventemperatureT as Eq.(4) yields a R t1/3 law. The further improvements 1 definedbythecoexistencecurve(Fig.1). Thenthedrops (see[11]andrefs. ∝therein)ofthismodelinfluencemainly v: growjustbecausethesystemtendstominimizethetotal thenumericalfactorin(1)whichisnotimportantforthe i surface separating the phases (i.e. due to coalescence) present considerations. X and φ no longer depends on time. Hydrodynamic approaches. The originof the R ar Brownian coalescence. The Brownian mechanism t1 growth law, observed at high volume fraction wher∝e was considered first by Smoluchowski [7] for coagulation domainsareinterconnected,ismuchlessclear. Bymeans of colloids and was then applied to phase separation by ofadimensionalanalysisSiggia[9]hasshownthathydro- Binder and Stauffer [8] and Siggia [9]. According to this dynamics is needed to explain the kinetics. It assumed mechanism, the rate of collisions per unit volume due to the growth to be ruled by the Taylor instability of the the Brownian motion of spherical drops in the liquid of long tube of fluid, which breaks into separate drops and shear viscosity η is associatedthe growthrate with the rate ofthe evolution of the unstable fluctuations. This idea has been devel- oped by San Miguel et al. in [12]. However, it was not N =16πDRn2f(φ), (1) B clear how this process was related to the growth. Another approach has been considered by Kawasaki where n is an average number of drops per volume, R and Ohta [3] who used a model of coupled equations of is the average radius of the drops and D is the diffusion hydrodynamics and diffusion. It was assumed that the 2 X' 0 --1.5 -1 -0.5 ψ0 0.5 1 1.5 2 ) (K cee Initial position T-T c-0.02 CoexistencurvDits1c/o3n- Intte1r- Dits1c/o3n- -0.9063 d0 d0 nected connected nected -0.04 5 6 5 0 R 0.1 0.2 0.3 .18 R 7 d 5 0.66 c+ 0.7 0.72 c- 0.76 0 ψ d Concentration 0 1 FIG. 1: Coexistence curve for a model two-phase system .28 R 1 (density-matched cyclohexane–methanol) from Ref. [1]. The dotted curve is the calculated boundary (see text) between thet1/3 andt1 growthregionswhichcorrespondstoφ=0.26. d0 The triangles are the experimental data from Ref. 1. The 2.3 "Coalescence" curves corresponding to φ=0.15 (random percolation limit) 75 position andφ=0.35 (thevaluewhichgivesthebestfittotheexper- X imental data from Ref. 1) are presented also for comparison. Thevolumefraction oftheminorityphaseatthepoint(c,T) FIG. 2: The positions of the drop surfaces at the beginning can be calculated as φ = 1/2−|c−c |/(c− −c+) where c (dotted line) and at the end (solid line) of the simulation. ψ c c (=0.707 for thiscase) is thecritical concentration. is the coalescence distance, d0 is the initial distance between ′ thedrops, X –X is theaxis of cylindrical symmetry. growth is controlled by diffusion and the hydrodynamic correction to this process was calculated. However, the Browniandiffusion but rather flows induced by previous translational movement of the drops due to the pressure coalescence. We use the conceptof “coalescence-induced gradient was not taken into account. The motion of the coalescence” as introduced by Tanaka [4] who, however, liquid was supposed to be induced by the concentration thought that induction by the hydrodynamic flow was gradient only. At the same time, it is well known that not relevant, having stated that coalescence takes place the concentration variation does not enter the equations after the decay of the flow. We consider here a coales- ofhydrodynamicsofthe liquidmixture inafirstapprox- cence process between two drops and study numerically imation[13]. Moreover,it is evidentthatathighvolume thegeneratedflowanditsinfluenceonthethirdneighbor- fraction the coalescence process induced by the transla- ingdrop. Thefullydeterministichydrodynamicproblem tional motion of the drops becomes very important. within the creeping flow approximation (which is well Recently, several groups [14–18] performed large scale justified near the critical point [9]) was solved. The free direct numerical simulations by using different ap- boundary conditions were applied on all drop interfaces proaches to solve coupled equations of diffusion and hy- whose motion is driven by surface tension. At each time drodynamics. Some recovered the linear growth law stepthevelocityofeachmeshpointontheinterfacecon- [15, 17], whilst the others were notable to reachthe late tours was computed using a boundary integralapproach stages of separation and measured the transient values [19]. When the new positions ofthe interfaces havebeen of the growth exponent (between 1/3 and 1). In spite calculated, the procedure was continued iteratively. The of these efforts, the physical mechanism for the linear details of the solution will be presented elsewhere [20]. growth has not been clarified. To our knowledge, the Webeginthesimulationwhencoalescencestartsbetween simulationsnevershowedtwoasymptoticallawsdepend- two drops of size R (Fig. 2), i.e. when the drops of the ing on the volume fraction: the exponent is either larger minority phase approach to within a distance of coales- than 1/3 when accounting for hydrodynamics or 1/3 for cence ψ which corresponds to the interface thickness of pure diffusion. Thus the precise threshold in φ separat- the drop [9]. We choose for the simulation R = 10ψ. ingthe t1 andt1/3 regimesisnotpredictedeitherbyany Then we place another drop at the distance d from the 0 theory or by simulation. composite drop (defined as the aggregateof two coalesc- Simulation of coalescence. We show here that a ing drops) and envelope these two drops by a spherical t1 growth can originate from a coalescence mechanism shell to mimic the surrounding pattern of tightly packed whose limiting process between two coalescences is not drops. Thusthedistancebetweenthedropsandtheshell 3 has been chosen to be d also. The surface tension σ is always repulsive due to the lubrication force. Thus they 0 supposed to be the same for all the interfaces. Unfor- tend to be as far from each other as possible. More- tunately, due to the prohibitively long computing times, over, experiment shows a liquid-like order for the drop we could not simulate the process of coalescence of two positions. Such a correlation explains why the drops do spherical drops. Instead, we had to use a configuration not percolate [1] when the volume fractionφ reaches the with cylindrical symmetry with respect to the axis X′ random percolation limit (φ 0.15). ≈ – X (see Fig. 2) which is expected to retain the main Sincenoquantitativeinformationisavailabletodeter- features associated with the spherical shape. In the be- mineb,wecalculateitsupperandlowerbounds. Ideally, ginning of the simulation the composite droplooks more we can assume that the drops are arranged into a reg- like a torus. The spherical shell approximation can be ular lattice, with the vertices as far from each other as justified by the fact that the main effect of the assembly possible. This is the face centered cubic lattice where of surrounding drops (as well as of the sphericalshell) is b=π/3√2 0.74andwhichcorrespondstothe fullyor- ≈ to confine the motion of the neighboring drops – see [20] dered structure. We can also consider as a lower bound foranadvanceddiscussion. Asetupwith onlytwodrops the random close packing arrangementfor spheres of ra- without either a shell or surrounding drops would not dius R + d /2. This corresponds to the absence of a 0 enable a new coalescence even for the smallest interdrop short-range order [22] and implies b 0.64. We note ≈ distances. Though we cannot control quantitatively this that the value of b is not very sensitive to the particular approximation,it is the simplest one which captures the spacearrangement. Inthefollowingweadoptthemedian main features of the process. value b=0.69. Afirstimportantresultfromthesimulationisthatthe Generalization of the hydrodynamic model. flow generated by the first coalescence is able to gener- Now we can generalize the above hydrodynamic mech- atea secondcoalescencebetweenthecompositedropand anism for an arbitrary shape of the drops. The self- the neighboring drop (Fig. 2). This means that the lu- similarity of the growth implies the following relation brication interaction with the surrounding drops (with for the characteristicsizes of the drops between i-th and the shell in our model) make attract the composite and (i+1)-thcoalescences: R(i+1) =βR(i), where β is a uni- the neighboring drops (note that the second coalescence versal shape factor, which depends on φ only. We can does not take place between this neighboring drop and rewrite also the Eq. (6) for the time between the coales- the shell). This leads to the formation of a new elon- cencesinthe formt(i) =α(φ)ηR(i)/σ, whereα(φ) isalso c gated droplet. When the drops are close enough to each the universal function. The last expression conforms to other the composite drop will have no time to relax to a the scaling assumption which implies the independence spherical shape since a new coalescence takes place be- of t(i) on the secondlength scale ψ. Then after n coales- c fore relaxation. An interconnected pattern naturally fol- cences lows. Incontrast,whenthedropsarefarfromeachother (d /R 1),thesecondcoalescencewillnevertakeplace. n−1 0 The dr≫oplets take a spherical shape and the liquid mo- R=βnR(0), t= Xtc(i) tion stops. It is also clear that if d /R < l , which we i=0 0 G call “geometric coalescence limit”, coalescence necessary and occursduetogeometricalconstraint. Wefindl 0.484 G ≈ [20], [21]. β 1 σ R=R(0)+ − t, (8) The second important result is that the coalescence α · η takes place also for l < d /R < l where l is a value G 0 H H which we call “hydrodynamic coalescence limit”. It is where R(0) is the initial size of the drop. Noting that defined as a reduced initial distance where the time be- β = 21/3 for the spheres and β > 1 for the long tubes, tweentwocoalescences(tc)becomesinfinite. Sincethere we can take β 1.1 for the estim∼ate. Since α 10 for ∼ ∼ is only one length scale (R) in the problem, tc can be φ = 0.5, as it follows from Fig. 3 and Eq.(7), we obtain written in the scaled form R 0.01σ/ηt,whichcompareswellwiththe experiment ∼ [23], which gives 0.03 for the numerical factor. t =αηR/σ, (6) c Competition between two mechanisms. Using where α is a reduced coalescence time which depends on now Eq. (7) we can relate l to a volume fraction φ . H H d /R only (Fig. 3). It is clear that the described hydrodynamic mechanism 0 The quantity d /R is relatedto the volume fractionof worksonlywhenφ>φ whiletheBrowniancoalescence 0 H the drops (minority phase) φ: takes place in the whole range of φ. Below we shall con- φ=b[1+d /(2R)]−3. (7) sider the regime for whichφ>φH inorderto obtainthe 0 position of the boundary between t1/3 and t1 regions on The constantb depends onthe spacearrangementofthe the phase diagram. drops. The hydrodynamic interaction between them is Taking into account the competition between the two 4 60 Now we aim to estimate the function G(φ) by using for α(φ) the calculated function in Fig. 3 along with Eq.(7). 50 It turns out that the function G(φ) well fits the power law 40 G(φ) (φ φ )−δ (12) H ∝ − forφ>φ whereφ 0.26,δ 0.33andthedivergency H H α 30 comes from α(φ). Fr≈om (12),≈it is easy to deduce that (11) is valid when 0 < φ φ < 10−6. In practice, this H 20 meansthatforallφ>φ −,theh∼ydrodynamicmechanism H only will determine the growth from the very beginning 10 of the drop coarsening. Alternatively, for φ < φH, the l l drops will grow according to the Brownian mechanism g H only. This explains the sharp transition in the kinetics 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 (t1 t1/3) which is controlled by the volume fraction → d /R of the minority phase as observed in [1] and [2]. The 0 curve which corresponds to the threshold value φ=0.26 is plotted in Fig. 1. It shows a reasonable agreement FIG. 3: The simulation data on the dependence of the re- duced coalescence time α on d0/R. The vertical lines show withtheexperimentaldatainspiteofourveryparticular thegeometric and hydrodynamiccoalescence limits. choice of the form and arrangementof the drops. It should be mentioned that our model can be applied toanysystemwherethegrowthisduetothecoalescence mechanisms, we consider the relation ofliquiddropsinsideanotherfluid(phaseseparation,co- dn agulation, etc.). = (N +N ), (9) dt − B H Oneoftheauthors(V.N.)wouldliketothankthecol- laboratorsof SPEC Saclay for their kind hospitality and instead of (4), where N is the rate of the coalescences H Minist`eredel’EnseignementSup´erieuretdelaRecherche due to the hydrodynamics which can be calculated by of France for the financial support. using Eq.(8) and the relationbetween n and R (Eq.(5)). The latter, however, depends on the shape of the drop. Weassumethatintheearlystagesthedropsarespherical and we use Eq.(5). It should be noted that the Eq.(9) rewritten for the scaled wavenumber exactly coincides ∗ On leave from: Bogolyubov Institute for Theoreti- withthe semi-empiricalequationsuggestedby Furukawa cal Physics, National Ukrainian Academy of Sciences, [24]. 252143, Kiev,Ukraine; e-mail: [email protected] The BrownianmechanismdominateswhenN >N . [1] Y. Jayalakshmi, B. Khalil and D. Beysens, Phys. Rev. B H Lett. 69, 3088 (1992). Inthevicinityofthecriticalpointonecanusetwo-scale- [2] F. Perrot, P. Guenoun, T. Baumberger, D. Beysens, factor universality [25] expression σ = k T/γξ2 where B Y. Garrabos and B. Le Neindre, Phys. Rev. Lett. 73, ξ is the correlation length in the two-phase region, γ 688 (1994). ≈ 0.39 is a universal constant. Then we reduce the last [3] K. Kawasaki, T. Ohta, Physica A, 118, 175 (1983); inequality to T. Koga, K. Kawasaki, M. Takenaka and T. Hashimoto, Physica A,198, 473 (1993). R2/ξ2 <G(φ),(10) [4] H. Tanaka, Phys. Rev.Lett. 72, 1702 (1994). γ [5] J. D. Gunton, M. San Miguel and P. S. Sahni, in:Phase where G(φ)=Cφf(φ)α(φ) and C = . 5π(β 1) Transitions and Critical Phenomena, Volume 8, edited − by C. Domb and J. L. Lebowitz (New York: Academic We do not know much about the function f(φ) which Press, 1983), p.267. has been discussed in Eq.(1). However, it is unphys- [6] I. M. Lifshitz and V. V. Slyozov, J. Phys.Chem. Solids, ical to assume that it exhibits singularities or a steep 19, 35 (1961). behavior. We shall use the constant value (3) in the fol- [7] M. von Smoluchowski, Z. Phys. Chem. 92, 129 (1917). [8] K. Binder and D. Stauffer, Phys. Rev. Lett. 33, 1006 lowing. We recall that all our considerations are valid (1974). only when the drop interfaces have already formed (late [9] E. D.Siggia, Phys.Rev.A 20, 595 (1979). stages of growth), which means that the initial radius of [10] X. Zhang, R.H. Davis, J. Fluid Mech. 230, 479 (1991). a drop cannot be less than the interface thickness, i. e. [11] H. Hayakawa, Physica A 175, 383 (1991). 4ξ [26]. Itfollowsreadilyfromtheinequality(10)that [12] M. San Miguel, M. Grant and J. D. Gunton,Phys. Rev. g≈rowth would obey the law R t1/3 when A 31, 1001 (1985). ∝ [13] L.D.LandauandE.M.Lifshitz,FluidMechanics(Perg- G(φ)>16. (11) amon, London, 1959). 5 [14] S.PuriandB.Du¨nweg,Phys.Rev.A45, R6977(1992). finedaccording toFig. 2as afunction of R,ψ andd0.If [15] F.J.Alexander,S.Chen,andD.W.Grunau,Phys.Rev. thereisnocoalescence,thecompositedropeventuallybe- B 48, 634 (1993). comesaspherewithadiameterD whichcanbededuced c [16] O. T. Valls and J. E. Farrell. Phys. Rev. E 47, R36 out of its initial volume, because the latter is conserved (1993). during the evolution. The l value is determined by the G [17] A.ShinozakiandY.Oono,Phys.Rev.E48,2622(1993). conditionD =D +2R+3ψwhichmeansthattwodrops s c [18] C. Appert, J. F. Olson, D. H. Rothman and S. Za- of the final spherical shape must fit inside the spherical leski, Spinodal decomposition in a three-dimensional, shell with thegaps of thewidth ψ. two-phase, hydrodynamic lattice gas, Preprint (1995). [22] J. H.Berryman, Phys. Rev.A 27, 1053 (1983). [19] C. Pozrikidis, Boundary integral and singularity method [23] P. Guenoun, R. Gastaud, F. Perrot, D. Beysens, Phys. forlinearizedviscousflow(CambridgeUniv.,Cambridge, Rev. A 36, 4876 (1987). 1991). [24] H. Furukawa,Adv.Phys.34, 703 (1985). [20] V. S. Nikolayev and D. Beysens, ”Coalescence-induced [25] M. R.Moldover, Phys.Rev.A 31, 1022 (1985). coalescence”byhydrodynamicflow,submittedtoJ.Fluid [26] D.Beysens,M.Robert,J.Chem.Phys.87,3056(1987), Mech. (1996). ibid. 93, 6911 (1990). [21] The shell diameter D at the initial moment can be de- s