See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/223002320 Rupturing of bitumen-in-water emulsions: Experimental evidence for viscous sintering phenomena ARTICLE in COLLOIDS AND SURFACES A PHYSICOCHEMICAL AND ENGINEERING ASPECTS · JANUARY 2001 Impact Factor: 2.75 · DOI: 10.1016/S0927-7757(00)00699-3 CITATIONS 8 READS 12 7 AUTHORS, INCLUDING: John Philip Indira Gandhi Centre for Atomic Resea… 179 PUBLICATIONS 3,176 CITATIONS SEE PROFILE Jordan T Petkov Unilever 64 PUBLICATIONS 925 CITATIONS SEE PROFILE Available from: John Philip Retrieved on: 03 February 2016
A: Physicochemical and Engineering Aspects 176 (2001) 185–194
Rupturing of bitumen-in-water emulsions: experimentalevidence for viscous sintering phenomena
L. Bonakdar a, J. Philip a, P. Bardusco a, J. Petkov a, J.J. Potti b, P. Meleard a,F. Leal-Calderon a,*
a Centre de Recherche Paul Pascal, Centre National de la Recherche Scientifique, A6enue Albert Schweitzer, 33600, Pessac, Franceb PROBISA, Pol. Ind. ‘Las Arenas’, C/Ronda 9, 28320 Pinto, Spain
Emulsions are prepared by shearing two immis-cible fluids in the presence of a surfactant. Theexcess energy associated with large interfacial areacreated in droplet formation makes the emulsionsmetastable. Emulsions are widely used in a varietyof applications because of their ability to trans-port or solubilize hydrophobic substances in awater continuous phase (e.g.: painting, paper
coating, road surfacing, lubrication, etc.). Emul-sion technology drastically simplifies the pourabil-ity of many hydrophobic materials, which arealmost like solids at room temperature. For exam-ple, the bitumen used for road surfacing is verydifficult to manipulate at room temperature dueto its high viscosity. It has to be heated to highertemperature to lower the viscosity in order topour it easily during application. Instead, a bitu-men-in-water emulsion is quite easy to handle dueto its low viscosity at room temperature. Themain requirements of such kind of emulsions arevery good stability during storage and rapidbreaking upon application on the road.
L. Bonakdar et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 176 (2001) 185–194186
Breaking or coarsening of an emulsion is anirreversible process that has been a topic of in-tense research for many years. The irreversibledestabilization of emulsions can be due to eithercoalescence [1] or Ostwald ripening [2]. When thetwo phases are poorly miscible as it is the casewith bitumen and water, coarsening is only due tocoalescence phenomena. Coalescence consists inthe rupture of the thin liquid film in between twoadjacent droplets through the nucleation of asmall channel. This first nucleation step is fol-lowed by a shape relaxation driven by surfacetension which causes the two droplets to fuse intoa unique one. The characteristic time for shaperelaxation is governed by the competition betweensurface tension and viscous dissipation and isgiven by: Tr8hR/g, where h is the viscosity of thedroplets, R is their characteristic radius and g istheir surface tension [3]. The total destruction ofan emulsion involves a very large number ofcoalescence events, each of them followed byshape relaxation. The spatial and temporal distri-bution of coalescence events may strongly varyfrom one system to the other. However, twolimiting destruction scenarios may be identified.The first one occurs when the time separating twocoalescence events (channel nucleation) is largecompared to the relaxation time Tr (low dropletviscosity). Then a gradual increase of the dropletsize is observed as a result of successive coales-cence events followed by rapid shape relaxation.In this paper, we shall focus on the other limitingmechanism which occurs when the droplet viscos-ity is extremely high. In this case, Tr may becomelarge enough compared to the time separating twocoalescence events. In such conditions, a gel isinitially formed made out of irreversibly con-nected droplets. Then the gel continuously con-tracts in order to reduce its surface area until thetotal separation of dispersed and continuousphases.
The contraction phenomenon in our case isreminiscent of the sintering process observed inceramics and aerogels [4,5]. It has been known forquite long time that powders made of fine packedparticles can be sintered at temperatures well be-low the melting point of the same macroscopicsolid phase. During sintering the materials be-
comes tougher or denser by reducing the totalsurface area. In the case of gels and glasses, thesintering occurs due to viscous flow of matter,whereas in ceramics and metals it could be due toevaporation, surface diffusion and viscous flow.This phenomenon has been widely used for theproduction of dense ceramics materials from pow-ders, the manufacturing of refractory materials atlow temperatures, the coating of electronic com-ponents and optics, etc. In this paper, we presentexperimental evidence for viscous sintering phe-nomena in a gel formed by bitumen emulsiondroplets. When a destabilizing agent is added tothe initially stable emulsion, it forms a gel, whichcontracts with time by preserving the geometry ofthe container. Therefore, the observed contractionphenomena mimic the classical sintering phenom-ena. This observation demonstrates that the vis-cous sintering can occur in organic materialshaving much lower viscosity than inorganic ones(six to eight orders of magnitude smaller thanceramics and glasses). We study the shape relax-ation of two coalesced droplets using a mi-cropipette technique as well as the kinetics ofcontraction of the macroscopic gel. The lineardependence of the contraction rate on the viscos-ity of the bitumen is in agreement with the viscoussintering theory.
2. Experimental section and discussion
2.1. Materials
We used two NYNAS bitumens with penetra-tion grades 180/220 and 80/100 (the penetrationgrade is an indication of the fluidity obtainedfrom the sinking of a needle in asphalt in normal-ized conditions). The low shear viscosity of bitu-mens was obtained at different temperaturesranging from 20°C to 90°C using two differentmethods. At lower temperatures we measured thesedimentation velocity of a small iron sphere (ra-dius=0.5 cm) immersed in the asphalt and weused Stokes law to deduce the viscosity. A con-trolled stress rheometer (Carrimed, CSL 100)equipped with the cone and plate geometry waspreferred at higher temperatures. The values ob-
L. Bonakdar et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 176 (2001) 185–194 187
Fig. 1. Temperature dependent viscosity of two different bitumens (80/100 and 180/220) from NYNAS company.
tained by these techniques were consistent to eachother. The temperature dependent viscosities ofthe bitumens are shown in Fig. 1. Tetradecyltrimethyl ammonium bromide (TTAB, critical mi-cellar concentration (CMC)=3.5×10−3 M) pur-chased from Aldrich was used as surfactant.
2.2. Study of the shape relaxation of twocoalesced droplets
The aim of this experiment is to follow theshape relaxation process between two highly vis-cous bitumen droplets. For that purpose, a mi-cropipette technique, previously applied forstudying vesicle–vesicle interactions [6] was em-ployed (see Fig. 2). A crude emulsion is producedby shearing for a few seconds a hot mixture ofwater and asphalt (95°C) with a turbulent ultra-turrax mixer. The obtained emulsions have verywide droplet size distribution ranging from 0.1 to100 mm. The emulsion is then diluted with waterand the TTAB concentration is set to 1 CMC.The emulsion is introduced in a cell that is prop-erly positioned in the observing field of a phasecontrast optical microscope (Zeiss, Axiovert 100,objective 50× ). Small pipettes were prepared bymeans of a micropuller (Narishige, Japan). Thetips have to be as smooth as possible since if asharp edge is present, droplets may stick at the tipbecause of edge piercing. The pipettes are fixed on3-direction micromanipulators (Zeiss, Germany)
in order to ensure a free xyz-movement in theexperimental chamber. Emulsion droplets are eas-ily sucked at the pipette tips (inner diameter �5–10 mm) by reducing the water pressure inside thecapillary using the device schematically repre-sented in Fig. 2. Two droplets of around 20 mm indiameter are captured by means of two distinctpipettes (coupled to 2 independent micromanipu-
Fig. 2. Schematic representation of the micromanipulationsetup. A pipette with a small diameter tip is connected by itsopposite side to a pressure-controlling device. This device isrepresented by the vertical glass tube in the right part of thedrawing. It consists of a water reservoir in which a piston isimmersed. The water level in the reservoir sets the hydrostaticpressure Ppip at the tip of the pipette. This latter may be easilyvaried by simply displacing vertically the piston. In theconfiguration of the scheme, we have PpipBPout (Pout beingthe pressure in the cell), and thus an emulsion droplet (blackcircle) suspended in a solution may be sucked at the tip of thepipette.
L. Bonakdar et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 176 (2001) 185–194188
lators) and brought into contact. The TTAB con-centration in the continuous phase (1 CMC) issuch that the droplets do not stick on pipettes butcoalescence between two droplets occurs oncethey are pressed one against the other. Then, onepipette is detached while the other one holds thetwo fusing droplets in the microscope field (seeFig. 3). A video camera is used to follow thecontraction kinetics. At regular intervals, imagesare grabbed and stored in a computer and laterthe dimensions of the fusing droplets at a time tare measured using standard image processingsoftware. As can be seen in Fig. 3 and schemati-cally in Fig. 4, the shape relaxation process in-duces a gradual decrease of the axial length L(t)along the line that joins the droplet centers. Thecontraction process was followed using the so-called anisotropy factor p(t) defined as: p(t)=H(t)/L(t), where H(t) is the largest dimension inthe direction perpendicular to the line joiningdroplet centers (Fig. 4). In the following calcula-tions, the initial radii of the two droplets will benoted R and bR with b]1. Comparison betweensystems having different initial anisotropy ratiosp0 may be performed by calculating the normal-ized anisotropy (1−p(t))/(1−p0). In Fig. 5, weplot the evolution of the normalized anisotropy asa function of time for the two chosen bitumenswith penetration grades equal to 180/220 and80/100. In these experiments:
R=11.8 mm and b=1.5for bitumen 180/220,
R=15.2 mm and b=1.4for bitumen 80/100.
It can be observed that the normalized an-isotropy is linearly decreasing with time in thefirst stages of contraction but deviates from lin-earity in the latest stages. This evolution is in
Fig. 3. Shape relaxation of two bitumen droplets (180/220)undergoing coalescence (a) t=0, (b) t=30 s, (c) t=300 s. Fig. 4. Scheme representing the shape relaxation process.
L. Bonakdar et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 176 (2001) 185–194 189
Fig. 5. Normalized anisotropy for the two bitumens as afunction of time R=11.8 mm and b=1.5 for bitumen 180/220; R=15.2 mm and b=1.4 for bitumen 80/100.
=23 mJ/m2.The g/h values are very low and justify the
extremely low shape relaxation process (Fig. 5).Indeed, the relaxation characteristic time is of theorder of some hours for 10 micron-sized dropletsof bitumen 80/100 and of the order of some minutesfor bitumen 180/220. For comparison, dodecanedroplets of the same size stabilized with the samesurfactant should relax their shape with a rate 108
to 109 faster!
2.3. Contraction experiment
We now explore the consequences of this ex-tremely low relaxation process on the macroscopicevolution of a destabilized emulsion. For thatpurpose, we use monodisperse model emulsionswhich are made suddenly unstable upon additionof a suitable chemical. In general, when there is noenergy barrier for coalescence, the droplets coalesceas soon as they collide under the effect of Brownianmotion. This non-activated coarsening has beenidentified in the late stages of phase separations orin strongly unstable emulsions. The main limitwhich was most extensively studied to date corre-sponds to systems in which the characteristic shaperelaxation time Tr is shorter compared to the timeTb separating two droplet collisions. In this limit,it was found both theoretically and experimentallythat the average droplet size scales with time as t1/3
in 3-D systems [8]. A very different scenario isexpected in the limit where Tr is much larger thanTb. In this limit, the coarsening is limited by shaperelaxation leading to very different structures andkinetics than in the previous case. Though this limitis frequently encountered in systems like emulsionsof highly viscous substances (asphalt, colophon) orphase separations in binary mixtures of polymers,it has not been systematically explored so far.
First, a crude bitumen emulsion was prepared bygently shearing surfactant and bitumen at a temper-ature of 95°C [9]. The composition of the bitumen,TTAB and water was 80:6:14 by weight.
qualitative agreement with the theoretical calcula-tions of Martinez et al. [7]. These authors havecalculated the linear shrinkage along the axis con-necting the centers of two particles L0−L(t)/2R,where the subscript 0 refers to initial configuration.They found that the initial rate of shrinkage isalmost linear with time whatever the initial dropletsize ratio b :
(L0−L(t))2R
:ag
hRt for
gthRB1,
where a is a constant close to 1/3. Since in the earlystages of shrinkage H(t):2bR, we deduce that thenormalized anisotropy should vary as
1−p1−p0
:1−ab
(1+b)gthR
.
From the experimental slope of (1−p)/(1−p0)versus time, we deduce the g/h values (whichcharacterizes the rate of relaxation) as well as
L. Bonakdar et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 176 (2001) 185–194190
After emulsification we obtained a polydisperseemulsion with size distribution ranging from 0.1to 2 mm. The droplet size distribution was mea-sured using a Malvern Particle sizer (MastersizerS). From the crude initial emulsion, we obtainedquasi-monodisperse samples of different sizes byusing a fractionation technique [10]. Fig. 6 shows
a typical concentrated bitumen emulsion obtainedafter fractionation. After the size fractionationprocess, the continuous phase of the emulsioncontains excess of surfactant. We centrifuged theemulsion four times and each time the continuousphase was replaced by TTAB surfactant solutionat a given fixed concentration. The emulsion ob-tained after centrifugation was perfectly stableover a period of months.
In order to study gelation and the followingcontraction phenomenon, we introduce the emul-sion of known initial bitumen volume fraction ina cylindrical vessel and we add NaOH at knownconcentration. Since the emulsion is stabilizedusing a cationic surfactant (TTAB), the dropletinterface is positively charged. Naturally, bitumencontains some acidic molecules which are alsopresent at the bitumen–water interface. By addingNaOH in the continuous phase, the pH is raisedand the acidic dissociation gives rise to somenegative charges that may totally neutralize theinterface. In the absence of electrostatic repulsion,the droplets become unstable and coalesce. Justafter the addition of NaOH, the emulsion is agi-tated for some seconds and stored at a giventemperature. Initially, the system remains liquid-like, but after sometime the emulsion does notflow any more. At this stage, microscopic obser-vation reveals that the droplets stick together andform a three-dimensional gel network (Fig. 7).Once this network is formed, the gel starts tocontract by reducing its surface area. In this pro-cess water is expelled from the space filling net-work. The contraction remains remarkablyhomothetic meaning that it preserves the geome-try of the container (Fig. 8). We have verified thispoint by performing experiments in a variety ofcontainers such as rectangular, cylindrical, etc.The contraction of the gel made of 80/100 bitu-men at three different times (0, 1 and 2 h), afterthe addition of NaOH, is shown in Fig. 8. Theinitial bitumen volume fraction of the emulsionwas 10%, the NaOH concentration was 0.2 M andthe TTAB concentration was CMC/10. Theamount of NaOH introduced was determined em-pirically in order to get rapid gelation of thesamples (within a few seconds). By measuring thedimensions of the gel, we can evaluate its internal
Fig. 6. Microscopic image of a quasi-monodisperse bitumenemulsion.
Fig. 7. Microscopic image of a gel network obtained just aftergelation.
L. Bonakdar et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 176 (2001) 185–194 191
Fig. 8. Digital camera images of the gel at various time intervals after the addition of NaOH. Bitumen volume fraction=10%.
bitumen volume fraction f. In order to ensurethat we measure the correct volume fraction, weprepare a series of identical samples and, at regu-lar intervals, we remove the gel from the containerand measure the volume fraction by gravimetry.Both methods give identical results within 93%.
Fig. 9 shows the volume fraction of oil in thegel as a function of time for an emulsion with aninitial volume fractions equal to 9% at T=30°C.For this experiment, we used the bitumen withpenetration grade 80/100. The NaOH concentra-tion was 0.2 M and TTAB was at CMC/10. Theradius of the emulsion droplets was 0.25 mm.When NaOH is introduced in the emulsion, adelay of about 30 s has to be expected before thegel starts to shrink. Once the contraction starts,we can clearly distinguish two different regimes.The contraction at the initial stages is quite rapidand exhibits a roughly linear evolution. However,the contraction becomes much slower towards thefinal stages of contraction. This evolution is inperfect agreement with previous observations con-cerning the sintering of solid inorganic particles[11].
There are number of theoretical models for thesintering mechanisms in the case of spheres, cylin-
ders and many other geometrical shapes andconfigurations [5,7,11,12]. Generally, these ap-proaches are difficult to extend to complex disor-dered systems due to geometrical constraints.Quite recently, the sintering was explained interms of connected fractals aggregates that areformed by cluster–cluster aggregation [13]. Weaim now to explain the contraction mechanism inemulsions using the so-called ‘cylindrical model’,which has been found to be sufficiently accurate
Fig. 9. Volume fraction of bitumen in the gel as a function oftime for bitumen 80/100 (initial volume fraction=9%). NaOHand surfactant concentration were 0.2 M and CMC/10 respec-tively. T=30°C. Droplet radius=0.25 mm.
L. Bonakdar et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 176 (2001) 185–194192
Fig. 10. Volume fraction of bitumen in the gel as a function ofreduced time K(t− t0). The solid line corresponds to theoryand the dashed line has a slope of 1.
E: f= (3pha2/l)(dl/dt)2. (1)
The superscript dot indicates a derivative withrespect to time. The energy change due to thereduction of surface area (Es) is given by
E: s= (g dSc/dt), (2)
where g is the interfacial tension and Sc is thesurface area of a full cylinder. The energy balancerequires the following condition:
E: f+E: s=0. (3)
From Eqs. (1) and (2) we get the rate of densifica-tion as
(g/hl0)(1/f0)1/3(t− t0)
=& k
0
2 dx/(3p−82x)1/3x2/3, (4)
where x=a/l. For a cubic cell, x is related to thecylinder volume fraction as
f=3p(a/l)2−82(a/l)3. (5)
f corresponds to the measured volume fraction ofbitumen in the gel. t0 is the fictitious time at whichx=0. In Eq. (4), (g/hl0)(1/f0)1/3=K is a constantfor a given initial volume fraction f0 and initialcylinder height l0.
When the ratio of cylinder radius to its height isequal to 1/2, the neighboring cylinders touchesand the cell contains only closed pores. The corre-sponding theoretical density (volume fraction) ofthe sample would be 0.942. Therefore, the cylin-drical model can explain the densification up to94.2%. However, this value may change depend-ing upon the geometry of the unit cell. For exam-ple, the above values for octahedron andtetrahedron and inverse tetrahedron are respec-tively 0.872, 0.978 and 0.814 [14].
Fig. 10 shows the evolution of the gel volumefraction f of bitumen in the gel as a function ofreduced time K(t− t0). The solid line representsthe theoretical curve obtained using Eq. (4). Forf values between 0.2 and 0.8, the theoreticalcurve is roughly linear with a slope of 1 (seedashed line). Equivalently, within the same f
range, the volume fraction should vary linearlywith time with a slope equal to K. The experimen-tal data from Fig. 9 where recalculated in order to
for the intermediate stages of sintering. Scherer etal. [5,12] developed a simple approach to modelthe sintering mechanism in systems, especiallyglasses or ceramics, containing open pores withinthe framework of Frenkel’s concept [3]. Themodel considers a cubic array formed by inter-secting cylinders which are made up of strings ofparticles. The initial cylinder radius correspondsto the average radius of the particles in the mate-rial. Although the choice of cubic array is anidealized approximation compared to the compli-cated microstructures formed in actual situations,studies have shown that the geometry of the unitcell chosen has very little influence on the kinetics.Extensive calculations reveal that the choice ofdifferent unit cells (octahedral, tetrahedron, in-verse tetrahedron) has little influence on the sin-tering kinetics [14]. To reduce their surface area,the cylinders tend to become shorter and thicker.According to Frenkel’s concept, the rate of vis-cous sintering can be calculated by equating theenergy dissipated in viscous flow to the energygained by reduction of surface area. In thesecalculations, it is assumed that the geometry ofthe cell is preserved. A brief description of thecalculations is as follows.
If h is the viscosity of the dispersed phase, l anda are the length and radius of the cylinder respec-tively, the energy dissipated in viscous flow (E: f) isgiven as
L. Bonakdar et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 176 (2001) 185–194 193
be plotted in reduced coordinates (f= f(K(t−t0)). K is deduced from the initial slope of thecurve f= f(t) (for f between roughly 0.2 and 0.6)and t0 is only a ‘translational’ term which allowsto translate the curves towards the same abscissarange. In Fig. 10, we observe that the experimen-tal data are in reasonable agreement with thetheoretical ones. The figure clearly shows that theinitial stages of densification (up to 60%) correctlyfollow the cylindrical model but however, a devia-tion appears at the final stages of densification.We believe that the deviations are the conse-quence of the formation of closed pores or tosome temporal evolution of the surface propertiesof the gel (in terms of surface tension). Moreover,
it is probable that the model becomes a roughapproximation at high levels of densification.
Although the densification kinetics have beencompared with the cylindrical model, to the bestof our knowledge, no direct experimental studiesthat confirm the effect of viscosity on the densifi-cation rate have been performed. The main con-straint for performing such experiments are due tothe difficulties associated with the measurement ofvery high viscosities in inorganic materials (�1013
Pa s). Since the viscosity of our system was muchlower, we could measure it experimentally andvaried over three orders of magnitude in thetemperature range from 20°C to 60°C (see Fig. 1).Therefore, we have been able to explore the influ-ence of viscosity on the rate of contraction bychanging the temperature. The 80/100 penetrationgrade bitumen was used for this set of experi-ments. Again, the droplets radius was 0.25 mmand the initial volume fraction of bitumen wasequal to 9%. The NaOH and TTAB concentra-tions were 0.2 M and CMC/10 respectively. InFig. 11, we report the evolution of the gel volumefraction of bitumen as a function of time atdifferent temperatures. It can be observed thatlowering the viscosity has the effect to increase therate of contraction. This result confirms that theobserved contraction is controlled by the viscousflow of bitumen through the gel network. Accord-ing to viscous sintering theory, the slope K of thedensification curve should vary as the inverseviscosity at fixed initial volume fraction. Fig. 12shows the evolution of K as a function of viscosityin a log–log scale. The experimentally observedslope of −0.85 is in reasonable agreement withthe expected value, which is −1. Since viscositywas varied with temperature, some slight differ-ence with respect to the theoretical behavior couldbe expected due to the change in surface tensionwith temperature. The relative effect of tempera-ture on surface tension is generally much lowercompared to its effect on viscosity and this is whythe scaling mainly reflects the variation ofviscosity.
Finally, in order to see whether this contractionphenomenon is a universal one, we have preparedanother oil in water emulsion using highly viscouscolophon oil (oil extracted from the resin of pine,
Fig. 11. Temporal evolution of the gel volume fraction as afunction of temperature for bitumen 80/100. Initial volumefraction=9%. NaOH and surfactant concentration all caseswas 0.2 M and CMC/10 respectively. Droplet radius=0.25mm.
Fig. 12. Slope (K) of densification curve as a function ofviscosity for bitumen 80/100.
L. Bonakdar et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 176 (2001) 185–194194
viscosity=105 Pa s at room temperature). As inthe case of asphalt, this oil also contains someacidic groups and we could observe similar con-traction mechanism when the emulsion was al-lowed to break by adding NaOH. Therefore, weconclude that the contraction mechanism is auniversal phenomenon in all emulsion systemsmade up of highly viscous oils.
3. Conclusion
In conclusion, we present experimental evidencefor viscous sintering phenomena in emulsions. Webelieve that this coarsening process takes placeduring the breaking of bitumen emulsions whenrupturing agents are added to the initially stableemulsions. The sintering process may be of greattechnological importance in the field of emulsionssince it makes possible to transform an initiallyliquid emulsion, into a dense and highly viscousmaterial within a short period of time and atroom temperature. This is important when fabri-cating porous materials, because of the difficultyand expense of achieving high temperatures.Lower temperature processing may also be usefulin all cases where coatings must be depositedwithout damaging the substrates. Beside the fieldof applications which may be very wide, the sin-tering process still raises fundamental questionsthat we aim to investigate. Among them, we canmention the evolution of the gel topology (forma-
tion of closed pores) and its influence on the rateof contraction.
Acknowledgements
This work is part of the OPTEL project, finan-cially supported by the commission of the Eu-ropean Communities within the framework of theBRITE EURAM III project.
References
[1] B. Deminiere, A. Colin, F. Leal-Calderon, J. Bibette,Phys. Rev. Lett. 82 (1999) 229.
[2] W. Ostwald, Z. Phys. Chem. 37 (1901) 385.[3] J. Frenkel, J. Phys. (Moscow) 9 (1945) 385.[4] J.K. Mackenzie, R. Shuttleworth, Proc. Phys. Soc. Lon-
don 62 (1949) 833.[5] G.W. Scherer, J. Am. Ceram. Soc., 60 (1977) 236; 60
(1977) 243.[6] E. Evans, D. Needham, J. Phys. Chem. 91 (1987) 4219.[7] J.I. Martinez-Herera, J. Derby, J. Am. Ceram. Soc. 78
(1995) 645.[8] K. Binder, D. Stauffer, Phys. Rev. Lett. 33 (1974) 1006.[9] M.P. Aronson, M.F. Petko, J. Colloid Interface Sci. 159
(1993) 134.[10] J. Bibette, J. Colloid Interface Sci. 147 (1991) 474.[11] G.W. Scherer, C.J. Brinker, P.E. Roth, J. Non Cryst.
