PHYSICAL REVIEW E

VOLUME 57, NUMBER 1

JANUARY 1998

Structure and dynamics of electrorheological uids

Thomas C. Halsey
Exxon Research and Engineering, Route 22 East, Annandale, New Jersey 08801

Randall Kamien
Department of Physics, University of Pennsylvania, Philadelphia, Pennsylvania 19104 Received 14 June 1996; revised manuscript received 26 December 1996 We have used two-dimensional light scattering to study the structure and dynamics of a single-scattering electrorheological uid in the quiescent state and in steady and oscillatory shear. Studies of the quiescent uid show that particle columns grow in two stages. Particles rst chain along the electric eld, causing scattering lobes to appear orthogonal to the eld, and then aggregate into columns, causing the scattering lobes to move to smaller angles. Column formation can be understood in terms of a thermal coarsening model we present, whereas the early-time scattering in the direction parallel to the eld can be compared to the theory of line liquids. In simple shear the scattering lobes are inclined in the direction of uid vorticity, in detailed agreement with the independent droplet model of the shear thinning viscosity. In oscillatory shear the orientation of the scattering lobes varies nonsinusoidally. This nonlinear dynamics is described by a kinetic chain model, which provides a theory of the nonlinear shear rheology in arbitrary shear ows. S1063-651X 97 10112-X PACS number s : 47.50. d

I. INTRODUCTION

Electrorheological ER uids 1,2 are made by suspending particles in a liquid whose dielectric constant or conductivity 35 is mismatched in order to create dipolar particle interactions in the presence of an ac or a dc electric eld. ER uids rapidly solidify, or at least increase their viscosity dramatically, in response to an electric eld, due to the formation of particle chains that bridge the electrodes. The millisecond response of ER uids has piqued the interest of engineers, who are now trying to incorporate these uids into practical fast electromechanical actuators, such as ber spinning clutches and active shock absorbers. To predict the behavior of ER uid-based devices a designer must have a good understanding of the both the solidication kinetics and the response of the activated uid to mechanical stress, especially when ow is induced. We have previously reported preliminary light-scattering studies of the evolution of structure in a quiescent uid 6 , the steadystate structures that form in shear 7 , and the nonlinear dynamics of chain structures in oscillatory shear 8 . In this paper we extend those investigations and give a complete account of the light-scattering studies we have made as well as the theoretical ideas that can be invoked to understand the data. The rst experiments we report concern the evolution of structure after an electric eld is applied to a quiescent ER uid. This occurs in two principal phases: the fast aggregation of particles into chains, which occurs in milliseconds, and the slow coalescence of chains into columns, which occurs in minutes. Chain formation has been studied by timeresolved studies of the transmission of light through an opaque ER uid 9 , birefringence and dichroism 10 , and uid permittivity 11 . The light-scattering measurements we
1063-651X/98/57 1 /756 20 /$15.00 57

report compliment these measurements since they allow determination of the slow column formation as well. In the Halsey-Toor theory of column formation 12 it is shown that after particle chains form and span the electrodes, the long-range dipolar interactions between vicinal chains are screened by the image dipoles created by the conducting electrodes and the chains then interact via a short-ranged potential that arises from one-dimensional Landau-Peierls charge-density uctuations. The chains then coalesce in the plane orthogonal to the eld to form a three-dimensional solid 13 . Our measurements of the power-law growth of columns as a function of applied voltage and uid concentration support this general description of coarsening, but the observed voltage dependence has led to the consideration of the case wherein column coalescence occurs before a chargedensity uctuation can relax. The anisotropic bow tie light-scattering pattern we observe is similar in appearance to that predicted by the theory of line liquids. By analyzing our data in ways suggested by this theory we show that there is a strong correspondence of the predictions of line liquid theory to scattering data taken shortly after the quench. For completeness we mention that within the columns the colloids eventually form a crystalline solid 13 that recent calculations show is body-centered tetragonal 14 . This bct structure has been observed in a laser diffraction experiment from a column of large silica spheres 15 . Of course, a glassy structure may result from a rapid, deep quench. When an initially quiescent ER uid is subjected to steady or oscillatory shear, the electrode-spanning columns break into volatile structures that continuously fragment and aggregate, and generally tilt away from strict eld alignment in the direction of uid vorticity. There are two fundamentally different theoretical models of structure and rheology in shear:
756 1998 The American Physical Society

57

STRUCTURE AND DYNAMICS OF . . .

757

the equilibrium droplet model 16 and the kinetic chain model 17 . In both of these models we use the point dipole approximation, which is reasonably accurate for our colloidal silica uid, which has negative dielectric contrast. In the droplet model, which was developed for stationary shear, a free energy is minimized in order to compute the equilibrium size, shape, and orientation of particle structures that are presumed to be ellipsoidal. In the kinetic chain model, which can be applied to both stationary and non-stationary shear, a mechanical stability condition determines the metastable size and orientation of structures that are presumed to be chains. The predictions of these models are contrasted as we compare them to our light-scattering data. Light-scattering measurements in steady shear 7 show how the steady-state size and orientation of these particle structures depends on the shear rate, with the size decreasing and the tilt angle increasing as the shear rate increases. These changes are responsible for the shear thinning of the viscosity of ER uids 16 . We have extended these measurements to include the voltage dependence. We report light-scattering studies of the structural dynamics in oscillatory shear ow. These measurements should be of special interest to those interested in modeling the uid stress response to nonstationary shear ows. Direct rheological studies 18,19 have been complicated by the difculty in nding a linear-response regime. It has been shown that if a strict linear viscoelastic regime exists at all, it is conned to strain amplitudes smaller than 10 2 . Thus the framework of linear viscoelasticity, a phase-shifted stress in response to an applied strain, is of limited utility for ER uids. Powell 20 has presented rheological data that demonstrate the nonlinear nature of the stress response and these agree quite well with the light-scattering data we present here. In our measurements of the orientation dynamics in oscillatory shear we nd a quasilinear response regime at small strain amplitudes and are able to determine experimental conditions where the droplet orientation is in or out of phase with the strain. This quasilinear regime is successfully described by the independent droplet model, wherein the droplet size does not change during a shear cycle. However, at strain amplitudes larger than about 0.25, chain volatility during a single strain cycle becomes signicant and the orientation dynamics becomes strongly nonlinear. In this regime the kinetic chain model gives a very reasonable description of the observed dynamics and agrees well with direct rheological measurements 20 of the stress.
II. EXPERIMENT A. Sample preparation

shaken and allowed to react without stirring for 2 h. The hydrophilic silica spheres are then coated with 5.0 ml of the organophilic silane coupling agent 3- trimethoxysilyl propyl methacrylate via a condensation reaction 21 . After a 24-h vacuum distillation of water and ammonia at 50 C the spheres are centrifuged at low acceleration 35 g for 7 h, the supernatant decanted, and the soft colloidal solid is redispersed in 4-methylcyclohexanol, again chosen to closely index match the spheres. The sample is centrifuged again at 35 g for 16 h, the supernatant decanted, and the solid resuspended in 4-methylcyclohexanol. To prevent settling the nal sample is rotated slowly until used. Scanning electron microscopy and elastic and quasieleastic light-scattering measurements indicate that 0.7- m-diam silica spheres are easily formed at high silica concentrations under mild hydrolysis with 0.5 M NH4OH. The elastic lightscattering data are consistent with a Gaussian sphere radius R distribution having R /R of 10.5%. Light-scattering intensity measurements indicate a refractive index increment of dn/dc 0.0017 ml/g, which is small enough to ensure single scattering from concentrated dispersions: Indeed, depolarization of the scattered light was negligible. To measure the surface charge of the colloids, electrophoresis measurements were made in a Pen Kem Laser Zee microelectrophoresis apparatus. Since we were unable to observe any electrophoresis with this apparatus we simply applied a 1-kV/mm electric eld to the particles and observed their behavior through a Nikon Microphot-FXA optical microscope. Even at these high electric elds we were unable to observe electrophoresis of these particles, although at high applied frequencies eld-induced particle chaining was observed and found to be reversible by Brownian motion alone, indicating that contact interactions between these particles are much smaller than k B T. The two samples used in the study measured 20 and 34 wt. % by thermal gravimetric analysis, although the highconcentration sample was diluted to 11 wt. % for the kinetics studies. The uid used in the oscillatory shear study is 7.5wt. % silica, which is 3.0 vol %, assuming a silica specic gravity of 2.5. This corresponds to a mean separation between silica sphere centers of 6R and a spacing of 11R between initially formed chains.
B. Two-dimensional light scattering

The sample used in this study is a model ER uid we developed for light scattering 6 , microscopy, and electrorheological measurements 16 . The colloids in our model uid are synthesized by the base-catalyzed nucleation and growth of monodisperse silica spheres from tetraethoxysilicon TEOS . To reduce the Keesom interactions that lead to aggregation, this synthesis was conducted in a mixed organic solvent that index matches the growing spheres. Specically, to prepare a 200-ml solution we combine 21.2 ml of 29.5wt. % NH3 with 39.2 ml of formamide, 117.2 ml of benzyl alcohol, and 22.4 ml TEOS. The solution is mixed but not

To study the kinetics of phase separation requires the ability to determine structure as a function of time. Traditional one-dimensional light-scattering instruments must repetitively scan through a sequence of angles, with the result that data are acquired in an interval during which the structure is evolving, so that temporal resolution is severely compromised. Furthermore, the ER phase transition gives rise to anisotropic scattering and therefore a two-dimensional detector is required. To meet these demands we have developed a light-scattering instrument that is based on readily available video and computer technology. A 454.5-nm argon-ion laser beam, focused with a 40-cm focal length lens, illuminates the sample, the scattered light impinges on an opaque diffusing screen, is collected by a xed-gain Pulnix video camera, and is stored on a video cassette recorder at a spatial resolution of 640 480. The

758

MARTIN, ODINEK, HALSEY, AND KAMIEN

57

video frames are then grabbed from the tape and eight-bit digitized by a Perceptics PixelBuffer card on a Macintosh Quadra 950. The frame grabber card immediately transfers these images to a 32-Mbyte dual-ported PixelStore memory board that resides on both the slow computer bus and a very fast direct bus from the frame grabber that can support the data transfer rate of 10 Mbyte/s. The digitized images have a dynamic range of 256, a spatial resolution of 512 480, and 1 a temporal resolution of 30 s. The length scale regime that can be studied with 454.5-nm light is from 2 /q 22 to 0.95 m, where q 4 sin( /2)/ is the scattering wave vector, is the scattering angle, and is the wavelength in the scattering medium. The scattered light that ultimately arrives at the charge coupled device array is not the true scattered intensity that emerges from the scattering volume. In fact, the measured scattered intensity increasingly underestimates the true intensity as the scattering angle increases. The loss of scattered light is due to a number of causes and although each of these is small, they combine to give a correction factor of almost two at the highest measured wave vector, relative to zero angle. We correct for the incident polarization of the laser, the polarization and incident-angle-dependent cell and scattering screen reectances, the angular distribution of the emerging light from the diffusing screen, camera vignetting, the Jacobian for light refraction, and the Jacobian of the projection of a sphere onto a at screen twice . We calculate each of these corrections analytically and apply these to a 512 480 constant intensity image to create a 512 480 template. Each scattering image was then corrected by multiplying by the template on a pixel-by-pixel basis. Calibration runs with a uniform scatterer yielded an essentially constant scattering image after these corrections were applied, with only a small drop in the intensity at high q.
C. Scattering cells

FIG. 1. Within milliseconds of an electric-eld quench into the two-phase region the scattered intensity from an electrorheological uid shows two distinct lobes the maximum scattering angle is 8 here . These lobes, which are orthogonal to the electric eld, indicate the presence of an unstable concentration uctuation that is due to column formation. This scattering pattern is the twodimensional analog of the spinodal ring commonly observed in the rst-order phase separation of three-dimensional systems.

and the strain frequency can be varied by adjusting the microstepping controller speed. The detector response time limits the maximum frequency to 1 Hz. The electric eld is supplied by a square wave from a Trek power supply driven by Wavetek signal generator. Voltages are reported peak to peak. The dielectric constant of 4-methylcyclohexanol is 13.5 and silica is 4.
D. Strain phase determination

The scattering cell for the quiescent uid studies consists of a black nylatron body with a cylindrical cavity into which 1.0-mm-diam black anodized cylindrical aluminum electrodes were threaded. Flat glass microscope slides were pressed against rubber O rings to provide a seal. A 12.0-kHz sine wave was applied to the 0.72-mm electrode gap to induce particle chaining. The scattering cell for the shear studies consists of an inner 40 2 -mm at circular electrode that is concentric to a 42-mm hole in an outer electrode, creating a radial electric eld in a 1.0-mm gap. The outer electrode is sandwiched between plastic and both electrodes are embedded between glass plates, with a uid-lled 2.0-mm gap between the inner circular electrode and each glass plate. The radial electric eld is parallel to the shear gradient in the uid. For the steady shear studies a dc servo motor was used to drive a pulley on the inner electrode. In the oscillatory shear studies the inner electrode was caused to oscillate sinusoidally by a long rod connected to a 25-mm lever on the electrode shaft, which is driven open loop by an adjustable eccentric shaft on a powerful 300-oz-in. microstepping motor. The strain amplitude can be varied by adjusting the eccentric

One problem in this experiment is the phasing of the applied strain with the scattering data. To solve this problem, we devoted a small corner of the scattering screen to an optical strain phase clock. This clock was created by running a second pulley, whose rotational axis is orthogonal to the scattering screen, synchronously with the stepping motor. Inside this second pulley is mounted a prism that deects by several degrees a He-Ne laser beam directed toward the scattering screen. As the pulley turns, the laser beam scribes a small circle on the scattering screen. The strain phase clock can be set by means of a rotational adjustment of the prism. Each image thus contains both the scattering data and the absolute strain phase.
III. QUIESCENT FLUID

The rst measurements we report concern the growth and structure of the uid in the quiescent state. Preliminary measurements of the structure orthogonal to the eld were reported earlier 6 , here we extend those results and present an analysis of the structure parallel to the eld.
A. Domain structure orthogonal to the eld

Scattering data taken shortly after an electric-eld quench, shown in Fig. 1, demonstrate an unstable concentration uctuation orthogonal to the electric-eld lines in the uid. The two scattering lobes have an intensity maximum at some

57

STRUCTURE AND DYNAMICS OF . . .

759

FIG. 2. Intensity slices through the lobes, i.e., orthogonal to the electric eld, indicate that the peak intensity increases with time and moves to larger length scales smaller q . The scattering functions are noisy because uctuations are slow; the number of particles in the scattering volume is small.

nonzero wave vector, indicating strong spatial correlations between initially formed chains in the plane orthogonal to the eld. The scattering patterns fades when the eld is turned off, indicating the reversibility of particle chaining. The Bragg scattering one would expect from chains of regularly spaced particles is at values of q larger than observable with our instrument, but can be seen on the image screen. Reversible chaining is also observed in direct optical imaging. The distinctive scattering lobes are supercially the twodimensional counterpart to the spinodal ring observed in three-dimensional systems and thus compel a comparison to this model. In fact, if the laser beam is directed along the eld lines, by using transparent electrodes, a scattering ring can be observed. We will show that more information can be obtained in our scattering geometry. The time evolution of the structure factor can be analyzed by taking a slice through a lobe in the direction orthogonal to the electric eld lines, as shown in Fig. 2. Each of these single video frames is intrinsically noisy because the number of chains in the scattering volume is small, but it is nonetheless clear that the scattered light increases in intensity and the peak position moves to smaller q as time evolves. An elementary question is whether or not the domain structures scale, i.e., merely enlarge with time while maintaining the same morphology. In a scattering experiment two fundamental quantities are obtained: a characteristic domain length L and the domain mass within a volume of size L d , where d is the spatial dimension. In our case the characteristic length can be identied with the inverse peak position L(t) 2 /q max(t) and the characteristic mass is the peak intensity I max . The collapse obtained by plotting the scattering data against axes normalized by these quantities Fig. 3 demonstrates that the domains scale.

FIG. 3. A master curve is obtained when the scattered intensity data are plotted on dimensionless axes, indicating scaling of the domains. Averaging these data results in the inset scattering curve, which demonstrates a q 3 fall off the data on the high-q side of the peak. This falloff, which is Porods law for a two-dimensional system, indicates sharp nonfractal interfaces on the columns. The scattering function on the low-q side of the peak increase as q 2 . This is expected for spinodal decomposition and is a consequence of a conservation law.

We can analyze the data further to understand the structure of the domains. For example, if the aggregating chains stick upon contact, without rearranging to minimize their electrostatic energy, then one might expect mass fractal, twodimensional cluster-cluster aggregation to describe the structure of a cross section of a column. It is also possible that sheetlike cross sections evolve or even surface fractals. In fact, the scaled and averaged scattering data inset in Fig. 3 have a high-q shoulder that decays as q 3 , which is Porods law for two dimensions. Porods law indicates the presence of sharp nonfractal interfaces, as expected for a system that is spinodally decomposing. By comparison, mass fractal scattering would yield a high-q decay that is slower than q 2 for a two-dimensional system. It is also possible to test for the formation of fractal domains by the relationship of the characteristic mass and 2.25 length. The data in Fig. 4 demonstrate that I max qmax , suggesting that within experimental error the standard scaling relation for spinodal decomposition I t
d q max t f q/q max t ,

f x

d 1

for x 1

applies since d 2. However, this result is simply a consequence of the growth of nonfractal domains that have a sizeindependent morphology and so is more general than spinodal decomposition.

760

MARTIN, ODINEK, HALSEY, AND KAMIEN

57

FIG. 4. The peak intensity I max scales with the peak position as 2.25 I max qmax . Within the errors of this experiment this indicates that d the scaling relation I max qmax is obeyed since the dimension d of this system is 2. B. Domain growth

FIG. 5. The kinetics of structural coarsening is well described by the power law L(t) L(0) 1 (t/ ) 2/5 , where L(t) 2 /q max is the characteristic length. For spinodal decomposition in a system with a conserved order parameter one expects L(t) t 1/3. An important conclusion of these experiments is that the rate of coarsening is voltage dependent, as quantied in the inset.

Experimental studies of the domain growth kinetics were complicated by the intrinsically noisy scattering signal, which led to unacceptable errors in judging the peak position and intensity from the scattering data. To reduce the noise in the characteristic length L(t) 2 /q max(t) we computed the moments I k q l ,q u
qu ql

q k I q dq.

This integral converges only for k 2, so we used the zeroth and rst moments. In our experiment the lower integration limit q l 0.33 10 3 nm 1 is determined by the beam stop radius and the upper limit q u 0.529 10 2 nm 1 is set by the camera position. These experimentally xed integration limits cause truncation errors in the determination of the moments, a point that deserves some consideration. Truncation errors are of no consequence in determining the functional form of the growth kinetics, provided that the relative error is independent of the domain size. For example, the moments can be determined self-consistently by taking advantage of the domain scaling implied by Eq. 1 . This is accomplished by scaling the integration limits with the domain size, so that aq l bq u q max(t), where a and b are arbitrarily chosen to maximally exploit the available data. To do this we xed a and b, determined an initial value for q max by integrating over all of the data, computed a new q l and q u , recomputed q max and continued to iterate. These partial scaled moments are strictly proportional to the true moments, but in practice we found that simply integrating over the nite data gave moments that scaled similarly to the partial scaled moments, i.e., q max(t) I 1 (aq max ,bq max)/ I 0 (aq maxbqmax).

The time dependence of the characteristic length is shown in Fig. 5 for peak-to-peak voltages of 0.56, 1.25, and 2.5 kV across the 0.72-mm gap. At the earliest times the characteristic length is about L(0) 1.9 mm, after which a linear increase of L(t) with t 2/5 is then observed at all voltages, with the growth rate increasing with voltage. In fact, a nonlinear least-squares t to L(t) L(0) 1 (t/ ) gives an average exponent of 0.42, but we plot against 2/5 since this is close. For spinodal decomposition in a system with a conserved order parameter it is thought that after a linear, CahnHilliard-Cook regime there is a nonlinear growth regime 1 with a characteristic length that increases like q max t1/3. Thus, although the electrorheological phase transition has many of the salient features of two-dimensional spinodal decomposition with a conserved order parameter, the growth kinetics appears to be somewhat faster than t 1/3. The dependence of the growth rate on voltage is shown in the inset graph in Fig. 5. The rate increases slightly less than linearly with voltage and so rules out the eld-squared dependence one might naively expect by noting that the solvent friction is eld independent and that a dipole attraction scales as the eld squared. However, these observations can be understood within the framework of uctuation-induced coupling between columns, discussed below. Finally, we completed studies of the dependence of column formation on the colloid volume fraction c . The results in Fig. 6 show that the initial length scale decreases with increasing concentration, but that the growth rate increases with concentration, causing a crossover in the growth curves. These curves were best linearized with a growth exponent of 0.6, the disparity between this exponent and the 0.4 value in Fig. 5 being due to the intrinsic noise in the experiment. To rst order the initial length scale should

57

STRUCTURE AND DYNAMICS OF . . .

761

E2

k B TR
4

at a transverse distance from the column. Thus there is a uctuating force E dip per unit length between columns of radius R separated by a distance , which may be either attractive or repulsive and is of the order of magnitude F dip k B T ER 3/5
3

FIG. 6. The coarsening kinetics was also determined as a function of the initial particle volume fraction the legend gives the colloid concentration in wt. % . As expected, the initial column spacing L(0) decreases with increasing concentration, whereas the column growth rate increases with increasing concentration, causing a crossover in the growth curves. The best power-law t to these data is L(t) L(0) 1 (t/ ) 3/5 . This coarsening exponent of 3/5 is greater than the value of 2/5 obtained in our rst experiments. This variation is due to the large intrinsic noise in the experiment; a realistic estimate of the growth exponent is 0.5 0.1.

since the dipole moment per unit length of a column is ER 2 . The dominant force for columns separated by a distance 1 comes from uctuations with k , thus the coherence length of these uctuations will be . Now adjacent sections of length , which will be pulled in different directions by this force, cannot move independently of one another because there is a strong restoring force, on the energy scale of 2 2 3 E R , which keeps the columns parallel to the eld. We do expect, however, that sections of length can move independently, where is determined by balancing this restoring force with the statistical average of the thermal force for a column of length . We thus conclude that satises F dip
2

E 2R 3

decrease as 1/ 1/2 , the mean separation between single c chains that span the electrodes, and the data roughly bear this out. The faster growth rate occurs because closer chains feel a greater attraction.
C. Thermal theory of coarsening

which leads immediately to ( / ) 1/5. The time scale for the columns to be drawn together by this force can be obtained by balancing this thermal force for a column of length against the viscous force F vis C 0 v , where 0 is the solvent viscosity, v is the velocity of the column, and C is a drag coefcient at most logarithmically dependent on R 23 . We thereby obtain a collision time
0 c 3/2

1/10 27/20

The roughly linear dependence of the coarsening rate on electric eld and the approximately root time dependence of the length scale of the condensed phase can be accounted for semiquantitatively if we suppose that coarsening is driven by thermally generated dipole moments. The electric eld of perfectly ordered chains or columns of dipoles is short ranged, decaying exponentially as one moves away from the column in a transverse direction, with a range that is of the order of magnitude of the lattice spacing of the particles in the column. However, chains or columns of dipoles are essentially one-dimensional solids and are thus subject to strong Landau-Peierls uctuations. If the phonon eld longitudinal or transverse of a column is u(k), with k the wave vector along the column, then equipartition implies that 22 u k
2 2

k BT E

We should compare this time to the coherence time of the 1 uctuations of the dipole moment on a scale k . By the uctuation-dissipation theorem we expect that the time for these uctuations to dissipate is f 1/Dk 2 , where D k B T/ 0 R is a characteristic diffusion coefcient for the particles 24 . We thus conclude that for large we will have f c and the uctuations will persist long enough to drive coarsening. We now obtain the estimate t At
5/9

k BT , E 2R 3

where R is the radius of the column and 2 E 2 R 3 is the energy scale for density uctuations of the column. These density uctuations lead to the appearance of a uctuating electric eld near the column. It is easy to show that this electric eld is of the order of magnitude 12

with A ( E) 4/5, in qualitative agreement with the results above. We conclude that the evolution of structure in an ER uid proceeds by an unstable concentration uctuation whose characteristic size scale increases with a power of time. The observed growth exponent of 0.5 0.1 is slightly larger than the 1/3 value expected for spinodal decomposition and somewhat smaller than the 5/9 value predicted by extending the Halsey-Toor theory to account for electric-eld-dependent uctuations in interchain interactions. However, the fact that

762

MARTIN, ODINEK, HALSEY, AND KAMIEN

57

FIG. 7. This diagram shows the denition of scattering wave vectors q and q used to describe the detailed shape of the twodimensional scattering function. The contour lines are from real scattering data.

the characteristic growth rate increases almost linearly with the eld is further evidence that the modied Halsey-Toor theory accounts for the structural evolution in this system.
D. Structure parallel to the eld

FIG. 8. The width of the scattering lobe in the q direction is a minimum at the value of q where the intensity is a maximum. The scattering width is apparently time independent for values of q q ,max . The solid lines are the data smoothed by Fourier ltering.

We have thus far limited our discussion to the scattered intensity as a function of the wave vector q in the direction parallel to the scattering lobes, which is perpendicular to the electric eld. Nelson and Seung 25 have studied the statistical mechanics of line liquids and have noted a striking resemblance of the shape of our scattering images to predictions for line liquids. In this section we explore these connections quantitatively. Before we start our discussion it is useful to examine Fig. 7 to see how a wave vector q that is parallel to the applied eld for each value of q is dened. There are several experimental issues to address. Is the shape of the scattering lobes scale invariant? How does the width of a scattering lobe vary along its length? What is the functional form of the scattered intensity in the parallel direction, that is, I(q ,q c)? We will examine each of these issues experimentally and then will compare our results to theory. The rst issue we will investigate is the variation of the scattering width parallel to the eld as a function of the scattering wave vector q . To do this it is useful to dene the parallel moment q 1/2
0

We can determine whether the shape of the scattering function, hence the column structure, is scale invariant. A scale-invariant scattering function will have a minimum width q 1/2 2 proportional to the peak position. It is convenient to dene the length L 2 / q 1/2 2 and determine if this is proportional to our previously dened length L . Data for a uid subjected to a 0.56-kV/mm eld are shown in Fig. 9. Despite the noise these data convincingly show that L L . Similar results are obtained at larger elds, so on

q 1/2I q ,q

dq
0

I q ,q

dq .

We have chosen to work with root moment to eliminate convergence problems with the integrals, as the intensity falloff is expected to be Lorentzian in the direction parallel to the eld. The results are shown in Fig. 8 for a sample subjected to a 0.56-kV/mm eld at various aging times. A deep minimum in the width of the scattering function is observed at a value of q that decreases with time and a master curve appears to be developing for values of q past the minimum. The minimum scattering width occurs at the value of q that maximizes the scattered intensity I(q 0,q ) and this value of q is proportional to the moment q dened previously.

FIG. 9. The length L is proportional to L , indicating that the shape of the scattering data, and thus the domain structure, is scale independent. Note that there is roughly a one-decade difference in length scales, however, with correlations along the chains extending much further.

57

STRUCTURE AND DYNAMICS OF . . .

763

FIG. 10. The intensity integrated in the q direction should increase linearly with q , and this is found at all times. The same result is found at all voltages.

experimental grounds we conclude that the scattering function and column structure are scale invariant. According to theory of line liquids, the integrated scattered intensity 0 I(q ,q )dq should increase linearly with q for q q ,max . The data shown in Fig. 10 bear this out. This result is due to an underlying conservation law, as discussed below. Finally, the theory of line liquids predicts that the line shape in the q direction is Lorentzian. Determining the line shape is difcult in such a noisy system, so we tried to take full advantage of our data in order to reduce the noise. To average the signal we i chose a value of q , ii computed the width q 1/2 2 and height I(q 0,q ) of a 1-pixel-wide slice in the q direction at this q , and iii nonlinearly binned the normalized data q / q 1/2 2 versus I(q ,q )/I(q 0,q ) the nonlinear bin widths were chosen to keep the signal-to-noise ratio xed . This was repeated for each value of q to obtain a signal-averaged data set at one particular coarsening time. In Fig. 11 we show the resultant time dependence of the line shapes. At early times the data are well described by a Lorentzian, but at later times they decrease more rapidly than Lorentzian for large q / q 1/2 2 . This may indicate that the line liquid theory only holds shortly after chain formation. Taken as a whole, the agreement between the experimental data and the line liquid theory is compelling at short times.
E. Line liquid theory

FIG. 11. The line shape of the scattering data in the q direction is t to a Lorentzian dashed lines at various times after the eld quench. At early times the t is quite good, but at late times the scattered intensity decreases more rapidly than a Lorentzian.

the bottom of the sample, there is an equation of continuity for the areal density (x,y,z), dened so that in an xy cross section at height z 0 , x,y,z 0 dx dy N z0 , 9

where N(z 0 ) is the number of dipole spheres in that section. If the chains are continuous from one electrode to the other, then it is clear that N(z 0 ) will be independent of z 0 . As with the continuity of charge this implies that t 0, 10

where t (x,y,z) is a vector in the xy plane and can be interpreted as the xy projection of the local tangent vector to the chains at (x,y,z) 26 . The conservation law implies that when q x q y 0, z 0 or, in other words, (0,0,q z ) 0, where (q ) (q ). This implies that the structure 0 function S q q q 11

The intensity plots of the structure function of the ER uid at short times after the eld is applied is reminiscent of the equilibrium structure factors of line liquids 25,26 . However, since the interactions between the chains of dipoles are attractive 12 , it is clear that once the dipole chains are formed there is no equilibrium state until the dipole spheres condense into their nal crystalline form 1315 . Let us rst review some generic features of an ensemble of directed lines. Most notably, if the lines go from the top to

will vanish when q x q y 0. This is a generic feature of line liquids, independent of equilibrium, and is seen here in the data. Of course if the chains are broken or contain branches, the continuity equation will be modied by the addition of a source term. For large values of q the dipole chains will appear to be unbroken if the density of chain ends and branch points is low enough. At smaller q the structure function will cross over to that of an isotropic liquid, at long enough length scales even extended but nite objects will appear as points 26 . Equally generic to the structure function of line liquids is a peak at some value of q corresponding to the average

764

MARTIN, ODINEK, HALSEY, AND KAMIEN

57

interchain spacing. This is simply due to the incipient liquid or crystalline order that will emerge at lower temperatures. Again, the data show these peaks. To consider the dynamics of the coarsening, we note that the chains form almost immediately. Thus we can view the coarsening of the dipole uid as the coarsening of a line gas suddenly quenched into the crystal phase and ignore the complication of chain formation. Note that although the number of dielectric spheres is constant with time, the number of chains need not be as they can join and coalesce even in the absence of chain ends and branch points. Thus we suggest a nonconserved coarsening dynamics. We model the change in areal density according to the Langevin equation x t F x x,t , 12

S2 q

ei ei

qxx qy y

S x,y,z 0 dx dy S x,y,z dq z , 2 z dx dy dz 17

qxx qy y

S q x ,q y ,q z

where (x,t) is a noise term with correlation chosen to ensure that the system comes to thermal equilibrium, namely, x,t x ,t k BT
3

where the last equality comes from the convolution theorem. The data have been reduced to this form by integration along q z . We see again two generic features of the line liquid. The plot of I(q ) q 2 /S 2 (q ) versus q should rise linearly at small q and then have a dip at q * , the wave vector of the incipient crystalline order at the relevant density. According Bq 2 Cq 4 , in agreeto the model outlined above, I(q ) ment with the data for small q . We may also look at the time dependence of I(q ;t). Using Eqs. 15 and 17 we have I q ;t e2
r B t

I q ;0

18

x x

t t .

13 for small times. While there is no time at which an equilibrium line liquid exists, at short enough times the system is not far from the dilute gas that it started as on its way to a crystal. A more detailed theory of the coarsening would require the introduction of nonlinearities along the lines of Langer, Bar-on, and Miller 28 .
IV. STEADY SHEAR

Using the model free energy 26 for a line liquid, we are led to the linearized equation for the structure function 27 : S q ;t t Cq 2 q2 z q2

r S q ;t

2 k B T, 14

where r 0 and C and K are constants related to the stiffness and density of the dipole chains 26 . The nonlocal nature of the kernel in Eq. 14 is a consequence of the conservation law. For short times Eq. 14 can be solved and we nd that S q ;t e
w q t

S q ;0

1 e

w q t

S q;

15

with w(q ) 2 Cq 2 K(q 2 /q 2 ) r . Thus the structure z function will have the form of a line gas with an exponential rq 2 the modes deq -dependent decay. For Cq 4 Kq 2 z 4 2 2 Kq z rq the modes grow. Thus we cay, while for Cq expect that for very small q z there will be a regime of q for which the scattered intensity will grow at early times. The structure function at zero time should be that of a line gas, namely,
2 2 0q 4

S q ;0

Bq 2

Cq

Kq 2 z

16

The shear thinning of the uid viscosity is perhaps the most basic aspect of electrorheology. For the colloidal silica uid studied here we found 16 that at low applied elds the 2/3, where is the shear viscosity shear thinned as rate, whereas at high applied elds the standard result 1 was obtained. The discovery of the anomalous 2/3 shear thinning exponent prompted the development of the independent droplet model, which gives this exponent. This model makes some very specic predictions about the droplet size and orientation as functions of the shear rate and eld and these can be determined by light scattering. Likewise, the kinetic chain model, originally developed to account for oscillatory shear, gives a shear thinning exponent of 1 and makes dramatically different predictions for the chain orientation. In the following we report measurements of the orientation of particle structures and compare these to the kinetic and equilibrium models.
A. Droplet fragmentation and orientation

where B 0 is related to the average spacing and this expression is valid for small q . At larger q there will be additional terms leading to the peak that represents the average chain spacing. This form also predicts that at xed q the structure function will fall off along q z with a Lorentzian line shape, in agreement with the data. The three-dimensional structure function gives us information about the two-dimensional structure as well. Since the two-dimensional structure function S 2 (x,y) S(x,y,z 0) we have

When an ER uid is subjected to shear, the columns fragment and tilt in response to the hydrodrodynamic forces, resulting in several changes in the light-scattering pattern Fig. 12 . First, the coarsening of the lobes stops and the scattering pattern reaches a steady state, as shown in Fig. 13. Second, the scattering pattern is rotated in the direction of uid vorticity. Finally, the peak of the scattering lobes decreases to q 0 Fig. 14 , showing that the quasiperiodic intercolumn correlations are destroyed. These observations indicate rotated structures whose spatial correlations have

57

STRUCTURE AND DYNAMICS OF . . .

765

FIG. 12. When an ER uid is subjected to steady shear the columns fragment into droplets that tilt in the direction of the uid vorticity to some equilibrium angle max relative to the electric eld. A more subtle change is that the maximum of the scattering lobes moves to zero scattering wave vector, indicating a loss in spatial correlation between droplets.

been destroyed by shear. By measuring the degree of rotation as a function of shear rate, we can directly test models of the steady-state uid structure. To characterize the scattering data it is necessary to have a systematic method of determining the angle max by which the scattering lobes of the sheared uid are rotated relative to those of the quiescent uid. In the following, positive max indicates a rotation in the direction of the uid vorticity. To determine max we rst divide a time-averaged scattering image into 360 wedges that each subtend 1 of arc. The scattered intensity in each wedge is computed by integrating the intensity from the experimentally determined limits q l to q u . We have termed this technique longitudinal scattering analysis 7 by analogy to the latitudes and longitudes of the

FIG. 14. A radial slice of the scattered intensity taken through the center of the tilted lobes shows that the maximum scattered intensity occurs at zero-scattering wave vector, unlike in the quiescent uid, where a peak occurs at nite q.

FIG. 13. In steady shear the scattering pattern reaches a steady state, in contrast to the coarsening that occurs in the quiescent uid. These data show how the scattered intensity approaches a steady state after the shear rate is increased from 0.89 to 1.04 s 1. The decay time of 33 s is quite large compared to the reciprocal shear rate, indicating that the droplet-droplet collision time may be an important time scale.

Earth; if the laser beam is considered as the axis, then we have integrated over the latitudes to obtain the intensity as a function of longitude. Furthermore, the prime meridian is orthogonal to the direction of the electric eld. In Fig. 15 we contrast the longitudinal analysis for the quiescent uid with that of a uid sheared at a rate of 1.06 s 1. The scattering maximum of the sheared sample is shifted to positive and the scattering half-width is broader than in the sheared uid. The peak position and halfwidth were extracted by tting the Gaussian I( ) 2 2 max) /2 I(0)e ( to the data Fig. 16 , which worked well, despite the slight bilateral asymmetry of the scattering lobes. We rst investigated the dependence of the droplet orientation on the applied electric-eld frequency. This dependence is shown in Fig. 17 for a sample at 0.8 kV and a shear rate of 1.4 s 1. Because the angular displacement is nearly constant over this frequency regime, we can conclude that the polarizability is essentially frequency independent. We arbitrarily chose 1.0 kHz as the standard operating frequency for our studies, although some shear rate studies were also done at 400 Hz. It is worth noting that because the particle polarizability is fast and the shear rate is slow, there is no chance of dephasing the particle dipoles by particle rotation. Particle dephasing would reduce the dipolar interactions at high shear rate. The dependence of the droplet rotation angle with shear rate is shown in Fig. 18. These data were taken at an applied frequency of 400 Hz and at a peak-to-peak voltage of 1.2 kV across the 1.0-mm gap. A nonlinear least-squares power-law t to the data gives max 0.326, so good linearity is obtained by simply plotting max against 1/3. This cube-root dependence is much weaker than one might naively expect. For a rigid rod in shear the electrostatic torque that tends to maintain eld alignment increases

766

MARTIN, ODINEK, HALSEY, AND KAMIEN

57

FIG. 15. The radially integrated intensity of scattered light, obtained from the longitudinal analysis, is shown for an unsheared and sheared sample in the polar coordinates (I, ). Positive angles are in the direction of uid vorticity and the small background scattering has been subtracted for clarity. Note that for the sheared sample the scattering pattern is angularly displaced, indicating column tilting; broadened, indicating a dispersity of column sizes; and skewed, primarily because of the nonlinear dependence of the orientation angle on droplet size.

FIG. 17. The dependence of the droplet orientation angle on the eld frequency is shown for a 7.5-wt. % sample at 0.8 kV applied across the 1.0-mm gap and a shear rate of 1.4 s 1. The angular displacement is nearly constant in this regime, indicating that the particle polarization is nearly frequency independent. Most of our shear and voltage studies were done at 1.0 kHz, although some shear rate studies were done at 400 Hz. Because the particle polarizability is fast and the shear rate is slow, the particle dipoles will not dephase by rotation in shear.

to rst order as the tilt angle , whereas the hydrodynamic torque increases as the shear rate . In steady state the droplet velocity is zero, so a torque balance gives the linear relation max 1. The sublinear dependence actually observed is

due to the tendency of chains or droplets to fragment in response to ow, as discussed below. The droplet or chain orientation angle can also be studied as a function of the applied electric eld at constant shear rate. Stronger elds should align the structures more closely

FIG. 16. The column orientation can be obtained from a Gaussian t lines to the data or from the position of the intensity maximum. The Gaussian does not t the wings of the data very well, but reliably nds the peak position, while being insensitive to noise. The intensity is radially integrated.

FIG. 18. The linear increase of the droplet orientation angle with the cube root of the shear rate bolsters the independent droplet model of the shear thinning viscosity. This sample was 7.5-wt. % silica and the applied voltage was 1.2 kV at 400 Hz across the 1.0-mm gap.

57

STRUCTURE AND DYNAMICS OF . . .

767

leads to a shear-rate-dependent ER uid viscosity. In the following the electric eld is oriented along the z direction and the shear ow is v (z) zx . We also assume a solvent dielectric constant s 1. The hydrodynamic torque about the y axis for a rigid ellipsoidal droplet rotating at a rate in a shear ow is given by 29 L 2 0V b 2n x c 2n z c 2 cos2 b 2 sin2 b2 c2 , 19 where V is the spheroid volume and n x,y,z are the shapedependent depolarization factors of the spheroid. The depolarization factors of a prolate spheroidal droplet can be expressed as n z n and n x,y (1 n)/2. For extremely prolate spheroids with c b, we can dene g b 2 /c 2 1 to obtain 1 29 n 4 g, where we have dropped a logarithmic term. Thus both g and n are suitable small parameters for expansion. If the droplet is in mechanical equilibrium, then the hydrodynamic torque L must balance the electrical torque K. The electrostatic torque K on a spheroid of dielectric constant inclined at an angle to the eld is 30 K 1 8
2

FIG. 19. The orientation angle is shown as a function of the applied eld, at constant shear rate, for a 7.5-wt. % silica uid. Because the uid is nonohmic at high voltages, it is expected that the particle and uid polarizabilities will be eld dependent, thus obscuring the scaling. Still, at the lower shear rate of 0.34 s 1 the data are in good agreement with the expected cube root law. At the higher shear rate the data are better described by a smaller exponent, near 0.2, perhaps due to the formation of a shear slip instability.

E 2 V sin 2

20

where E is the applied electric eld. If we assume that the droplet angle is small, then balancing the electrostatic and hydrodynamic torques gives 16 1 0 1 1 2E 2 g Mn . g 21

with the eld. Because the uid is non-Ohmic at high voltages, it is expected that the particle and uid polarizabilities may be eld dependent, so the scaling is expected to be somewhat obscured. Still, Fig. 19 shows that the data at a shear rate of 0.34 s 1 are in good agreement with the cuberoot law obtained in the shear-rate-dependent studies. We will now compare these observations with the predictions of the models.
B. Independent droplet model

Here we give a brief derivation of the independent droplet model rst developed to describe the shear thinning viscosity of an ER uid. This model neglects droplet-droplet hydrodynamic and electrostatic interactions and so is most appropriate for dilute suspensions at low Mason numbers, yet it remains informative for larger values of the concentration. This independent droplet model is based partly on the result of Halsey and Toor 13 for the shape of a particle droplet. By balancing depolarization effects against surface tension, one nds that such an independent droplet is roughly a prolate sphere, with the size c of the droplet in the direction parallel to the applied eld related to the size b of the droplet in the direction perpendicular to the eld by b r 1/3c 2/3, d where r d is the radius of an colloidal particle. In shear ow an ellipsoidal droplet will rotate so that its long axis is no longer parallel to the eld. The larger the droplet, the greater the rotation in a shear ow. This rotation reduces the depolarization energy of a droplet and a balance between depolarization energy and surface energy determines the characteristic droplet size. This size computation

2 2 Mn E 0 is the Mason number, which ex0 /2 0 c presses the ratio of hydrodynamic to electrostatic forces between two vicinal spheres in shear in terms of the dielectric contrast factor ( p c )/( p 2 c ), where p is the particle dielectric constant, c is the dielectric constant of the continuous phase, and 0 8.854 10 14 F/cm is the vacuum permittivity. Note that for a rigid droplet with a large aspect ratio in a large shear gradient, the tilt angle will be large, causing the droplet to gain polarization energy. It is therefore reasonable to expect the droplet to attempt to minimize its total energy by reducing its size, even at the expense of increasing its surface energy per unit volume. Having determined the tilt angle as a function of g b 2 /c 2 , we can now nd the size and aspect ratio of a droplet that minimize its total energy. This is done by balancing the depolarization energy which will be a function of n and , both small parameters against the surface tension of a droplet. The depolarization energy of a spheroid is 30

Fd

1 VE 2 1 8

1 1

1 n ,

22

where only the lowest-order terms in and n have been kept. This energy is minimized when 0 and n 0, so this term favors long, thin columns aligned with the eld. On the other hand, the surface energy term

768

MARTIN, ODINEK, HALSEY, AND KAMIEN

57

Fs

s 4

1 E 2S

23

1 /4

/8

favors large spheroidal droplets. Here S 4 (abc) 2/3 is the surface area of the spheroid and we expect r d 13,31 . s Minimizing the total energy F d F s gives g Mn2/3, b Mn 2/3, and c Mn 1. The droplet width is b c 2/3, as in the theory of the quiescent uid 6 . Also, since Mn/g Mn1/3, Eq. 4 indicates that Mn1/3 for M 1, which conrms that is small. Thus, as the shear rate increases or the electric eld decreases the droplet length c and its aspect ratio c/b Mn 1/3 decrease while the tilt angle increases. The dependence of the droplet orientation angle has a direct bearing on the uid viscosity. The eld enhancement of the viscosity is conveniently quantied by the dimensionless eld-specic viscosity F ( )/ 0 , where is the solution viscosity at innite Mn. To determine the droplet contribution to the shear part of the stress tensor we compute the hydrodynamic torque per unit volume in the uid. If the volume fraction of droplets is , then this is h 4 0 /g, which yields
F

where (3). To obtain some idea of the magnitude of the local-eld correction, the enhancement, over the dipole approximation, of the breaking strength of a chain in tension aligned along the eld direction is (3)(3 1 2 )/2. For 1 1 the enhancement is 2 this factor is 0.71, but for 7.54. The uid exerts a Stokes friction on each of the spheres and thus a hydrodynamic torque on the chain. Balancing this hydrodynamic torque against the electrostatic torque, obtained by summing the tangential component of the dipolar force along the chain, gives 3 (3)tan 8 MnN 2 for the tilt angle of a chain of 2N 1 spheres. Balancing the hydrodynamically induced tension at the chain center against the radial component of the dipolar force gives the mechanical stability constraint 3 3
2 1 cos 2

sin cos

16 MnN 2 .

27

Mn

2/3

24

for the eld-induced contribution to the shear-thinning viscosity. This calculation applies only to the intermediate Mason number regime where the Mason number is small enough for chains to form, yet not so small that these chains span the electrode gap.
C. Chain model

The chain angle thus increases with chain length. The longest stable chain will have a critical chain angle that depends on the dielectric contrast through tan 3 2
1 3 2 2

2 1 3 1

/4 /2

28

We have recently developed a simple chain model 17 of electrorheology that is based on a balance of electrostatic and hydrodynamic forces. This athermal model was originally based on the interaction between the induced dipole moments of dielectric spheres in solution. The dipole moment was originally computed for a single sphere in a liquid continuum, but we have modied this model to account for local-eld effects by self-consistently computing the dipole moment on an enchained sphere. The presence of vicinal spheres then alters the local eld and this has a substantial effect on the agreement of the model with experiment, as we shall see. In the chain model the particle structures are presumed to be chains of spheres that interact through dipolar forces with their neighbors. The self-consistent electrostatically induced attractive dipolar force is 17 F sc b 3 3
1 cos 2 2

Note that the singularity is outside the physical range of . The chain angle is also closely approximated by the linear relation sin c 2/5(1 3 /10). As increases over its 1 maximum physical range of 2 to 1, the critical chain angle increases from 31.3 to 59.3. A stable chain of maximum length N Mn 1/2 will be oriented at exactly the critical angle c . By contrast, the droplet model gives a ellipsoid length L Mn 1/3 and an orientation angle increasing as Mn1/3. The data clearly show a dependence of the chain angle on the cube root of Mn and so support the droplet model. However, electrode friction can affect the simple prediction of the chain model to give a linear increase of the critical angle with Mn.
Rheology

3 sin

25

where b 3 a 2 0 /8 Mn, (3) 1.202 is the Riemann zeta function, and r and are unit vectors parallel and perpendicular to the line of centers between the spheres. The remaining constants have to do with the local-eld effects and are given by 1
1 2

The viscosity can be computed in a straightforward fashion. The electrostatic torque on a single-particle pair is 2aF e, 2a (3) 3 b sin2 . There are 2N such pairs in a chain so the electrostatic torque per chain is e 3 a 2 0 c 2 E 2 L (3) 3 sin2 . In terms of the volume 2 0 fraction of spheres the stress in the sample is thus 9 2 2 E 0 (3) 3 sin2 c . Using our result for the criti8 0 c cal angle gives the shear eld specic viscosity
F

/8
2

c Mn

29

/4 1

/8

2,

/4

2,

26

1 3 4 where to a good approximation c 5 ( 2 ) 5/2 (3)(1 10 3 1 3 2 ) over the physical range of . 20 6 1 has been obtained in many exThe scaling periments, including those conducted on our silica uid at

57

STRUCTURE AND DYNAMICS OF . . .

769

FIG. 20. A typical scattering image contains a pair of scattering lobes and the strain clock. These scattering lobes are tilted in the direction of uid vorticity by some angle relative to the orientation of the scattering lobes obtained for the quiescent uid, which are orthogonal to the electric-eld vector.

high elds 16 . However, for the silica uid at low elds we 2/3, a result consistent with the elliptical nd droplet model 16 .
V. OSCILLATORY SHEAR

We have reported preliminary measurements of ER uids in oscillatory shear ow that demonstrate that the chain dynamics is highly nonlinear. The nonlinear dynamics was reasonably well described by a simple kinetic chain model, in which the nonlinear response is caused by large variations in the chain length during an oscillation. These chain length variations are caused by aggregation and fragmentation processes that occur during each cycle. At the time these results were reported we did not understand why the simple model we presented did not agree more quantitatively with experiment, nor did we have direct experimental evidence for the chain length variations assumed to drive the nonlinear response. In the time hence we have thoroughly examined the chain model, including such effects as local-eld corrections, multipolar interactions, and hydrodynamic screening. Experimentally, we have now determined the chain size variations that occur during each cycle. We will rst reexamine the kinetic chain model, showing how it can be modied to take into account these effects and then we will reanalyze previously reported data and present data on the uctuations in chain size during a shear cycle. We have previously generalized the droplet model to oscillatory shear at small strain amplitudes, where aggregation and fragmentation effects that occur during a cycle can be ignored. This model gives a sinusoidal response, but the single characteristic relaxation time of the system depends on the strain amplitude and shear frequency.
A. Data analysis

tering lobes are tilted by an angle relative to the orientation of the scattering lobes obtained for the quiescent uid. The angle was then obtained by one of two methods: locating the intensity maximum max or nding the median m , dened as that angle that divides the integrated scattered intenm sity into equal halves, so 90 I( )d I( )d . The lat90 m ter method is used when the strongly nonlinear dynamics is characterized by twin peaks. Under these circumstances simply locating the maximum peak causes discontinuities in the data.
B. Measurements

A scattering image contains a pair of scattering lobes and, of course, the strain phase clock, as in Fig. 20. These scat-

At relatively low strain amplitudes the response of the orientation angle max to the sinusoidal shear strain t) was nearly linear, demonstrating that frag0 sin(2 mentation and aggregation effects may be not be important in perturbative ows where 0 1. This quasilinearity is exemplied in Fig. 21, where Lissajous plots of tan( max) against are shown to be nearly elliptical. At the low shear frequency the chain orientation leads the strain by 57. Since a 90 phase shift would put the droplets in phase with the strain rate, we conclude that the polarization coupling to the electric eld, which tends to align the chains, dominates the hydrodynamic forces, so the chains deviate from eld alignment only at the highest shear rates. Conversely, at high frequencies the shear rate t) is much larger and the droplet orientation 2 0 sin(2 is nearly in phase with the uid shear since the hydrodynamic torque dominates the electrostatic torque. Note that under this condition the afne deformation limit is nearly achieved, as shown by the dashed line. At higher strain amplitudes the motion becomes clipped as the droplets fragment and aggregate during the cycle in order to maintain good electric-eld alignment. This nonlinear motion is evidenced by the parallelogram-shaped

770

MARTIN, ODINEK, HALSEY, AND KAMIEN

57

FIG. 21. Clockwise Lissajous plots of tan( max) against are virtually elliptical lines when the strain amplitude is small. At high strain frequency hydrodynamic forces dominate and the motion is nearly in phase with the strain. At low frequency the electrostatic forces dominate and the chains deviate from eld alignment in proportion to the instantaneous strain rate, thus being almost out of phase with the strain.

FIG. 23. At low voltages a different nonlinear response of the uid can be observed, with retrograde motion of fragmenting chains. The dashed line is a theoretical curve. The amplitude of motion is sensitive to the dipolar model and the method of data reduction.

Lissajous plots in Fig. 22. The dashed lines represent theoretical curves computed from the theory developed below. At lower voltages we observe a different nonlinear uid response, as shown in a Lissajous plot in Fig. 23, and against time in Fig. 24. The droplet motion leads the strain, indicating that electrostatic interactions dominate. This may seem surprising, given that the eld is small, but we have also reduced the strain frequency to 0.075 Hz. Another way to appreciate the nonlinear behavior of this system is the noncircular Lissajous plot of the chain angular

velocity versus the chain angle Fig. 25 . Again, the theoretical prediction is shown for comparison. Starting at maximum positive strain, the droplet half cycle can be described as follows. As the strain reverses, the droplets corotate with the uid and tilt to a maximum angle at roughly half the maximum strain on the return stroke, whereupon they fragment and undergo retrograde motion to realign with the electric eld. If inertial effects are neglected then the hydrodynamic torque equals the electrostatic torque, so at zero tilt angle the droplets corotate with the uid. Because the hydrodynamic torque is thus proportional to the droplet tilt angle, the large area within the Lissajous loop indicates that this nonlinear response is dissipative.

FIG. 22. Lissajous plots at strain amplitudes of 0 0.5, 1.6, and 3.2 have parallelogram shapes that indicate a clipping of the angular motion as chains fragment and align with the eld at high strains. The theoretical curves dashed lines are computed in the instantaneous equilibrium limit where k is large.

FIG. 24. The dogbone nonlinearity shown in Fig. 23 is plotted against time. The strain data are t to a sinusoid solid line and the chain orientation data are connected by a dashed line. The retrograde motion is quite evident in this representation.

57

STRUCTURE AND DYNAMICS OF . . .

771

FIG. 26. The coordinate system used in the kinetic chain model. The number of spheres in the chain is 2N 1, v k is the velocity of the kth bead, and v f is the uid velocity.
N

F h,

k 1

F k cos , sin
0a 2

6
FIG. 25. The nonlinear response is shown as a Lissajous plot of chain angle versus angular velocity. The solid line is the theoretical prediction of the chain model and is plotted against the top and right axes; the data are plotted against the bottom and left axes.

cos2

N2 k2 .

This tangential force is a maximum at the chain center, where in low-Reynolds-number ow it is balanced solely by the tangential component Eq. 25 of the dipole-dipole interaction force F sc, 3 3 a2
0 /8

Finally, we observed that the scattering lobes brighten considerably as they swing back through zero angle, indicating droplet aggregation. Likewise, as the lobes swing to their maximum tilt they diminish in intensity, indicating droplet fragmentation. All of these observations of the droplet motion point the way to the simple model of the dynamics that we shall now present.
C. Kinetic chain model

Mn

3 sin2

Balancing the tangential hydrodynamic and electrostatic forces at the chain center gives the damped oscillator equation
d sin2

cos2

where

3 3 . 16 MnN 2

30

The salient features of experimental results we have shown can be understood in terms of a kinetic model of the dynamics of volatile chains. We have presented this model elsewhere 17 for the case of xed induced dipolar interactions: Here we modify the basic model to account for localeld corrections. Local-eld corrections have the important consequence of changing the critical chain angle from a xed value of 39.2 to a range from 31.3 to 59.3, depending on the dielectric contrast factor . We consider a linear chain of 2N 1 spheres of radius a labeled from N to N in a coordinate system (x,z), the origin of which is centered on the zeroth sphere Fig. 26 . The z axis is in the direction of the electric eld and the x axis is in the direction of uid vorticity. The chain makes an angle to the x axis, so the position of the kth bead is 2ak sin ,2ak cos . The uid velocity is given by v (z) zx , where x is a unit vector, and the velocity of the kth bead is v k 2ak (cos x sin z) for a chain rotating at angular velocity . The shearing uid exerts a hydrodynamic force Fk 6 0 v (z) v k on the kth bead, where 0 is the liquid viscosity. This hydrodynamic force can be decomposed into a tangential component that causes chain rotation and a component that causes tension or compression. The tangential component of the hydrodynamically induced force between the kth and (k 1)th spheres is

The characteristic oscillator frequency d depends strongly on chain size. Physically acceptable values N must correspond to mechanically stable chains or fragmentation will occur. The radial component directed along the chain axis of the hydrodynamic force is
N

F h,r

k 1

F k sin ,cos

0a

sin 2

N2 k2 .

This force, which again is a maximum at the chain center, puts the chain in tension when 0, since the chain is tilted in the direction of shear, and in compression when 0. For the chain to be stable to fracture this hydrodynamically induced force must be smaller than the radial component of the electrostatic interaction F sc,r 3 3 a2
0 /8

Mn 3

2 1 cos

The maximum stable chain number is determined by balancing these forces at the chain center 3 3 8 Mn ,
1 cos 2 2

N max

sin2

0 31 0.

The maximum stable chain length is extremely dependent on chain orientation and strain rate, especially when driven by

772

MARTIN, ODINEK, HALSEY, AND KAMIEN

57

oscillatory shear. The maximum stable chain length diverges when the chain is aligned with the eld, when the instantaneous strain rate is zero, and when the chain is under compression. If a chain is far from its maximum stable size then its size will adjust by aggregation or fragmentation. We will describe the kinetics of aggregation and fragmentation by the phenomenological formula dN t dt k N t 1 N t
2

N2 t max

32

where because induced dipolar forces drive aggregation the rate constant it is useful to write k k 0 (3) 3 0 c 2 E 2 /8 0 , where k 0 is a concentration0 dependent constant with no implicit eld or viscosity dependence. The reasons for this denition of k 0 will become obvious in the following. This kinetic equation gives very different time dependences for aggregation and fragmentation. When the chain is much smaller than its maximum stable length, slow, powerlaw aggregation will occur with N(t) N(0) 2 2kt, in agreement with the root time prediction of See and Doi 32 , which they developed for the quiescent uid. If the chain is much larger than its stable length then fragmentation will occur exponentially quickly according to N(t) 2 N(0)e kt/N max. Note that the fragmentation rate k/N 2 is max proportional to the strain rate and is independent of the electric eld and viscosity. Of course, when the chain is at its maximum length no aggregation or fragmentation occurs since dN(t)/dt 0. Thus the phenomenological rate equation gives physically reasonable behavior while avoiding the complexities of the Smoluchowski equation. Equations 30 32 now comprise a set of coupled nonlinear equations that can be solved to model the dynamics of chains in shear ow. However, at this point there are four independent parameters in the system: the Mason number Mn, the strain frequency and amplitude 0 and the rate constant prefactor k 0 . A considerable simplication occurs by recognizing that solutions to the kinetic equation are of the form N(t) (3) 3 /16 Mn 1/2n( t). If all functions are expressed in terms of the dimensionless time s t this leads to the reduced damped nonlinear oscillator equations 1 sin2 n2 cos2 ,
2

FIG. 27. Lissajous plots change unexpectedly as the rate constant k is varied. For k 0 the oscillator reorientation rate d is a simple constant and the remaining nonlinearities are small enough that a nearly elliptical response is observed. For large k Lissajous plots approach a parallelogram, whereas for intermediate-k values retrograde motion is observed. 1. Electrorheology

Before we discuss the behavior of these equations we would like to discuss the predicted rheology. In terms of the volume fraction of spheres the eld-induced contribution to the uid stress is 6
0N 2

cos2

9 8

0 c

E2 0

3 sin2

. 34

From the right-hand side of this equation it is clear that the uid stress increases with the chain orientation angle. When 0, the chain contribution to the uid stress is zero and the chain comoves with the uid so cos2 . Because Eq. 33 shows that the dynamics of chain orientation is independent of electric eld and shear frequency, we conclude from Eq. 34 that the stress scales purely as the square of the electric eld.
2. Numerical results

k0 1 n
2

n2 n2 max 0 0.

, 33

n max ,

2 3

1 cos

3 sin2

The reduction to a three-parameter ( 0 ,k 0 , ) model is a result of the particular form of the rate equation we have chosen. The strain amplitude is xed in the experiment and can be computed, so this is really a single free parameter model. Finally, it is interesting to note that the chain orientation dynamics (t) is independent of Mason number, although the chain length dynamics N(t) is not.

We can now compare the predictions of this model with our experimental results. For this purpose we have used the dielectric constant of 13.5 for 4-methylcyclohexanol and 4 for silica, giving a negative value of . For small strains the response is sinusoidal, as expected, but for large strains Figs. 27 and 28 Lissajous plots are nonelliptical and dependent on the rate constant k. For k 0 the oscillator reorientation rate a is a simple constant and the remaining nonlinearities in Eq. 30 are small enough that a nearly elliptical response is observed. As k increases, the chains fragment and aggregate to achieve mechanical stability while trying to maximize their length and Lissajous plots approach a parallelogram. The instantaneous response limit large k of the local-eld-corrected model is in good agreement with the data in Fig. 22. This agreement is much better than the bare model because of the reduced orientation angle. Finally, for

57

STRUCTURE AND DYNAMICS OF . . .

773

FIG. 28. The orientational dynamics shown in Fig. 27 are plotted as functions of the strain rate. As the rate constant k increases, the nonlinear dynamics becomes more pronounced.

intermediate values of k an interesting crossover is observed wherein humps appear in the Lissajous plots. This hump is commonly observed in our data at low elds, as illustrated in Figs. 23 and 24. The nonlinear behavior of this system is also shown as a Lissajous plot of the chain angular velocity versus the chain angle Fig. 27 . Again, the detailed shape of the theory is somewhat different from the data, but the overall agreement is quite good. The chain orientation dynamics is a somewhat indirect test of the kinetic chain model since it is conceivable that one could construct another set of equations that gives the same overall behavior. However, the nonlinear behavior of this model is ultimately driven by the fragmentation and aggregation phenomena that occur during each shear cycle. The dynamics of fragmentation, which occurs at twice the oscillation frequency, is shown in Fig. 29. Chain aggregation is very pronounced just as the electrode is about to return to zero angle, with chains reaching a maximum size just after the maximum strain. During this time interval the chain is more or less comoving with the ow. Fragmentation occurs much before the electrode returns to zero angle because the chain is nearly 90 ahead of the strain and thus is already at its maximum tilt angle in the opposite direction. The powerlaw aggregation and exponential fragmentation leads to an obvious asymmetry in the chain size peaks. Large aggregates cause intense light scattering, so to measure the chain size we determined the wedge integrated maximum intensity I max as opposed to the peak position max for each scattering image. The peak intensity is plotted against time in Fig. 30 and against chain orientation in Fig. 31. There is a very close correspondence between these data and the prediction Fig. 29 of the kinetic chain model. The peak position relative to the applied strain is about right and even the peak asymmetry, due to slow aggregation and fast fragmentation, can be seen. Moreover, this pattern was observed in all of the data sets, except those taken at a strain amplitude of 0.25, despite the tremendous variations in Lissajous plots of the orientational dynamics.

FIG. 29. The chain length varies considerably during the strain cycle due to fragmentation at large chain angles and aggregation at small angles. The asymmetry of the peaks is due to the power-law aggregation kinetics being slower than the exponential fragmentation. The sinusoid is the strain. D. Phase bifurcation

The oscillatory shear measurements we have reported thus far have been at low to moderate elds. When the voltage is turned up to greater than about 1.0 kV we found that the scattering lobes split into two pairs. Direct observation indicated that the phase and amplitude of motion of the two pairs differed. After digitizing the scattering data we found that, due to the nite lobe width, the pairs of lobes could be distinguished from each other only when the strain amplitude

FIG. 30. When the maximum intensity solid line, plotted against the rst left axis is plotted against time, the resulting curve is very similar to the chain length uctuations of Fig. 29. The strain is the sinusoid plotted against the right axis and the chain orientation is the small dashed line plotted against the second left axis. These data, taken at a strain amplitude of 3.2, are representative of data taken at strain amplitudes down to 0.5. Intensity uctuations are small when the strain amplitude is 0.25.

774

MARTIN, ODINEK, HALSEY, AND KAMIEN

57

FIG. 31. The data of Fig. 30 are used to make a Lissajous plot of chain size versus orientation. The inset is a typical computed Lissajous plot of chain length versus orientation for the kinetic chain model. The asymmetry in the data is exaggerated in this representation.

FIG. 33. The difference between the tilt angle of scattering lobes that are in phase and out of phase with the applied strain reaches a maximum when the strain is at half maximum on the return stroke, i.e., 150 and 330.

is roughly half maximum on the return stroke i.e., at strain phase angles of about 150 and 330, where 0 is dened as the zero strain amplitude on the positive outgoing stroke . The resolved peaks are shown in Fig. 32. We attribute this phase bifurcation to the onset of a shear slip instability 33 , with one population of chains growing out from each elec-

trode and a second population of free droplets in a central high shear rate slip zone. If a shear slip instability is indeed the cause of phase bifurcation, then the free droplets in the shear slip zone should be subject to a large hydrodynamic torque and should thus be nearly in phase with the strain, whereas the bound chains should feel little hydrodynamic torque and should be out of phase with the strain. Because the scattering lobes cannot be resolved throughout most of the strain cycle, it is not possible for us to directly determine the phase of each component. Instead, we generated the response functions in(t) for in phase and out(t) for out of phase oscillators and plotted the difference (t) out(t) in(t) between these functions in Fig. 33. It is apparent that the scattering lobes will be resolved when , where is the width of the lobes. Thus, when is a maximum we have the best chance of resolving the peaks, and this does indeed occur at the half maximum strain on the return stroke. This is strong evidence in support of two structural components free droplets and bound chains that arise due to the formation of a shear slip zone.
VI. CONCLUSIONS

FIG. 32. As the voltage is increased the scattering pattern bifurcates into two pairs of lobes that oscillate at about 90 out of phase with one another. When the applied strain is at half maximum on the return stroke i.e., at 150 and 330 the two lobes can easily be resolved, as shown here. We attribute this phase bifurcation to the onset of a shear slip instability, with free droplets in the shear zone and bound chains attached to the electrodes. Direct microscopy studies on this uid support this conclusion.

We have presented light-scattering studies of an electrorheological uid in the quiescent state, steady shear, and oscillatory shear. Studies of the coarsening of the quiescent uid after a eld quench show that the growth of structure in many ways mimics the spinodal decomposition of a binary uid. The analogy with spinodal decomposition is strong; a peak appears in the structure factor, the high-q shoulder of which conforms to Porods law of scattering from sharp interfaces while the low-q shoulder increases as q 2 , and the domain size increases as a power of time. However, close examination reveals that the domain growth exponent is smaller than the 1/3 value predicted for spinodal decomposition in a system with a conserved order parameter. We nd

57

STRUCTURE AND DYNAMICS OF . . .

775

the kinetics data are better described by a thermal model of hierarchical clustering of columns into successively larger columns. The interaction between columns is presumed to arise from one-dimensional charge-density uctuations in the columns that persist long enough to allow columns to collide. This model also gives a good account of the increase in the growth kinetics with applied voltage. Studies of an ER uid in steady shear show that the structure reaches a steady state wherein droplets are rotated in the direction of uid vorticity at some angle relative to the applied electric eld. This angle is found to increase as the cube root of the shear rate, in agreement with a model we originally proposed for the shear thinning viscosity. In this model the equation of motion of elliptical droplets is found by balancing the hydrodynamic and electrostatic torques. The droplets reach their free-energy minimum by fragmenting to align with the electric eld until exposing more surface nally becomes too energetically costly. Our studies of ER uids in oscillatory shear demonstrate that the chain dynamics, and thus the electrorheology, is non-

linear. We have described a simple kinetic chain model of the dynamics that describes the approach of a chain to its maximum stable size by a kinetic equation. Much of our experimental data can be described by taking the instantaneous approach to stability; however, at low elds strong nonlinearities suggest that the approach to stability is slow compared to the shear period. This model is then used to compute the nonlinear rheology of an ER uid and it is concluded that light scattering is an indirect probe of stress. At high voltages we observe a phase bifurcation in the scattering pattern that we attribute to the onset of a shear slip zone. Free droplets in the shear slip zone oscillate out of phase with bound chains attached to the electrodes.

ACKNOWLEDGMENTS

The work of J.E.M. and J.O. was performed at Sandia National Laboratories and was supported by the U.S. Department of Energy under Contract No. DE-AC04-94AL85000.

1 H. Block and J. P. Kelly, J. Phys. D 21, 1661 1988 . 2 A. P. Gast and C. F. Zukoski, Adv. Colloid Interface Sci. 30, 153 1989 . 3 R. A. Anderson, in Proceedings of the International Conference on Electrorheological Fluids: Mechanism, Properties, Structure, Technology and Applications, edited by R. Tao World Scientic, Singapore, 1992 . 4 R. A. Anderson, Langmuir 10, 2917 1994 . 5 L. C. Davis, Appl. Phys. Lett. 60, 319 1992 . 6 J. E. Martin, J. Odinek, and T. C. Halsey, Phys. Rev. Lett. 69, 1524 1992 . 7 J. E. Martin, J. Odinek, and T. C. Halsey, Phys. Rev. E 50, 3263 1994 . 8 J. E. Martin and J. Odinek, Phys. Rev. Lett. 75, 2827 1995 . 9 J. M. Ginder, Phys. Rev. E 47, 3418 1993 . 10 K. L. Smith and G. Fuller, J. Colloid Interface Sci. 155, 183 1993 . 11 D. Adolf and T. Garino, Langmuir 11, 307 1995 ; D. Adolf, T. Garino, and B. Hance, ibid. 11, 313 1995 . 12 T. C. Halsey and W. Toor, J. Stat. Phys. 61, 1257 1990 . 13 T. C. Halsey and W. Toor, Phys. Rev. Lett. 65, 2820 1990 . 14 R. Tao and J. M. Sun, Phys. Rev. Lett. 67, 398 1991 . 15 T. Chen, R. N. Ritter, and R. Tao, Phys. Rev. Lett. 68, 2555 1992 . 16 T. C. Halsey, J. E. Martin, and D. Adolf, Phys. Rev. Lett. 68, 1519 1992 . 17 J. E. Martin and R. A. Anderson, J. Chem. Phys. 104, 4814 1996 .

18 J. E. Martin, D. Adolf, and T. C. Halsey, J. Colloid Interface Sci. 167, 437 1994 . 19 T. C. B. McLeish, T. Jordan, and M. T. Shaw, J. Rheol. 35, 427 1991 . 20 J. A. Powell, J. Rheol. 39, 1075 1995 . 21 A. P. Philipse and A. Vrij, J. Colloid Interface Sci. 128, 121 1987 . 22 L. Landau and E. M. Lifshitz, Statistical Physics, 2nd ed. Pergamon, Oxford, 1984 . 23 G. K. Batchelor, Introduction to Fluid Dynamics Cambridge University Press, New York, 1967 . 24 W. B. Russel, D. A. Saville, and W. R. Schowalter, Colloidal Dispersions Cambridge University Press, New York, 1989 , Chap. 2. 25 D. R. Nelson, Physica A 177, 220 1991 ; D. R. Nelson and H. S. Seung, Phys. Rev. B 39, 9153 1989 . 26 R. D. Kamien and D. R. Nelson, J. Stat. Phys. 71, 23 1993 . 27 J. D. Gunton and M. Droz, Introduction to the Theory of Metastable and Unstable States Springer-Verlag, Berlin, 1983 . 28 J. S. Langer, M. Bar-on, and H. D. Miller, Phys. Rev. A 11, 1417 1975 . 29 G. B. Jeffry, Proc. R. Soc. London, Ser. A 102, 161 1922 . 30 L. Landau and E. M. Lifshitz, Electrodynamics of Continuous Media, 2nd ed. Pergamon, Oxford, 1984 , Chaps. 1 and 2. 31 W. Toor and T. C. Halsey, Phys. Rev. A 45, 8617 1992 . 32 H. See and M. Doi, J. Phys. Soc. Jpn. 60, 2778 1991 . 33 D. F. Klingenberg and C. F. Zukoski, Langmuir 6, 15 1990 .