ConfinementInduced Colloidal Attractions in Equilibrium
Abstract.
The PoissonBoltzmann theory for colloidal electrostatic interactions predicts that charged colloidal spheres dispersed in water should repel each other, even when confined by charged surfaces. Direct measurements on highly charged polystyrene spheres, however, reveal strong, longranged confinementinduced attractions that have yet to be explained. We demonstrate that anomalous attractions also characterize the equilibrium pair potential for more weakly charged colloidal silica spheres sedimented into a monolayer above a glass surface. This observation substantially expands the range of conditions for which meanfield theory incorrectly predicts the sign of macroions' interactions, and provides new insights into how confinement induces longranged likecharge attractions.
Interactions among charged colloidal particles in water are incompletely understood despite more than a century of study. For example, PoissonBoltzmann theory predicts that similarly charged colloidal spheres should repel each other with a screenedCoulomb potential (1). This reasonable prediction was contradicted by Kepler and Fraden's experimental observation that highly charged microspheres attract each other, at least under some conditions (2). Optical tweezer measurements (3); (4) subsequently demonstrated that anomalous attractions only appear when charged spheres are confined to a plane by other charged surfaces, and not otherwise. Even so, confinementinduced likecharge attractions also are inconsistent with meanfield theory (5). How confinement inverts the sign of charged colloids' longranged interactions has proved a durable mystery.
This problem's intractability reflects the complexity of the macroionic environment. Under equilibrium conditions, chargestabilized colloidal particles' longrange interactions result from a combination of direct Coulomb repulsions and indirect interactions mediated by microscopic ions in solution. The spheres' effective pair potential, , is an average over the simple ions' degrees of freedom. The PoissonBoltzmann formalism performs this average in the meanfield approximation and so cannot account for fluctuations or correlations in the simple ions' distributions. Thermodynamically selfconsistent liquid structure calculations demonstrate that particleion and ionion correlations indeed can mediate longranged attractions (6) consistent with metastable phase separation in bulk suspensions (7). This mechanism's relevance to pairs of spheres confined by charged walls remains to be determined, however.
The apparent subtlety of confinementinduced likecharge colloidal attractions has inspired a critical reassessment of the experimental evidence. For example, all previous observations of confinementinduced attractions (2); (4); (8); (9) were performed on polystyrene (PS) microspheres with highly acidic surface groups. The only related study on confined silica spheres found no deviations from meanfield theory's predictions (10). Anomalous interactions therefore might be peculiar to highly charged spheres in general, or to PS in particular, in which case the effect would be far less generally important than originally supposed. Concern also has been raised that hydrodynamic coupling (11); (12) or other experimental artifacts (10) might explain the observed attractions, particularly because error estimates were not available for some of the measurements (2); (8).
This Letter describes digital video microscopy measurements of the equilibrium pair potentials for charged colloidal spheres under varying degrees of confinement. These measurements reveal strong and longranged likecharge colloidal attractions not only between PS spheres confined to the midplane between parallel glass walls, but also between more weakly charged silica colloid sedimented into a monolayer above a glass wall. Even though the sedimented spheres' dynamics should be influenced principally by the nearby wall, we find that intersphere attractions arise only in the presence of a second parallel wall, and that its influence is remarkably longranged. These effects cannot be ascribed to kinematic artifacts (11); (12) because the measurements are performed on dispersions in equilibrium. We also demonstrate that sample inhomogeneities have far too small an effect to account for our observations. Instead, our results are consistent with a pairwise confinementinduced attraction between likecharged colloidal spheres.
Our experimental system, shown schematically in Fig. 1, is prepared according to the methods of Ref. (3). A dilute deionized colloidal suspension is confined to a monolayer between the parallel surfaces of a glass microscope slide and a #1 glass coverslip. The edges of the glass surfaces are sealed with a highpurity UVcured adhesive (Norland Products Type 88) to form a closed sample volume, with access provided by glass tubes bonded to holes drilled through the upper glass slide. The separation, , between surfaces is measured by focusing a laser beam onto the glasswater interfaces using a piezoelectric controller to adjust the objective lens' height. Mixedbed ion exchange resin packed into the tubes helps to maintain total ionic strengths around in the visible sample area. The tubes are sealed after a sample is introduced, and the suspension is allowed to equilibrate at ambient temperatures on the stage of a Zeiss S100TV Axiovert microscope. The particles in an area are imaged with a 63 objective lens and a video eyepiece onto a lownoise monochrome CCD camera. A detail from a typical video frame appears in Fig. 1. The particles' motions are video taped before being digitized for analysis.
We use highresolution particle tracking algorithms (13) to locate the centroid of each particle in the plane to within 30 nd to link the particles' positions in a sequence of video images into timeresolved trajectories at 1/30 ntervals. A typical trajectory appears in Fig. 1. The distribution
(1) 
of particles in the field of view at time provides detailed information on the particles' dynamics under the combined influences of random thermal forces and their mutual interactions.
To extract from , we adopt the approach pioneered by Kepler and Fraden (2) and Vondermassen et al. (14) by first calculating the pair correlation function
(2) 
where is the areal density in a field of view containing particles, and the angle brackets denote averages over both angles and time. A typical example appears in Fig. 2(a).
The pair correlation function is related to the potential of mean force, , through the Boltzmann distribution,
(3) 
This reduces to the true pair potential in the dilute limit: . Because measurements must be made at finite concentrations, however, manybody interactions may introduce additional structure into the pair correlation function. In particular, layering in colloidal fluids causes oscillatory correlations even when the underlying pair potential is purely repulsive. The resulting minima in are not evidence of pairwise attractions, but simply describe manybody structural correlations.
To distinguish likecharge attractions from crowding, CarbajalTinoco et al. (8) introduced the idea of analyzing with the OrnsteinZernike (OZ) integral equation (15), which describes the evolution of manybody correlations from a hierarchy of pairwise interactions. Truncating the hierarchy results in approximations that may be inverted to obtain expressions for . Among these, the hypernetted chain (HNC) is found to be accurate for “soft” potentials while the PercusYevick (PY) is more accurate for shortranged interactions. The pair potential can be evaluated in these approximations as (16)
(4) 
where the convolution integral
(5) 
is solved iteratively, starting with . Evaluating directly rather than with numerical Fourier transforms reduces sensitivity to errors in (10).
Figure 2(a) shows a typical pair potential obtained with Eqs. (1)  (5) for a suspension of diameter polystyrene sulfate spheres (Duke Scientific Lot 22998, density ) at areal density confined to the midplane () between charged glass surfaces (17) separated by . These spheres' acidic surface groups dissociate almost completely in water, endowing them with a surface charge density of roughly 1 electron equivalent per (3); (13); (4). As in previous studies (2); (4); (8); (9), displays a minimum deep at a range of .
Excellent agreement between the HNC and PY approximations implies that the OZ formalism is reliable at the experimental areal density. Differences between the two constitute one contribution to the estimated error in .
The diamonds in Fig. 2(a) show HNC results for a similar suspension at one third the areal density, . Their agreement with the higherdensity data demonstrates the pairwise additivity of in this range of areal densities, and is consistent with optical tweezer measurements that found comparable confinementinduced attractions even at vanishing areal densities. This differs from densitydependent effects observed in more concentrated suspensions (18).
The data in Fig. 2(b) were obtained for silica spheres in diameter (Duke Scientific Lot 24169). These differ from PS spheres in two respects: they carry one fifth the surface charge density (17) and are twice as dense (). Consequently, they sediment into a monolayer roughly above the lower wall, with outofplane fluctuations smaller than (10); (17). This height does not vary perceptibly with wall separation for .
Previous measurements of these spheres' interactions in sedimented monolayers at showed no sign of interparticle attractions (10). The pair potential in Fig. 2(b) at similarly is monotonically repulsive. The meanfield prediction (1),
(6) 
yields good agreement for an effective charge number on each sphere, and DebyeHückel screening length , consistent with a total concentration of of monovalent ions at the experimental temperature of . These results agree both with previous measurements on comparable systems (10) and also with the charge regulation theory for interacting silica surfaces (17).
Reducing the interwall separation qualitatively changes , introducing a minimum consistent with a longranged attraction. As for the PS data in Fig. 2(a), the confined silica spheres' interactions are independent of sphere concentration, and consistent results are obtained with both HNC and PY approximations. The example at in Fig. 2(b) displays a minimum deep at . The inset to Fig. 2(b) reveals clearly resolved minima in for plate separations as large at . Such a longranged influence, extending to more than 100 screening lengths, is not likely to result from electrostatic coupling between the spheres and the distant charged wall. Instead, it might reflect the walls' contribution to the local salt concentration (13) or their influence on the silica spheres' effective charges (17). The nonmonotonic dependence of the potential well's depth on plate separation may also reflect variations in such factors as temperature and salt concentration from run to run. Comparable results also are obtained for slightly larger silica spheres (, Bangs Laboratories Lot No. 4186) under comparable conditions.
The accuracy with which wallmediated attractions are resolved in Fig. 2 can be quantified by considering sources of error in the measurement technique. Whereas accurate methods for measuring from particles' images are well established (13), those for estimating from involve subtleties which only recently have been explored in detail (10); (19). In particular, the number, , of particles in the field of view generally is insufficient to accurately assess at small separations. Instead, statistically independent snapshots are required to sample to within at with spatial resolution (10). If were undersampled in this thermodynamic sense, transient density fluctuations could distort .
The interval between statistically independent particle configurations is set by the time a particle of diffusivity needs to diffuse the mean interparticle distance. Thus the total time required to sample scales as (10); (19). For this reason, using the densest possible suspension consistent with reliable manybody corrections is highly desirable.
We estimate and thus from the particles' trajectories, by calculating the probability for a particle to travel a distance in the th direction over time . Based on fit values for , the sampling periods needed to attain accuracy better than with spatial resolution range from 20 in our most dense PS sample to 47 n our least dense silica sample, and were exceeded in all measurements. Variations in among subsamples of the full data sets are consistent with expected equilibrium fluctuations.
Even statistically accurate measurements can yield misleading results if external forces induce correlations among the spheres that mimic attractive interactions. However, twodimensional histograms of averaged over each experiment's duration reveal that the glass surfaces establish featureless potential energy landscapes to within for . Furthermore, typical drift speeds of are far too small to mediate measurable inplane hydrodynamic coupling (12). Consistency between fit values of further confirms the samples' freedom from uniaxial forces.
In light of the preceding consistency checks, the reproducible absence of attractions in silica samples at (17) rules out explanations for attractions in more confined samples based on variations in the spheres' properties. Nevertheless, such variations contribute to errors in and . In particular, a small population (%) of dimers increases measured correlations near contact. This is exacerbated in our silica data by a small (%) population of undersize particles whose height, , and outofplane fluctuations, , are larger than average and lead to projection errors at small separations (20). Such particles are easily identified by analyzing their apparent size and brightness (13). We recalculate without these outliers, deconvolve the result with the suspensions' 2.5% dispersion in diameters (21) and subtract the statistical error to set a lower bound on near contact. Recomputing and adding the difference between HNC and PY approximations in quadrature yields the upper estimates for in Fig. 2. Applying statistical and liquid structure corrections to the unmodified yields the lower limits.
Contrary to the predictions of mean field theory, our measurements demonstrate that confinement by parallel glass walls induces longranged pairwiseadditive attractions between similarly charged colloidal spheres in equilibrium. By accounting for all identified experimental artifacts, the quantitative error estimates ensure that these anomalous interactions are indeed clearly resolved. Explanations for the attractions based on kinematic effects such as hydrodynamic coupling similarly are excluded. When viewed in this light, the ability of a distant upper surface to substantially alter the interactions among silica spheres hovering just above the lower glass wall in our experiments is particularly noteworthy. The appearance of confinementinduced attractions between comparatively weakly charged silica spheres suggests that qualitative failures of mean field theory for macroionic interactions may be more common than previously supposed.
We are grateful to Sven Behrens for extensive conversations. This work was supported principally by the MRSEC program of the NSF through Grant Number DMR9808595. Additional support was provided by the NSF through Grant Number DMR9730189.
References

(1)
W. B. Russel, D. A. Saville, and W. R. Schowalter, Colloidal Dispersions (Cambridge University Press, Cambridge, 1989).

(2)
G. M. Kepler and S. Fraden, Phys. Rev. Lett. 73, 356 (1994).

(3)
J. C. Crocker and D. G. Grier, Phys. Rev. Lett. 73, 352 (1994).

(4)
J. C. Crocker and D. G. Grier, Phys. Rev. Lett. 77, 1897 (1996a).

(5)
J. E. Sader and D. Y. C. Chan, J. Colloid Interface Sci. 213, 268 (1999); E. Trizac and J.L. Raimbault, Phys. Rev. E 60, 6530 (1999); J. C. Neu, Phys. Rev. Lett. 82, 1072 (1999); J. E. Sader and D. Y. C. Chan, Langmuir 16, 324 (2000).

(6)
M. D. CarbajalTinoco and P. GonzalezMozuelos, J. Chem. Phys. 117, 2344 (2002).

(7)
A. E. Larsen and D. G. Grier, Phys. Rev. Lett. 76, 3862 (1996); A. E. Larsen and D. G. Grier, Nature 385, 230 (1997).

(8)
M. D. CarbajalTinoco, F. CastroRomán, and J. L. ArauzLara, Phys. Rev. E 53, 3745 (1996).

(9)
H. AcuñaCampa, M. D. CarbajalTinoco, J. L. ArauzLara, and M. MedinaNoyola, Phys. Rev. Lett. 80, 5802 (1998).

(10)
S. H. Behrens and D. G. Grier, Phys. Rev. E 64, 050401 (2001a). Longranged attractions were observed between silica spheres confined by elastomer surfaces in B. Cui, B. Lin, S. Sharma and S. A. Rice, J. Chem. Phys. 116, 3119 (2002). In this case, 2 mM added salt suppresses electrostatic coupling, and the observed attraction was ascribed to wetting effects at the hydrophobic surfaces.

(11)
T. Squires and M. P. Brenner, Phys. Rev. Lett. 85, 4976 (2000); P. Attard, Cur. Opin. Colloid Interface Sci. 6, 366 (2001).

(12)
Y. O. Popov, J. Colloid Interface Sci. 252, 320 (2002).

(13)
J. C. Crocker and D. G. Grier, J. Colloid Interface Sci. 179, 298 (1996b).

(14)
K. Vondermassen, J. Bongers, A. Mueller, and H. Versmold, Langmuir 10, 1351 (1994).

(15)
D. A. McQuarrie, Statistical Mechanics (University Science Books, Mill Valley, CA, 2000).

(16)
E. M. Chan, J. Phys. C 10, 3477 (1977).

(17)
S. H. Behrens and D. G. Grier, J. Chem. Phys. 115, 6716 (2001b).

(18)
M. Brunner, C. Bechinger, W. Strepp, V. Lobaskin, and von Grunberg. H. H., Europhys. Lett. 58, 926 (2002).

(19)
R. V. Durand and C. Franck, Phys. Rev. E 61, 6922 (2000).

(20)
R. Rajagopalan and K. S. Rao, Phys. Rev. E 55, 4423 (1997).

(21)
B. D'Aguanno and R. Klein, Phys. Rev. A 46, 7652 (1992); F. Lado, J. Chem. Phys. 108, 6441 (1998).