

We studied suspensions of silica spheres with diameter (Catalog # 8150, Duke Scientific, Palo Alto, CA) sedimented in deionized water into monolayers above the bottom wall of a 200 thick glass sample cell. The balance of electrostatic and gravitational forces maintained the spheres in a twodimensional layer at . As in previous studies [23,24], we extracted the sphere's equilibrium pair potential from their pair correlation function . Unfortunately, cannot be related directly to using Eq. (10) because we cannot be sure a priori that the monolayer is sufficiently dilute. Finite concentration introduces manybody correlations into and, in general,
(48) 
The average number of spheres separated from a given sphere by is , where is the twodimensional particle concentration. Given spheres in field of view , the total number of such pairs should yield the pair correlation function through , with . In a limited field of view, however, spheres near the edge have fewer neighbors to sample at range than those near the middle. Applying periodic boundary conditions would introduce spurious correlations, while limiting the calculation to a central region of the field of view would drastically degrade statistics. Instead, we count the pairs in a recorded image, normalize by the number of particles further than from any edge, and sum over angles to calculate .
The limited field of view contains very few particles when is small, so that typically is not much larger than 1 for a resolution . Limiting relative errors in to would require . Measuring to this accuracy therefore requires averaging uncorrelated snapshots of the suspension's structure. The time required for correlations to decay is roughly ten percent of the interval a particle needs to diffuse the mean interparticle separation. Thus the time required to adequately sample is .
On this basis, we suspect that some previously reported equilibrium measurements [23] of likecharge attractions may have been undersampled by as much as a factor of 20 in time. Excess correlations observed under these conditions might be interpreted, therefore, as transients, rather than as evidence for equilibrium attractions. This raises concern because the attractions reported in these studies are comparable to the quoted energy resolutions and thus could be particularly sensitive to incomplete averaging.
Because of difficulties in controlling concentration, temperature, and ionic strength constant over long periods, statistical accuracy strongly favors larger particle concentrations. On the other hand, manybody contributions obscure pair interactions for larger concentrations. The areal densities between and 0.1 chosen for the present study represent a compromise between statistical and interpretive accuracy. Adequate sampling for our lowest density monolayer required half and hour of sampling.
Even if is calculated appropriately, spurious correlations could be introduced by lateral variations in the glass wall's interaction with the spheres. Preferred regions would collect spheres, mimicking an intersphere attraction. We compared twodimensional histograms of recorded particle positions with analogous histograms for uniformly distributed random data sets with the same number of particles. Differences in the first two moments of these histograms vanish with increasing delay time between consecutive snapshots, suggesting that each particle's position becomes uncorrelated over time as expected. Thus the substrate potential appears to be featureless on the length and timescales of our experiment, to within our resolution. Previous studies [23,24] do not appear to have addressed this possible source of error.
Our experimental results for and at different concentrations appear in Fig. 9. The curves indicate a repulsive core interaction causing particle depletion extending to about . Beyond this, they reveal a preferred nearestneighbor separation between two and three diameters, as well as the onset of the oscillatory correlations typical of a structured fluid. The depth of the minimum in the potential of mean force depends on and so reflects at least some manybody contributions.
Provided that is free from experimental artifacts, reliable approximations for can be obtained from the OrnsteinZernike integral equation with appropriate closure relations [65]. Good results for ``soft'' potentials typically are achieved with the hypernetted chain (HNC) approximation, whereas the PercusYevick (PY) approximation is known to be a better choice for hard spheres. The pair interaction potential can be evaluated numerically as
(49) 
(50) 
Figure 10 shows the pair potential obtained for together with a fit of the HNC approximation to Eq. (9) for , , and an arbitrary additive offset. Having obtained comparable results in both the HNC and PY approximations lends confidence in the accuracy of . Corresponding pair potentials extracted from data at and 0.083 are essentially indistinguishable on this scale, and the extracted coupling constants are consistent with those obtained at lower areal density. The absence of minima in the pair potential confirms that oscillations in the potential of mean force (Fig. 9) resulted from crowding, while the underlying interaction is purely repulsive. The similarity of the results at different areal densities suggests that the interaction responsible for the monolayer's liquid structure indeed can be resolved into an additive pair potential. Values extracted for the charge number and screening length are consistent with surface charging due to silanol dissociation in a solution with an ionic strength around M.
Monte Carlo simulations of particles interacting with comparable pair potentials confirm our energy resolution to be about . This uncertainty is too large to resolve van der Waals attraction or image charge repulsions [49,51]. Longrange attractions of the previously reported strength [2,23,24,27], however, would have been resolved.