Researchers have developed a three-dimensional stochastic tiling model that accurately predicts mean nearest-neighbor distances across computer-generated uniform random distributions of hard spheres. Quantitative spatial analysis of microstructures relies heavily on the mean nearest-neighbor distance as a foundational clustering and ordering metric, particularly in materials containing dispersed particles, inclusions, or voids. In three-dimensional random-sequential-addition patterns spanning volume fractions from 0.0001 to 0.35, the newly introduced mathematical framework estimates center-to-center separations without relying on adjustable fitting parameters. The analytical technique builds directly upon previous two-dimensional formulations by extending polyhedral space-filling geometric calculations into three dimensions, providing materials scientists with a rigorous theoretical tool for interpreting complex spatial architectures.
Understanding spatial distributions in three dimensions has become increasingly critical as advanced characterization techniques such as X-ray microtomography routinely collect volumetric datasets of structural materials. Distance distributions between microstructural features directly govern bulk physical performance, including failure strain, yield behavior, and stiffness in particle-reinforced composites, as well as transport processes and particle coarsening in heterogeneous materials. By formulating a purely deterministic framework founded on rotational variations of space-filling geometric cells, the stochastic tiling approach provides an exact combinatorial pathway to estimate nearest-neighbor distance probability distributions from basic spatial geometry.
The structural foundation of the three-dimensional model begins with a space-filling polyhedral unit cell consisting of a cube of side length L containing a single sphere of diameter D. Adjusting the dimension of the cube directly alters the overall volume fraction of the system up to the cubic theoretical maximum of approximately 0.52. When the sphere is situated at a corner of the cube rather than its center, eight distinct rotational variants are established. The model constructs a twenty-seven-cube assembly in which a central cube contains a fixed sphere in its front right corner while twenty-six surrounding adjacent cubes adopt random rotational configurations. This geometric arrangement allows researchers to systematically decompose sphere center-to-center distances into six distinct spatial layers surrounding the central particle.
Across the lower five structural layers of the twenty-seven-cube assembly, rotational randomization yields fourteen distinct potential center-to-center distances distributed among ninety-seven spatial positions. These distances range from direct contact at distance D1 equal to sphere diameter D, through geometric combinations involving cube length L and sphere diameter D, up to distance D12, which represents the maximum relevant separation of a neighbor within an adjacent cell. Distances D13 and D14, corresponding to larger diagonal lengths, become mathematically relevant only at extremely low volume fractions below approximately 0.015. Because adjacent cubes cannot place a nearest neighbor farther than D12, distances beyond this threshold do not require evaluation in the nearest-neighbor calculation.
A primary mathematical complexity of extending stochastic tiling into three dimensions arises because the relative numerical ranking of these fourteen distances changes as the volume fraction changes. Specific volume fractions where two distinct interparticle distances become exactly equal are designated as crossover points, labeled C1 through C11. At each crossover boundary, the probability of a given distance being the absolute shortest distance to the central sphere shifts discontinuously. To prevent the prohibitive computational cost of running twenty-six nested loops, the investigators reduced the active calculation domain by restricting evaluations to cubes that can host distances below D12. Probability values for each distance within each crossover interval were computed numerically, yielding explicit probability-weighted formulations across volume fractions from approximately 0.10 up to the cubic theoretical maximum.
To rigorously test and validate the stochastic tiling estimates, computational reference models were generated using a random sequential addition algorithm in MATLAB. For each volume fraction evaluated between 0.0001 and 0.35, four distinct computational volumes containing 250,000 uniform spheres of arbitrary diameter 500 arbitrary units were constructed without particle overlaps or boundary shifts, resulting in pooled statistical ensembles of 106 instances per condition. Spheres were sequentially placed into cubic bounding boxes sized to match target volume fractions until the target density was achieved. Generating random sequential addition patterns above a volume fraction of 0.35 became computationally prohibitive because the configuration approaches the three-dimensional random sequential addition jamming limit of approximately 0.384.
Analysis of the Euclidean nearest-neighbor distances extracted from the computer-generated patterns revealed that three-dimensional distance distributions behave very differently from their two-dimensional counterparts. In two dimensions, nearest-neighbor distributions are known to be highly skewed and non-normal, requiring three-parameter skew normal fits. In contrast, three-dimensional distributions closely resemble truncated normal distributions that conform remarkably well to standard Gaussian functions when the hard-sphere cutoff at one diameter is ignored. At lower volume fractions, distribution truncation occurs predominantly in the left tail due to the physical exclusion boundary, whereas at higher volume fractions, truncation shifts toward the right tail as packing constraints compress available interparticle space.
Direct comparisons between the stochastic tiling predictions and the random sequential addition computational data demonstrate exceptionally strong quantitative agreement across the evaluated volume fractions. The maximum error difference between the predicted mean nearest-neighbor distance and the measured computational mean was approximately 1.22 percent, occurring at a volume fraction of 0.25. Across both smaller and larger volume fractions, deviations generally dropped below 1.0 percent. When extrapolated below volume fractions of 0.10 down to 0.03, the stochastic tiling model continued to yield reasonable approximations that outperformed conventional Poisson point-process estimates, despite relying on extrapolated probability regimes.
The study also compared the stochastic tiling solutions against established formulations developed by Tewari and Gokhale for poly-dispersed random patterns reduced to uniform systems. While both the stochastic tiling equations and the Tewari and Gokhale formulation produced roughly similar trajectories, both analytical approaches slightly underestimate actual means derived from computer-generated patterns in magnified comparison views. Neither model perfectly captures the curvature across intermediate volume fractions, pointing to underlying structural transitions within dense hard-sphere assemblies. In particular, comparative analysis between the fitted normal distribution means and the empirical nearest-neighbor distribution means revealed a distinct linear slope transition at a volume fraction of approximately 0.2445.
Further investigation of distribution standard deviations and modal behavior highlighted distinct spatial regimes governing three-dimensional packing. The standard deviation of the nearest-neighbor distribution exhibited a pronounced retrograde trajectory at high packing densities, departing from linearity at a volume fraction of 0.25 and reaching an abscissa minimum at 0.28, which may signal localized cluster formation. Furthermore, true distribution modes appeared only at volume fractions of 0.08 and lower, where the most frequent distance exceeded the hard-core exclusion limit. Above 0.09, maximum frequencies concentrated strictly at the hard-core contact limit, peaking smoothly near a volume fraction of 0.16. These transition boundaries suggest that comprehensive analytical models must incorporate piecewise mechanisms, establishing a quantitative baseline for ongoing extensions to variable-diameter microstructural systems.
Beyond establishing general spatial trends, the computational analysis incorporated rigorous statistical protocols to evaluate the simulated distance histograms. The researchers determined histogram resolutions using Freedman-Diaconis optimal bin sizes rounded to the nearest thousandth, extracting skewness, kurtosis, and median values alongside standard means. Across all evaluated patterns, distribution medians remained strictly lower than their corresponding mean values, accompanied by significant skewness. Curve fitting for the Gaussian approximations relied on the generalized reduced gradient nonlinear solver in Excel to derive distribution means and standard deviations. This fitting procedure encountered numerical constraints only at the highest tested volume fraction of 0.35, where the solver required an explicit non-negative mean constraint to converge, though the resulting curve still closely matched the simulated data.
The investigation also provided quantitative adjustments to existing analytical benchmarks for uniform packing. When contrasting the stochastic tiling predictions with the classic formulation from Tewari and Gokhale, the authors evaluated the governing expression scaled by the theoretical close-packed limit of approximately 0.7405 and the sphere number density. While the original benchmark applied baseline constants of 0.55396 and 1.02625, performing a direct least-squares fit against the generated random sequential addition data revealed that revised parameters of 0.545044 and 1.116187 reduced overall errors. Additionally, the mechanics behind the modal frequency maximum near a volume fraction of 0.16 were attributed to an interplay between distribution shifts: as packing density increases, the leftward movement of the normal distribution mean competes directly against the narrowing normal distribution standard deviation, maximizing boundary contact frequencies.
Physical interpretations of the high-density standard deviation retrograde offer further insight into spatial reorganization. After deviating from linear scaling at a volume fraction of 0.25 and reaching an abscissa minimum at 0.28, the standard deviation curve reached an ordinate minimum at a volume fraction of 0.33. The authors suggested that as particle separations compress slightly within local clusters, additional open space emerges outside those clusters, allowing the overall standard deviation to increase. Finally, the researchers noted specific computational constraints. Below volume fractions of 0.005, Gaussian distributions overpredicted frequencies near the contact boundary, likely because significantly larger spatial volumes require larger sampling ensembles. Similarly, combinatorial calculations on a laptop computer were computationally constrained to volume fractions above approximately 0.10, indicating that evaluating lower packing fractions will require greater computational resources or specialized combinatorial approaches.
Subject of Research: Three-dimensional stochastic tiling model for estimating mean nearest-neighbor distances in uniform random sphere distributions
Article Title: A stochastic tiling model of mean nearest-neighbor distances in three-dimensional uniform random distributions
Article References: Everett, R. K., & Zupan, M. (2025). A stochastic tiling model of mean nearest-neighbor distances in three-dimensional uniform random distributions. Journal of Materials Science: Materials Theory, 9(1), Article 2. https://doi.org/10.1186/s41313-025-00068-y
Image Credits: AI Generated
DOI: 10.1186/s41313-025-00068-y
Keywords: stochastic tiling, nearest-neighbor distance, microstructure analysis, spatial statistics, random sequential addition, hard-sphere distributions, volume fraction, materials theory, quantitative metallography, computational geometry, clustering metrics, spatial point patterns, Gaussian distribution, jamming limit, microstructural descriptors
News Source: Denise Maddox. (November 21, 2026). Stochastic Tiling Model Accurately Predicts Nearest-Neighbor Distances in 3D Microstructures. Scienmag.



