Pour sand, tap a jar of beads, or compact a powder, and the packing settles into a reproducible state with reproducible density fluctuations — even though nothing thermal is shaking it. For decades, measurements of the local (Voronoi) volumes around each particle in such jammed packings have shown a striking regularity: the volume fluctuations follow a Gamma distribution, across very different materials and preparation protocols. But why a Gamma? Was it a deep property of granular matter or a coincidence of how the measurements were fit? This paper answers that question.
The authors show — analytically, from the maximum-entropy principle, and experimentally, with x-ray tomography of ~30,000-particle tapped packings — that the intrinsic configurational density of states Ω(v) of jammed packings is itself a Gamma distribution when expressed in terms of the free volume
where υ is the local specific volume and υmin is the lower bound of locally accessible volume — the part of space that geometry and friction can never squeeze out. The density of states counts which mechanically stable configurations exist at all, independent of how the packing is driven:
Here k is a dimensionless shape factor (an effective number of geometric degrees of freedom) and θ sets the scale of structural fluctuations.
Within the Edwards ensemble (where volume plays the role of energy and compactivity X plays the role of temperature), the observed volume distribution is the intrinsic DOS weighted by a Boltzmann-like factor:
Because Ω(v) is Gamma, P(v) stays a Gamma — the exponential weighting only shrinks the scale parameter to θ̂ = (1/θ + 1/X)−1. So the experimentally famous Gamma shape is inherited from structural accessibility, while the preparation (tapping intensity) enters only through the state variable X. Fitting a single Gamma to data and reading physics off its parameters — standard practice until now — conflates these two distinct contributions.
The factorized form yields every thermodynamic quantity in closed form — the partition function Z(X) = [X/(X+θ)]k, the mean and variance, the compactivity X = θv̄/(kθ − v̄), and the configurational entropy. Most usefully, eliminating the state variable leaves a compactivity-independent structural relation:
Packings prepared at every tapping intensity must fall on this one straight line in the (σ, ῡ) plane. The experiments confirm the collapse, and its slope and intercept hand over k and υmin directly — no distribution fitting at all. With those parameters fixed, the theory predicts the full P(v) at every driving intensity, and the predictions match the measured histograms with no adjustable parameters.
The paper resolves a subtle but consequential ambiguity: should the Edwards ensemble conserve the absolute volume υ or the free volume v? Only the free-volume formulation gives a physically meaningful zero-compactivity limit (X → 0 collapses to v = 0, a unique reference state with no configurational freedom) and a consistent partition function — the two choices differ by a factor that grows exponentially as packings densify.
Compactivities and entropies from the factorized Gamma framework agree with two established model-independent methods — the overlap-histogram method and the variance–fluctuation relation — across all particle types tested. The Edwards fluctuation relation is in fact satisfied analytically by the framework. Smooth spheres of different sizes share nearly identical DOS parameters (k ≈ 62–70), while rougher and bumpier particles show smaller k and larger θ — friction widens the range of accessible configurations. The DOS is a measurable material property.
The deepest contribution is not that P(v) is Gamma — that was known empirically — but that the observed distribution factorizes into an intrinsic structural density of states and a state-dependent Boltzmann weighting. That separation mirrors equilibrium statistical mechanics, where a material is characterized by its density of states and its condition by a temperature:
This resolves the long-standing inconsistencies between compactivity values extracted by different methods, gives granular matter a quantitative, predictive statistical-mechanical foundation, and turns a decades-old empirical curiosity — the ubiquitous Gamma histogram — into a window on the configurational landscape of jammed matter.
Because the DOS parameters (k, θ, υmin) are driving-independent material fingerprints, and because the structural relation σ² = v̄²/k needs only a mean and a standard deviation — no distribution fitting — the framework lends itself to practical use far beyond the physics lab:
Summarize an entire packing by three physical numbers (k, θ, X) instead of a million Voronoi volumes. Monitor mean and standard deviation online — a departure from the structural line flags a change in the material itself. Lot release, batch comparison, drift detection.
Digital twins of pharmaceutical powders and tablets (a DOS fingerprint per batch), flowability prediction beyond Carr index and Hausner ratio, powder metallurgy and sintering outcomes from the initial pore structure, powder-bed 3D printing quality.
Soil state described by (k, θ) + compactivity instead of a zoo of empirical indices — compaction, settlement, liquefaction risk; porous rock characterization for reservoirs, CO₂ sequestration, and groundwater transport.
Battery electrodes are porous granular structures: the void distribution, not just porosity, governs ion transport and degradation. Filtration media (sand, activated carbon, ceramics): permeability, clogging, residence time.
P(v) tails mark weak spots and incipient rearrangements — silo clogging probability, granular robotics, rovers and diggers on lunar or asteroid regolith inferring sinkability and traction from the local free-volume statistics; granular foods handling.
The compact feature vector (k, θ, X, υmin) for machine-learning models of classification, anomaly detection, and process optimization; physics engines initializing realistic sand/snow packings from Ω(v) rather than empirical calibration.
Most speculative — and perhaps most far-reaching: the Gamma DOS follows from maximum entropy under generic constraints on a conserved "free resource." The same reasoning may apply wherever a complex system allocates spare capacity — porous materials, cellular tissues, transportation or communication networks, even memory in computing systems — suggesting a template for extending statistical mechanics to a broad class of athermal, constrained systems.
The interactive labs on this site implement this paper's entire framework: pour and tap packings, measure Voronoi specific volumes, watch P(v) build up over the free volume with its factorized-Gamma prediction, and read off k, θ, υmin, compactivity, Z(X), and entropy from the structural relation — just as in the experiments.