A colloidal viewpoint on the sausage catastrophe and the finite sphere packing problem

The finite sphere packing problem asks how to arrange a finite number of identical spheres to maximize packing efficiency within the smallest enclosing container (e.g., convex hull). While infinite packings are maximized by the FCC lattice (Kepler conjecture), finite packings exhibit different behavior. Fejes Tóth conjectured that for sufficiently high dimensions the densest finite packing becomes a linear “sausage” arrangement. In 4D, a transition (the “sausage catastrophe”) from linear to cluster packing occurs at large N; in 3D, prior studies suggest linear is optimal up to N ≤ 55 and at N = 57, 58, 63, 64, with clusters denser thereafter, though the exact cluster structures are largely unknown. This work aims to realize the finite sphere packing problem physically using colloidal hard spheres confined in flexible giant unilamellar vesicles (GUVs), quantify when linear, planar, or clustered arrangements are favored, examine bistability induced by membrane fluctuations, and test whether clusters can surpass linear packing in 3D near the predicted sausage catastrophe regime.

Historical and theoretical context includes the Kepler conjecture (FCC packing ~74% for infinite 3D packings) and its proof. Finite packings depend on container shape or the convex hull. Fejes Tóth conjectured that linear (sausage) packings are optimal for sufficiently high dimensions, proven for d ≥ 13387 and later for d ≥ 42, with no general proof for lower d. In 4D, a sausage-to-cluster transition occurs at very large N (upper bound reduced to 338,196). In 3D, studies indicated linear optimality for N ≤ 55 and specific N (57, 58, 63, 64), with clusters outperforming linear beyond these, but optimal cluster structures remained uncertain. Experimental realizations of sausage- and plate-like arrangements in physical systems have been scarce. Colloidal systems and GUVs provide a route to study excluded-volume packings under flexible confinement with measurable membrane properties and controllable osmotic conditions.

Experiments: Colloidal silica (2.12 µm) and sterically stabilized, fluorescent polystyrene (2.0 µm, 3% polydispersity) particles were encapsulated in GUVs using a modified droplet transfer method. Vesicles were formed from DOPC with fluorescent lipid (Liss Rhod PE) in paraffin or mineral oil-based lipid oil solutions. Inner solution: silica particles in 100 mM sucrose; outer solution: 110 mM glucose to induce hypertonic shock and obtain flaccid vesicles. Vesicles were imaged via confocal laser scanning and fluorescence microscopy; XY-Z stacks captured vesicle shape and particle arrangements, with temporal imaging at 2–5 fps. Some samples included non-adsorbing PAM polymer to immobilize vesicles via depletion to the substrate. Vesicle volume and surface area were extracted from 3D stacks using ImageJ/Fiji with smoothing, thresholding, Shape Smoothing plugin, and BoneJ Particle Analyser; axial scaling was corrected for spherical aberration (factor 0.83). Bending rigidity and membrane tension were measured by flickering spectroscopy, yielding κ ≈ 18 ± 6 kBT and λb ≈ 13 ± 7 nN m−2. Parameters and order metrics: The reduced volume v = Vv/Vs = 3√(4πVv²)/Av^{3/2} (0 < v ≤ 1) was used to compare vesicles of different sizes and surface areas. Anisotropy of particle arrangement was quantified by κ² = 3(ax²+ay²+az²)/[2(ax+ay+az)²] − 1/2 from the gyration tensor eigenvalues, distinguishing linear (κ² ≥ 0.5), planar (0.2 ≤ κ² ≤ 0.3), and cluster (κ² ≈ 0) conformations. Packing fraction within the vesicle η = N V0 / Vv spanned 0.12–0.28 in experiments/simulations for N ∈ [3,9]. Simulations (state diagram, small N): Molecular dynamics with a meshless, orientation-dependent membrane model representing lipids as spheres of diameter σ (unit length). The membrane had near-zero surface tension; explicit solvent was included outside to exert pressure and control shape. Colloids (diameter σc = 120) interacted via purely repulsive WCA potential. Systems with N ∈ [3,9] spheres were simulated across vesicle sizes to map conformations versus v. Simulations (cluster screening, larger N): To probe the sausage catastrophe regime, N ∈ [10,150] spheres were packed in tight spherical vesicles (η ≈ 0.4) using the same membrane model. Additionally, dense clusters from a Lennard-Jones cluster database were considered. A numerical optimization protocol approached the hard-sphere limit and computed the convex hull volume Vch to obtain ηch = Nv V0 / Vch. ηch values were compared to the ideal linear (spherocylinder) packing ηlin(N). Bond-order parameter g4 characterized structural ordering. Cluster families analyzed: Planar arrays (triangular, square, hexagonal) and 3D polyhedral clusters cut from FCC crystals, including tetrahedra (T), octahedra (O), and bipyramids (B), with regular and truncated variants. Notation X_n^k indicates removal of n particles from k vertices (X_n used for isotropic cuts); irregular, asymmetric truncations were also explored systematically near N ~ 56–70 to search for clusters surpassing linear packing.

• State diagram (N ∈ [3,9]): For flexible GUV confinement, three regimes emerge versus reduced volume v: clusters at high v (>0.9), plates at intermediate v, and linear (sausage) arrangements at low v. The stability window of linear arrangements broadens with increasing N; for N > 9, vesicles became excessively bent. Experiments and simulations agree.
• Bistability: Simulations reveal bistable regions where κ² toggles between plate–linear and plate–cluster values (demonstrated for N = 4), corroborated by time-lapse microscopy and simulations showing reversible plate-to-line transitions driven by membrane fluctuations.
• Packing in flexible vesicles: At nearly fixed surface area (simulations), linear arrangements consistently correspond to lower v (higher packing efficiency) than plate or cluster conformations for all N studied (3–9). Packing efficiency increases with N.
• Large-N cluster screening using convex hull packing fraction ηch: • Planar packings have ηch well below ηlin, consistent with optimality requiring 1D or 3D structures. • Tetrahedra: For regular tetrahedra, ηch crosses ηlin only at N = 84. Truncation increases ηch; the isotropically truncated T4 achieves ηch > ηlin at N = 68 (lowest-N tetrahedron surpassing line among regular truncations). • Octahedra: Regular and regularly truncated octahedra exhibit generally lower ηch than tetrahedra; the O10 reaches ηch > ηlin at N = 79. • Irregular/asymmetric truncations near the predicted sausage-catastrophe regime: Only asymmetrically sliced polyhedra achieve ηch > ηlin in the N ≈ 56–70 range. Notably, clusters with N = 58 and N = 64 surpass the linear (sausage) packing, overturning earlier suggestions that linear is optimal at these N. Overall, clusters with 56 ≤ N ≤ 70 (excluding N = 57 and 63) were identified that pack more efficiently than a line.
• Structural determinants: No simple correlation between ηch and individual polyhedral features (faces, edges, vertices); improved packing arises from subtle, nontrivial combinations of structural elements.
• Experimental parameters: Vesicle bending rigidity κ ≈ 18 ± 6 kBT; membrane tension λb ≈ 13 ± 7 nN m−2. Experimental/simulation packing fractions within vesicles spanned η ≈ 0.12–0.28; reduced-volume thresholds discriminated linear (low v), plate (intermediate v), and cluster (high v) regimes.
• Implication: The results provide direct, physics-inspired evidence supporting Fejes Tóth’s conjecture in 3D finite systems by exhibiting specific clusters denser than the sausage for select N.

The study addresses whether finite numbers of equal spheres pack most efficiently as compact clusters or as linear sausages when confined by the minimal convex hull or a flexible container. By realizing the problem experimentally with colloids in GUVs and corroborating with simulations, the work establishes that under flexible vesicle confinement, linear conformations maximize packing efficiency for small N (3–9) compared to plates or clusters, consistent with the sausage picture. Membrane fluctuations can induce bistability, revealing the delicate balance between container shape, surface area-to-volume ratio, and particle arrangement. Extending to larger N via convex-hull analysis, the authors demonstrate that while many regular 3D clusters do not outperform the line near N ~ 56–70, appropriately faceted, asymmetrically truncated clusters can, with explicit examples at N = 58 and 64 and more generally across 56 ≤ N ≤ 70 excluding 57 and 63. This practically demonstrates the early onset of the sausage catastrophe in 3D finite packings for specific cluster types and refines previous expectations about N values. These findings highlight that packing optimality depends sensitively on detailed cluster geometry rather than simple regular polyhedra and provide concrete targets for future mathematical proofs and experimental realizations in systems with strong attractions.

This work provides a colloidal, experimental-simulation realization of the finite sphere packing problem in flexible vesicles. It maps a state diagram versus reduced volume v showing linear, plate, and cluster regimes and identifies bistable transitions driven by membrane fluctuations. For N ≤ 9 in flexible vesicles, linear arrangements consistently yield higher packing efficiency than plate or cluster configurations. Systematic exploration of 3D clusters using convex-hull packing fractions reveals that asymmetrically faceted, ordered clusters can surpass the linear sausage near the predicted catastrophe regime, with explicit cases at N = 58 and 64 and more broadly within 56 ≤ N ≤ 70 (excluding 57 and 63). These results offer direct, physics-based support for Fejes Tóth’s conjecture in finite 3D packings and uncover previously unidentified optimal clusters. Future directions include developing mathematical proofs for these clusters and the full conjecture, computational generation of candidate clusters via conventional and machine-learning strategies (including diverse Barlow stackings), and experimental realization in nanoparticle systems with strong attractions (e.g., gold or platinum) where faceted clusters might self-assemble. The approach may also enable pre-assembly of colloidal building blocks for metamaterials and other functional structures.

• Experimental scope limited to N ≤ 9 due to excessive vesicle bending and computational cost; larger-N conclusions rely on simulations and convex-hull analysis.
• Precise control of vesicle surface area at fixed N is challenging experimentally; very high packing fractions (tight convex-hull-like confinement) are not experimentally accessible with GUVs.
• Imaging introduces artifacts (e.g., spherical aberration, membrane undulations) requiring corrections; residual errors affect reduced volume estimates (≈3–8%).
• Simulations use a coarse-grained, meshless membrane and WCA interactions rather than exact hard spheres; the hard-sphere limit is approached numerically for ηch comparisons.
• Faceted clusters that beat the sausage were identified via simulations/optimization; direct fabrication inside flexible vesicles is not feasible, and realization may require systems with strong attractions.
• No mathematical proof is provided; results are empirical/numerical demonstrations that motivate, but do not replace, formal proofs.