The STARFLAG handbook on collective animal behaviour, part 2 (Cavagna et al. 2008) codified the border-corrected toolkit for analysing the three-dimensional (3D) structure of large animal groups in the field — α-shape borders for non-convex groups, Hanisch border correction, the anisotropy factor γ for directional neighbour structure, and inscribed-sphere measurements of group dimensions — and left a set of unresolved issues running through the text: the hard-sphere model of the exclusion zone around each animal is a crude fit to the nearest-neighbour distance distribution; the biological interpretation of the anisotropy factor — its dependence on neighbour order and system, and whether the direction of maximal crowding describes other systems more clearly — remains to be established; α-shape scale selection is objective but laborious, and the inscribed-sphere measurement of group dimensions relies on the same α-shape border; geometric border effects and genuine behavioural differences of border individuals are not disentangled; the invariance of aspect ratios (1:3:6 in starlings) is untested in other species; how the orientation of the group axes relative to gravity and velocity changes under external stimuli such as predation is unexplored; and the framework cannot deliver unbiased statistics for small groups, where nearly all animals lie on the border. Substantial progress exists on interaction anisotropy in a second bird species and its biophysical mechanism, on model-based evidence that border animals carry an inward motion bias, on force-inference methods that supersede hard-sphere descriptions in other systems, and on collective escape under predation; the systematic cross-species and cross-context test of 3D structural laws with geometric border effects controlled, the empirical question of border-animal behaviour, unbiased statistics for small groups, and automated border-definition methods remain open.

Background

The study of collective animal behaviour must compare theoretical models with field data, and the richest data are the three-dimensional (3D) positions of every individual in a moving group. The STARFLAG project (an EU-funded collaboration based in Rome) reconstructed the 3D positions of up to a few thousand starlings, Sturnus vulgaris, filmed in the field with calibrated stereoscopic cameras; its methods handbook, part 2 (Cavagna et al. 2008), reviews the analytical tools that became a de-facto standard for 3D studies of large groups, while the companion papers report the empirical findings (Ballerini et al. 2008a,b).

The central technical obstacle is the border of the group. Points at the surface have empty space on one side, so their neighbour statistics are biased, and the border's weight scales with the surface-to-volume ratio, which shrinks only as the group grows (proportionally to one over the cube root of the number of animals). If border points are naively included, purely geometric artefacts appear — for example an apparent decrease of the nearest-neighbour distance (NND, the distance from an animal to its closest neighbour) with group size, or spurious directional anisotropy. The handbook therefore imports border-corrected estimators from spatial point-process statistics, notably the Hanisch estimator, which includes in each animal's neighbour statistics only those neighbours closer than the animal's own distance to the border. The border itself is defined by the α-shape, a generalization of the convex hull that carves out concavities of a chosen radius R, giving a non-convex boundary for non-convex groups; choosing R objectively requires scanning the internal density as a function of R and locating sudden jumps.

On top of this, the handbook defines the anisotropy factor γ: the squared projection of the normalized flock velocity onto the eigenvector of the neighbour-orientation matrix associated with its smallest eigenvalue, i.e. the direction in which neighbours are least crowded. A spatially random arrangement averages γ = 1/3; larger values mean neighbours avoid the direction of motion. In starlings γ exceeds 1/3 for the first few neighbours and returns to 1/3 at a range that is topological (counted in number of neighbours, about six to seven) rather than metric (in metres) (Ballerini et al. 2008a). The short-range dearth of neighbours is modelled as an exclusion zone of impenetrable hard spheres of radius r0 fitted to the NND distribution, acknowledged to be a crude approximation of a soft zone of repulsion. Group shape is summarized by aspect ratios of the three principal dimensions, measured for non-convex groups via the largest sphere inscribable in the α-shape; in starlings the average aspect ratio is 1:3:6, the shortest axis lies close to gravity and orthogonal to velocity, and the ratio stays relatively stable as group size varies (Ballerini et al. 2008b). Similar elongation along the direction of travel was reported for fish schools (Partridge et al. 1980).

The handbook itself leaves open what this framework cannot yet settle. Those unresolved issues — spanning model realism, biological interpretation, computational practicality, cross-species generality, and the small-group regime — are the subject of this problem record.

Problem Statement

The problem is to establish which three-dimensional structural features of moving animal groups are general biological properties, which are species- or context-specific, and which are artefacts of measurement geometry — and to extend the measurement framework itself where it currently fails. The source (Cavagna et al. 2008, the STARFLAG handbook, part 2) leaves the following issues open, comprising the scientific comparisons (2), (4), (5) and (6) that constitute the problem and the methodological enablers (1), (3) and (7) that serve them:

(1) Exclusion zone: replace the crude hard-sphere model fitted to the nearest-neighbour distance distribution by more realistic models of the zone of repulsion around each animal, carefully balancing the trade-off between increased realism and a larger number of parameters.

(2) Anisotropy: determine the biological significance of the anisotropy factor γ (what its departure from the isotropic value 1/3 indicates about the animals' interactions), how it depends on neighbour order and on species, and whether the direction of maximal crowding (rather than minimal crowding) gives a clearer description of neighbour anisotropy in systems other than starlings — where "clearer" means that the anisotropy factor measured along the direction of maximal crowding departs further from the isotropic value 1/3, over the same neighbour orders, than the factor measured along the direction of minimal crowding.

(3) Computational scalability: make the objective determination of the α-shape radius R (currently a laborious density-versus-R scan) and the largest-inscribed-sphere measurement of group dimensions practical for very large groups.

(4) Border biology: disentangle geometric border effects from genuine biological differences — determine whether individuals at the border of a group behave differently from interior individuals once the trivial geometric bias of the border is removed.

(5) Shape: test the generality of the starling findings on aspect ratios (average 1:3:6, relatively invariant under size changes) in other species.

(6) Orientation dynamics: characterize how the orientation of the group's principal axes relative to gravity and velocity changes under external stimuli, especially predation.

(7) Small groups: develop analysis approaches that give unbiased structural statistics for small groups (tens of animals), where the surface-to-volume ratio is so large that border correction leaves too few interior points for reliable estimation.

The common objective unifying these components is the geometry-controlled comparative biology of 3D group structure: components (2), (4), (5) and (6) are the scientific questions about generality and mechanism, and components (1), (3) and (7) are the methodological enablers needed wherever the current framework is the limiting factor (refined interaction models, scalable border definition, small-group statistics). "Geometry-controlled" here means that each comparison is made with the source's border-corrected estimators — the Hanisch-corrected neighbour count evaluated against the α-shape border — so that geometric border bias is removed before any claim is drawn. The four scientific components are posed jointly, each as a question about generality and mechanism under this geometry-controlled comparison, carried out across more than one species and more than one ecological context rather than inferred from a single system. For component (2), the question is to characterize, for each system studied, the biological significance of the anisotropy factor (what its departure from the isotropic value 1/3 indicates about the animals' interactions), its dependence on neighbour order and on species across neighbour orders up to the topological range, and the maximal-versus-minimal-crowding comparison on the clarity criterion defined above; for component (4), the question is whether border individuals behave differently from interior individuals once the geometric bias is removed; for component (5), the question is whether both the average aspect ratio and its invariance under size changes match the starling benchmark, with the measured values obtained with the same border-corrected estimators and compared against the size-to-size scatter the source itself reports for starling flocks; and for component (6), the question is how the three principal axes orient relative to gravity and velocity in the absence of, and under at least one external stimulus such as predation.

References

  1. Andrea Cavagna, Irene Giardina, Alberto Orlandi, Giorgio Parisi, Andrea Procaccini. The STARFLAG handbook on collective animal behaviour: 2. Three-dimensional analysis. Animal Behaviour 76(1):237–248, 2008. DOI: 10.1016/j.anbehav.2008.02.003 (arXiv:0802.1674). https://doi.org/10.1016/j.anbehav.2008.02.003
  2. M. Ballerini, N. Cabibbo, R. Candelier, A. Cavagna, E. Cisbani, I. Giardina, V. Lecomte, A. Orlandi, G. Parisi, A. Procaccini, M. Viale, V. Zdravkovic. Interaction ruling animal collective behavior depends on topological rather than metric distance: Evidence from a field study. Proceedings of the National Academy of Sciences 105(4):1232–1237, 2008. DOI: 10.1073/pnas.0711437105. https://doi.org/10.1073/pnas.0711437105
  3. Michele Ballerini, Nicola Cabibbo, Raphael Candelier, Andrea Cavagna, Evaristo Cisbani, Irene Giardina, Alberto Orlandi, Giorgio Parisi, Andrea Procaccini, Massimiliano Viale, Vladimir Zdravkovic. Empirical investigation of starling flocks: a benchmark study in collective animal behaviour. Animal Behaviour 76(1):201–215, 2008. DOI: 10.1016/j.anbehav.2008.02.004 (arXiv:0802.1667). https://doi.org/10.1016/j.anbehav.2008.02.004
  4. Brian L. Partridge, Tony Pitcher, J. Michael Cullen, John Wilson. The three-dimensional structure of fish schools. Behavioral Ecology and Sociobiology 6(4):277–288, 1980. DOI: 10.1007/BF00292770. https://doi.org/10.1007/BF00292770
  5. Hangjian Ling, Guillam E. Mclvor, Kasper van der Vaart, Richard T. Vaughan, Alex Thornton, Nicholas T. Ouellette. Local interactions and their group-level consequences in flocking jackdaws. Proceedings of the Royal Society B 286:20190865, 2019. DOI: 10.1098/rspb.2019.0865. https://doi.org/10.1098/rspb.2019.0865
  6. Hangjian Ling, Guillam E. Mclvor, Kasper van der Vaart, Richard T. Vaughan, Alex Thornton, Nicholas T. Ouellette. Costs and benefits of social relationships in the collective motion of bird flocks. Nature Ecology & Evolution 3:943–948, 2019. DOI: 10.1038/s41559-019-0891-5. https://doi.org/10.1038/s41559-019-0891-5
  7. Daniel J. G. Pearce, Adam M. Miller, George Rowlands, Matthew S. Turner. Role of projection in the control of bird flocks. Proceedings of the National Academy of Sciences 111(29):10422–10426, 2014. DOI: 10.1073/pnas.1402202111 (arXiv:1407.2414). https://doi.org/10.1073/pnas.1402202111
  8. D. J. G. Pearce, M. S. Turner. Density regulation in strictly metric-free swarms. New Journal of Physics 16:082002, 2014. DOI: 10.1088/1367-2630/16/8/082002. https://doi.org/10.1088/1367-2630/16/8/082002
  9. J M Lewis, M S Turner. Density distributions and depth in flocks. Journal of Physics D: Applied Physics 50:494003, 2017. DOI: 10.1088/1361-6463/aa942f (arXiv:1902.08181). https://doi.org/10.1088/1361-6463/aa942f
  10. James G. Puckett, Douglas H. Kelley, Nicholas T. Ouellette. Searching for effective forces in laboratory insect swarms. Scientific Reports 4:4766, 2014. DOI: 10.1038/srep04766. https://doi.org/10.1038/srep04766
  11. Charlotte K. Hemelrijk, Hanno Hildenbrandt, José Reinders, Eize J. Stamhuis. Emergence of oblong school shape: models and empirical data of fish. Ethology 116(11):1099–1112, 2010. DOI: 10.1111/j.1439-0310.2010.01818.x. https://doi.org/10.1111/j.1439-0310.2010.01818.x
  12. Alessandro Attanasi, Andrea Cavagna, Lorenzo Del Castello, Irene Giardina, Stefania Melillo, Leonardo Parisi, Oliver Pohl, Bruno Rossaro, Edward Shen, Edmondo Silvestri, Massimiliano Viale. Collective behaviour without collective order in wild swarms of midges. PLoS Computational Biology 10(7):e1003697, 2014. DOI: 10.1371/journal.pcbi.1003697. https://doi.org/10.1371/journal.pcbi.1003697
  13. Hungtang Ko, Abigail Girma, Yangfan Zhang, Yu Pan, George Lauder, Radhika Nagpal. Beyond planar: fish schools adopt ladder formations in 3D. Scientific Reports 15:20249, 2025. DOI: 10.1038/s41598-025-06150-2. https://doi.org/10.1038/s41598-025-06150-2
  14. Andrea Procaccini, Alberto Orlandi, Andrea Cavagna, Irene Giardina, Francesca Zoratto, Daniela Santucci, Flavia Chiarotti, Charlotte K. Hemelrijk, Enrico Alleva, Giorgio Parisi, Claudio Carere. Propagating waves in starling, Sturnus vulgaris, flocks under predation. Animal Behaviour 82(4):759–765, 2011. DOI: 10.1016/j.anbehav.2011.07.006. https://doi.org/10.1016/j.anbehav.2011.07.006
  15. Daniel O'Coin, Guillam E Mclvor, Alex Thornton, Nicholas T Ouellette, Hangjian Ling. Velocity correlations in jackdaw flocks in different ecological contexts. Physical Biology 20, 2022. DOI: 10.1088/1478-3975/aca862. https://doi.org/10.1088/1478-3975/aca862
  16. Marina Papadopoulou, Hanno Hildenbrandt, Rolf F. Storms, Claudio Carere, Simon Verhulst, Charlotte K. Hemelrijk. A mechanistic understanding of collective escape in starling flocks. Communications Biology 9:1055, 2026. DOI: 10.1038/s42003-026-10173-4. https://doi.org/10.1038/s42003-026-10173-4
  17. Venu Divin, Jitesh Jhawar. Predator Presence Influences The Geometry of Rosy Starling Flocks. Integrative and Comparative Biology 66, 2026. DOI: 10.1093/icb/icag027. https://doi.org/10.1093/icb/icag027
  18. Jérôme Hem, Alexis Poncet, Pierre Ronceray, Daiki Nishiguchi, Vincent Démery. Learning general pair interactions between self-propelled particles. arXiv:2507.13667, 2025. https://arxiv.org/abs/2507.13667