Information transfer in finite flocks with topological interactions
Introduction
The scale and grace of bird flocks are nature’s most impressive displays, arising from seemingly chaotic flight paths of individual birds. Flocking behaviour is not restricted to just birds, either, with many other species displaying similar movements, from schools of fish [1] to colonies of bacteria [2]. The phenomenon of collective motion, and its constituent components, are subject to much research [3].
While new stereographic camera techniques and equipment allow research into large real-world flocks, much of the literature continues to use abstract models, such as the Standard Vicsek Model (SVM) [4] or the Inertial Spin Model [5]. These models approximate real-world systems as point particles whose decision making processes are encapsulated by local neighbour interaction rules and errors (noise). The SVM is perhaps the most minimal such model in that it contains only a single rule obeyed by all particles: assume the average direction of the local neighbourhood of radius with some random perturbation added [4], where the magnitude of the noise determines whether the flock is coherent or not. The alternate model, the Topological Vicsek Model, behaves likewise, with the local neighbourhood instead defined as the closest neighbours.
An implication of local interaction rules is that observing one flock member can reveal information about its nearby flock mates—for instance, we can better guess, or be more certain of, neighbouring headings. Information Theory allows this reduction in uncertainty to be quantified. Mutual Information (MI) [6] defines information sharing, the symmetric and instantaneous reduction in uncertainty about the heading of one flock member when any other member is observed. Information transfer instead quantifies the temporal reduction in uncertainty between flock members, that is, how knowledge of the current heading affects estimation of a future heading, which we measure here with Global Transfer Entropy (GTE) [7].
Information sharing among flock members is crucial to flock formation and stability, yet in the SVM it behaves unexpectedly in these finite flocks [8], [9]: MI is shown to diverge as flocks become increasingly ordered, while information transfer converges to a finite, non-zero value, where both are expected to vanish in a highly ordered flock.
The unexpected behaviour in the SVM is due to continuous symmetry in the system. Fluctuations in a highly ordered flock allow the flock orientation to proceed on a random walk without affecting overall order. When observed over small periods of time, an ordered flock is confined to a small region of phase space – that is, the system is restricted to only a small number of all possible states or flock configurations – producing more expected results—namely, MI and GTE tend to zero as flocks become more ordered, as seen in [10] for MI. However as observation time is increased, flock orientation gradually drifts on a random walk, eventually becoming uniform over in the long-term (infinite) limit (See Fig. 1), exploring the entire phase space along the way. When the system is confined in phase space, it is said to have broken ergodicity; the assumption that behaviour averaged over time is the same when averaged over phase space. Thus over short observation periods, the SVM breaks ergodicity, while in the long-term observation limit, it is restored, leading to continuously-broken ergodicity [11] and the phenomena of diverging MI and non-zero GTE. The central question of this letter is whether the same holds for the TVM.
The SVM flock members interact with neighbours within a fixed radius, i.e., metric interactions. However, it is becoming increasingly apparent that this is not the case for many real-world flocks. Ballerini et al. [12] show via detailed 3D recordings of starlings that birds in real-world flocks instead interact in a topological manner. That is, a bird will interact with its closest six to seven neighbours, regardless of distance. Similar work has shown that 3D schools of fish, and (effectively) 2D herds of sheep and deer also utilise topological interactions [13], [14]. Niizato and Gunji [15] further shows that in “imperfect” biological organisms there might not be a strict delineation between the two, as both interaction methods can be seen as “uncertain” approximations of each other.
Here we apply the information theoretic metrics, MI and GTE, to a topological variant of the Vicsek Model (TVM) in which flock members take the average direction of their closest neighbours rather than those within units.
By noting a fundamental instability in the TVM over long time scales [16] we show here that the information theoretic behaviours of TVM are not only comparable to those of its metric counterpart, but in fact converge to the long-term limiting behaviours with shorter observation windows.
Section snippets
The topological Vicsek model
The two dimensional TVM comprises a set of point particles (labelled ) moving on a plane of linear extent, , with periodic boundary conditions. Each particle moves with constant speed, , and interacts only with the closest neighbours. Positions, , and headings, , are updated synchronously at discrete time intervals according to respectively, where is constructed from and the constant , is the
Methodology
We start in Section 3.1 with definitions of the information theoretic quantities we employ in this study, along with the dimensionally reduced forms outlined in [8], [9]. Following this, we discuss experimental simulation details in Section 3.2.
Results
Fig. 2 shows the long-term MI, , estimated in simulation according to Eq. (5), which behaves similarly to the metric case, particularly, the absence of any peak at the phase transition, and divergence to as .
The long-term GTE, , on the other hand behaves completely differently to the metric case around the phase transition. In the metric case, the noise value, , at which reached the convergence value scaled with both the velocity and the peak in , where higher velocities
Conclusion
Analysis of the topological Vicsek model over short observation windows reveals that information transfer more rapidly approaches the long term-limit behaviour in the TVM than its metric counterpart. This phenomenon arises due to an inherent instability in the topological Vicsek model which enables much faster exploration of the total phase volume. Future work would measure the information theoretic quantities of more stable models, perhaps even those measured from real-world flocks.
While the
CRediT authorship contribution statement
Joshua M. Brown: Software, Investigation, Methodology, Visualization, Writing - original draft. Terry Bossomaier: Conceptualization, Project administration, Writing - review & editing. Lionel Barnett: Conceptualization, Writing - review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
Joshua Brown would like to acknowledge the support of his Ph.D. program and this work from the Australian Government Research Training Program Scholarship.
The National Computing Infrastructure (NCI) facility provided computing time for the simulations under project e004, with part funding under Intersect and the Australian Research Council Linkage Infrastructure scheme.
Joshua Brown received his Ph.D. in Computer Science in 2019 from Charles Sturt University, Bathurst, Australia. His research interests are computational modelling, optimisation, and information theory.
References (23)
- et al.
Traveling bands of chemotactic bacteria: a theoretical analysis
J. Theoret. Biol.
(1971) - et al.
Collective motion
Phys. Rep.
(2012) - et al.
Metric–topological interaction model of collective behavior
Ecol. Modell.
(2011) - et al.
Review of data structures for computationally efficient nearest-neighbour entropy estimators for large systems with periodic boundary conditions
J. Comput. Sci.
(2017) Schooling fishes: The school, a truly egalitarian form of organization in which all members of the group are alike in influence, offers substantial benefits to its participants
Am. Sci.
(1978)- et al.
Novel type of phase transition in a system of self-driven particles
Phys. Rev. Lett.
(1995) - et al.
Flocking and turning: a new model for self-organized collective motion
J. Stat. Phys.
(2015) A mathematical theory of communication
Bell Syst. Tech. J.
(1948)- et al.
Information flow in a kinetic Ising model peaks in the disordered phase
Phys. Rev. Lett.
(2013) - et al.
Anomalous behaviour of mutual information in finite flocks
Europhys. Lett.
(2018)
Information flow in finite flocks
Sci. Rep.
Cited by (5)
Enhancing synchronization of self-propelled particles via modified rule of fixed number of neighbors
2023, Physica A: Statistical Mechanics and its ApplicationsEffects of adaptive acceleration response of birds on collective behaviors
2022, Journal of Physics: Complexity
Joshua Brown received his Ph.D. in Computer Science in 2019 from Charles Sturt University, Bathurst, Australia. His research interests are computational modelling, optimisation, and information theory.
Terry Bossomaier is professor of computer systems at Charles Sturt University. His research interests range from complex systems to computer games. He has published numerous research articles in journals and conferences and several books, the most recent, on information theory appearing in December 2016.
Lionel Barnett is a member of the Sackler Centre of Conciousness Science at the University of Sussex, United Kingdom. He has a number of publications with particular interest in evolutionary computation and Granger causality. He is also co-author with Terry Bossomaier of a Transfer Entropy book published in 2016.