An arXiv preprint describes a simulated flock that stayed almost fully aligned for long stretches before briefly falling into disorder when the timing of interactions varied. The pattern appeared in the model's dissipative-delay regime, which the study distinguishes from a conservative regime. Under conservative delays, including a constant delay, the same simulations instead showed rapid fluctuations between highly ordered and weakly ordered flocking states. The result is a model-based association between delay dynamics and collective order, not a demonstration that real animals, people or physical active matter behave this way.
That intermittent pattern is the central finding. The paper reports long-lived episodes of nearly complete polar order interrupted by brief disordering events, but it does not quantify the duration or frequency of those episodes. That leaves the result as a qualitative description of how the simulated trajectories unfold, rather than a numerical estimate of how often a flock would lose alignment. The supplied evidence also does not give a real-world event rate or effect size.
Inside the simulation
The researchers used a model of self-propelled particles in a two-dimensional unit square with periodic boundary conditions. Its default setup contained N=1000 particles. The analysis focused on collective motion with internal dynamics and delayed, Vicsek-like alignment, while the particle interactions were organized through a model network. These choices define the setting in which the reported pattern appeared; they do not show that the same behavior must occur in other systems.
To generate the trajectories, the equations were solved with the trapezoidal rule and the Lobatto IIIC method, with linear interpolation for delayed terms. Random constant initial histories were used, and transients were relaxed for at least 1,000 delay periods before the reported averages were calculated. Those details describe how the model was run; they do not add experimental measurements to the study.
How the delays were compared
To compare delay regimes, the reported scan varied the mean delay, tau0, while fixing the delay amplitude at A=0.9. The highlighted comparison used dissipative delays with A=0.9 and tau0=1, versus conservative delays with A=0 and tau0=1. The main readouts were the time-averaged flocking order, a measure of collective alignment, and the time-averaged synchronization error, a measure of mismatch in the particles' internal dynamics.
Stability was examined with a master-stability function, a calculation of whether perturbations away from synchronized flocking die out or grow across the network's transverse modes. In the paper's criterion, synchronized flocking was stable when Lambda(sigma mu_alpha) was below zero for every transverse mode. The analysis also examined recovery rates over ranges of flocking order to track what happened after the group became less ordered.
An intricate map of collective order
Across the delay scan, collective order depended on the delay parameters in a fractal way. For a general reader, the practical meaning is that the pattern was intricate rather than a single smooth change from disorder to order. The supplied analysis does not state exact plotted order values, so the result identifies the structure of the parameter dependence without providing a numerical effect size.
The delay space was classified using the Lyapunov exponent of the access map, a measure used here to classify the delay dynamics. Dissipative regions had lambda[R]<0, while conservative regions had lambda[R]=0. The text linked that classification to the fractal dependence of collective order. In the authors' interpretation, the temporal structure of the interaction delay is an organizing feature of the model's collective motion, but the evidence does not isolate delay modulation as the only possible route to intermittent flocking.
Under dissipative delays, the simulations produced long-lived intervals of nearly complete polar order, interrupted by short disordering events. The conservative-delay runs, including the constant-delay setting, showed rapid switching between highly ordered and weakly ordered states. The comparison was qualitative: the supplied text does not report exact fluctuation amplitudes or frequencies for either regime.
A feedback explanation
The proposed explanation is a feedback process inside the model. A transient loss of order reshuffles the particles' neighbors; those changed local connections are followed by restored synchrony, allowing the collective state to become ordered again. The paper presents this as a self-organized loop linking transient disorder, neighbor reshuffling and recovered synchrony. Because the mechanism comes from model analysis and simulations, it has not been independently tested in a physical active-matter system.
One recovery diagnostic pointed in the same direction. At flocking orders substantially below one, the conditional mean synchronization recovery rate became strongly negative. The exact rate values are not stated, so this finding describes the direction of the reported relationship, not its size. It is a model calculation about how the simulated system moves through different order levels, rather than a measured recovery time in an experiment.
Two kinds of chaotic motion
The researchers also studied particle mobility with the dynamics uncoupled. Headings in the turbulent-chaotic regime had a short-time diffusion coefficient that scaled as Dshort ~ 1/T, whereas the diffusion coefficient for laminar-chaotic headings approached a finite value as T increased. No fitted coefficients or uncertainty estimates were reported for this comparison. The mobility result was used to distinguish the two chaotic regimes within the model, not to quantify how far a real flock would travel.
Changing the interaction radius while keeping the remaining parameters fixed preserved a broad separation between the laminar-chaos and turbulent-chaos order. This provides a robustness check for that parameter, but the study did not quantify how the result changes with particle number, propulsion speed, turning time, coupling, nonlinear response or network density. Whether the feedback explanation survives those changes remains an open question.
Where the result holds—and where it does not
The study's boundaries are important. Its evidence consists of numerical trajectories, analytical reductions, parameter sweeps and uncoupled mobility calculations in the specified active-particle model. The setup uses a two-dimensional periodic domain, soft-disk repulsion and a particular network construction, while the stability analysis uses a fast-switching approximation that replaces the time-dependent network with its time average. These choices limit direct generalization beyond the modeled assumptions.
Conventional statistical uncertainty measures, confidence intervals, hypothesis tests and formal power calculations were not reported. Many exact outcome values were presented graphically rather than stated numerically in the text, and the duration and frequency of disorder episodes were also left unquantified. The reported conclusions should therefore be read as qualitative or model-based where the supplied analysis does not give a numerical effect or uncertainty estimate.
The document is an arXiv preprint, version 1, dated 20 August 2026. It offers a prediction generated by a simulated system, not evidence that intermittent flocking occurs in humans, animals or physical active matter. The next test would be whether synthetic or biological active-matter systems show the same pattern, and whether quasiperiodic, random or chaotic delay modulation changes the collective behavior.
Taken together, the paper's proposed idea is that time-varying delay structure can serve as a control parameter and organizing principle for collective motion. Its strongest claim is about the behavior of this specific set of equations: dissipative and conservative delay regimes were associated with different patterns of order. Whether that organizing principle transfers outside the model remains unresolved.
Paper data and sources
Original title: Intermittent Flocking and Fractal Collective Order Induced by Time-Varying Delays
Authors: David Müller-Bender, Rahil N. Valani
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text