Mechanistic origins of scale-invariant spontaneous behavior: the neural substrate, the scaling exponent, and the cutoff time

The timing of spontaneous (task-free) animal and human behavior is scale-invariant: the distribution of rest-interval durations obeys a power law $p(t)\sim t^{-(\beta+1)}$ truncated at a macroscopic cutoff time $T$, with the probability of remaining at rest decaying as $t^{-\beta}$, generated by a nonstationary hazard rate $r(t)=(1/t)f(t/T)$. Proekt, Banavar, Maritan and Pfaff (PNAS 2012) showed that this form follows generically from separation between a microscopic timescale $\tau$ and an emergent macroscopic timescale $T$, without fine tuning, and confirmed it in continuously recorded mouse behavior ($\beta\approx0.7$, $T\sim1000$ s). What remains open is the mechanism: which physiological and neural processes constitute the elementary process producing the hazard form, what sets the exponent $\beta$ (evolutionary advantage versus dynamical/statistical constraint), what sets the cutoff $T$ (the conjectured metabolic origin and its $M^{1/4}$ body-mass scaling are untested), and how explicit tasks, social interaction, and multi-state kinetics such as sleep modify or destroy the scaling regime.

Background

Many natural behaviors are spontaneously intermittent: an animal in an unchanging environment alternates between locomotion and rest with no obvious characteristic timescale. When the intervals (dwell or rest times) between consecutive behavioral events are pooled, their distribution often follows a power law, $p(t)\sim t^{-(\beta+1)}$, meaning the temporal pattern looks statistically the same at every magnification — a property called scale invariance, familiar from equilibrium systems near critical points (continuous phase transitions). Power-law-timed behaviors had been reported for human communication and foraging (Lévy flights — random walks whose step lengths are drawn from a power law), and were widely interpreted as consequences of queuing priorities, resource distributions, or biological criticality.

Proekt, Banavar, Maritan and Pfaff (2012) reanalyzed spontaneous locomotion of nine mice recorded continuously for ~22 days each in a featureless cage, plus published human and Drosophila data. They observed that successive rest intervals are essentially uncorrelated, so the dynamics can be characterized by the hazard function $r(t)$ — the probability per unit time that the animal starts moving at time $t$, given that it has been at rest since $t=0$. Because $r$ has units of inverse time while its only available arguments are a microscopic timescale $\tau$ (below which rest cannot be defined), the elapsed time $t$, and a maximum macroscopic timescale $T$, dimensional consistency alone forces the form $r(t)=(1/t)f(t/T)$ whenever the timescales are well separated, $\tau\ll t<T$. For small $z$, $f(z)$ approaches the constant $\beta$, with a leading correction $\sim z^{\delta}$, $f(z)\approx\beta + z^{\delta}$ (the paper's example $f(z)=z$ is the case $\beta=0$, $\delta=1$), so the survival probability (the probability of still being at rest at time $t$) decays as a power law, $p_>(t)\sim t^{-\beta}$, and the rest-interval density correspondingly as $p(t)\sim t^{-(\beta+1)}$, modified near $t=T$ by an exponential ($\delta=1$) or stretched-exponential ($\delta<1$) cutoff; when $\beta=0$, the distribution retains only the characteristic timescale $T$. Thus scale invariance requires no fine tuning to a critical point: it is generic for any nonstationary Markov process with a large separation between microscopic and emergent macroscopic timescales. In the mice, maximum-likelihood analysis confirmed a power law up to a cutoff of order 1,000 s with $\beta\approx0.7$, strikingly close to the human value from phone records; the same statistics held at every cage location and during the initial habituation phase, and the scaling broke down during the light (sleep-dominated) phase, as a three-state generalization predicts. The authors conjectured that $T$ is set metabolically (an animal must eventually drink or eat), so that $T$ should scale across species as body mass to the quarter power, and that the activation of all behaviors reflects an elementary, undifferentiated process of the nervous system (related to generalized arousal, i.e., the nonspecific baseline activation of the brain and spinal cord that precedes any specific motivated behavior).

The paper explicitly left open five linked mechanistic questions: (i) which physiological and circuit-level processes set the cutoff time $T$; (ii) why the observed exponents fall in a narrow range (the paper notes distributions with $\beta+1<1$ are nonnormalizable while $\beta+1>3$ is hard to detect statistically, but whether exponents additionally reflect evolutionary selection is unknown); (iii) which neural substrate implements the nonstationary hazard and produces the small-$z$ coefficient $\beta$ of $f(z)$; (iv) whether the framework extends to behaviors driven by explicit tasks, social interaction, or structured environments, and how extrinsic priorities modify $f(z)$ and $T$; and (v) how multi-state kinetics modify or eliminate the approximate power-law regime (as already seen during sleep).

Problem Statement

Identify the biological mechanism behind the scale-invariant activation dynamics of spontaneous behavior established by Proekt et al. (2012). The mechanism sought is a physiological and/or neural-circuit account that (i) derives the nonstationary, scale-separated hazard rate $r(t)=(1/t)f(t/T)$ — where $r(t)$ is the probability per unit time of initiating activity at rest-duration $t$, and $f$ is a dimensionless scaling function of $z=t/T$ — from identified processes, including the elementary process in the nervous system that governs activation of behavior and the shape of $f(z)$ at small $z$; (ii) explains what sets the macroscopic cutoff time $T$, including a test of the conjectured metabolic determination (with $T$ scaling across species approximately as body mass to the quarter power); (iii) explains the observed narrow range of scaling exponents (e.g., $\beta\approx0.7$ for mouse and human rest intervals, $\beta+1$ predominantly between 1 and 3 across behaviors), adjudicating between evolutionary selection on exponent values and dynamical or statistical constraints; and (iv) delimits the framework's generality by quantifying how explicit task priorities, social interactions, structured environments, or multi-state kinetics (e.g., sleep) modify $f(z)$ and $T$ or eliminate the power-law regime. Such an account must be consistent with the spontaneous-behavior statistics already reported in the cited literature and yield at least one falsifiable prediction; the four requirements (i)–(iv) are independently refutable sub-claims.

References

1. Proekt A, Banavar JR, Maritan A, Pfaff DW (2012) Scale invariance in the dynamics of spontaneous behavior. Proceedings of the National Academy of Sciences 109(26):10564–10569. doi:10.1073/pnas.1206894109, https://doi.org/10.1073/pnas.1206894109
2. Melanson A, Mejias JF, Jun JJ, Maler L, Longtin A (2017) Nonstationary stochastic dynamics underlie spontaneous transitions between active and inactive behavioral states. eNeuro 4(2):ENEURO.0355-16.2017. doi:10.1523/ENEURO.0355-16.2017, https://doi.org/10.1523/ENEURO.0355-16.2017
3. Jones SA, Barfield JH, Norman VK, Shew WL (2023) Scale-free behavioral dynamics directly linked with scale-free cortical dynamics. eLife 12:e79950. doi:10.7554/eLife.79950, https://doi.org/10.7554/eLife.79950
4. Rabus A, Masoliver M, Gruber AJ, Nicola W, Davidsen J (2024) Non-trivial relationship between behavioral avalanches and internal neuronal dynamics in a recurrent neural network. Chaos 34(5):053104. doi:10.1063/5.0201838, https://doi.org/10.1063/5.0201838
5. Costa AC, Sridhar G, Wyart C, Vergassola M (2024) Fluctuating landscapes and heavy tails in animal behavior. PRX Life 2:023001. doi:10.1103/PRXLife.2.023001, https://doi.org/10.1103/PRXLife.2.023001
6. Bialek W, Shaevitz JW (2024) Long timescales, individual differences, and scale invariance in animal behavior. Physical Review Letters 132:048401. doi:10.1103/PhysRevLett.132.048401, https://doi.org/10.1103/PhysRevLett.132.048401
7. Bulanchuk P, Koay SA, Romani S (2025) A class of Markovian self-reinforcing processes with power-law distributions. arXiv:2510.19034, https://arxiv.org/abs/2510.19034
8. Nakamura T, Takumi T, Takano A, Hatanaka F, Yamamoto Y (2013) Characterization and modeling of intermittent locomotor dynamics in clock gene-deficient mice. PLoS ONE 8(3):e58884. doi:10.1371/journal.pone.0058884, https://doi.org/10.1371/journal.pone.0058884
9. Gu C, Coomans CP, Hu K, Scheer FAJL, Stanley HE, Meijer JH (2015) Lack of exercise leads to significant and reversible loss of scale invariance in both aged and young mice. Proceedings of the National Academy of Sciences 112(8):2320–2324. doi:10.1073/pnas.1424706112, https://doi.org/10.1073/pnas.1424706112
10. Yadav CK, Doreswamy Y (2017) Scale invariance in lateral head scans during spatial exploration. Physical Review Letters 118:158104. doi:10.1103/PhysRevLett.118.158104, https://doi.org/10.1103/PhysRevLett.118.158104
11. Abe MS (2020) Functional advantages of Lévy walks emerging near a critical point. Proceedings of the National Academy of Sciences 117(39):24336–24344. doi:10.1073/pnas.2001548117, https://doi.org/10.1073/pnas.2001548117
12. Muñoz MA (2018) Colloquium: Criticality and dynamical scaling in living systems. Reviews of Modern Physics 90:031001. doi:10.1103/RevModPhys.90.031001, https://doi.org/10.1103/RevModPhys.90.031001
13. Yadav CK, Doreswamy Y (2025) Brain and behavior share features of a critical state dynamics. BioSystems 256:105551. doi:10.1016/j.biosystems.2025.105551, https://doi.org/10.1016/j.biosystems.2025.105551
14. Gómez CM, Altahona MA (2026) Power law and competitive model for behavioral characterization of incompatible spontaneous behavior in Drosophila melanogaster. Behavioural Processes 238:105372. doi:10.1016/j.beproc.2026.105372, https://doi.org/10.1016/j.beproc.2026.105372
15. Grabenhorst M, Poeppel D, Michalareas G (2026) The anticipation of imminent events is time-scale invariant. Proceedings of the National Academy of Sciences 123:e2518982123. doi:10.1073/pnas.2518982123, https://doi.org/10.1073/pnas.2518982123