When Musicians Lock In: Coupled Oscillators and the Physics of Ensemble Synchronisation
Abstract
The Kuramoto model explains how heterogeneous phase oscillators can develop collective synchrony. It is a useful analogy for ensemble timing, applause, and delayed feedback, but it does not by itself predict a universal networked- music latency limit or prove neural coupling between musicians.
Contents
The problem is ancient and the language for it is recent. In an ensemble, musicians continuously predict, listen, and adjust. What does a model of phase oscillators reveal about that process, and what does it erase?
This post extends the latency discussion in Latency in Networked Music Performance. The bridge is explicitly a model, not a claim that musicians are identical clocks.
Two Clocks on a Board
In February 1665 Christiaan Huygens noticed that two of his new pendulum clocks, hung a foot or two apart, beat together without ever drifting. The Journal des Sçavans account of 16 March, drawn from his letter to his father, calls it “une espece de sympathie” (“a kind of sympathy”) [[1]]. The pendulums, it says, did not swing in parallel but “par des mouvemens contraires”, in a stable anti-phase relation. Huygens first blamed the air; on 6 March he named a slight shaking (branslement) of the two chairs on which both clocks rested [[2]].
A modern version of the experiment shows that the outcome is not fixed. With two metronomes on a freely moving board, Yang and colleagues obtained in-phase and anti-phase locking about equally often from random starting phases when the metronomes’ natural frequencies differed by 3 mHz, and anti-phase in 98 of 100 runs at 14 mHz [[3]]. Anti-phase synchrony is not an unconditional law of two pendulums.
The scene supplies the useful abstraction. Each unit has a phase and a tendency to advance at its own rate. Interaction changes those phases. An ensemble adds perception, anticipation, hierarchy, expressive timing, and learning, none of which is contained in a wooden beam.
The Kuramoto Model
For $N$ globally coupled phase oscillators, the standard Kuramoto equation is [[4]]
$$\frac{d\theta_i}{dt}=\omega_i+\frac{K}{N}\sum_{j=1}^{N} \sin(\theta_j-\theta_i). \tag{1}$$Here $\theta_i$ is oscillator $i$’s phase, $\omega_i$ its natural frequency in $\mathrm{rad\,s^{-1}}$, and $K$ a uniform coupling strength in the same unit. The $1/N$ scaling keeps the mean-field coupling finite as the population grows.
For musicians, one may interpret phase as position within a pulse cycle and coupling as timing adjustment from perceived partners. That interpretation is not a fitted model of a quartet. Real interaction is directional and delayed; players have roles, memory, multiple timescales, and non-sinusoidal responses.
The complex order parameter
$$r e^{i\psi}=\frac{1}{N}\sum_{j=1}^{N}e^{i\theta_j} \tag{2}$$summarises phase concentration. $r=1$ means identical phases. Values near zero can indicate phases spread around the circle, although a finite ensemble rarely gives exactly zero and the same $r$ can hide different phase patterns. For $N$ independent, uniformly random phases, $r$ is of order $N^{-1/2}$ [[5]]: on average about 0.45 for four players and 0.09 for a hundred (my calculation). Multiplying Eq. (2) by $e^{-i\theta_i}$ and taking the imaginary part turns the sum in Eq. (1) into $r\sin(\psi-\theta_i)$, so
$$\frac{d\theta_i}{dt}=\omega_i+Kr\sin(\psi-\theta_i). \tag{3}$$This reduction is exact for the globally coupled equation. Its musical meaning remains conditional on the analogy.
The Onset of Collective Synchrony
In the infinite-population model with a smooth, symmetric, unimodal frequency density $g(\omega)$ centred at zero (a frame rotating at the centre frequency puts it there), incoherence loses stability at
$$K_c=\frac{2}{\pi g(0)}. \tag{4}$$In Kuramoto’s argument as Strogatz retells it [[5]], a steady state with constant $r$ and $\psi=0$ locks the oscillators with $|\omega_i|\le Kr$ at $\sin\theta_i=\omega_i/(Kr)$; the faster ones drift and, by the symmetry of $g$, cancel in the order parameter. The locked ones give
$$r=Kr\int_{-\pi/2}^{\pi/2}\cos^2\theta\,g(Kr\sin\theta)\,d\theta.$$Its non-zero branch starts where $r\to0^+$, which gives Eq. (4). The next order gives, for small positive $\mu=(K-K_c)/K_c$,
$$r\approx\sqrt{\frac{16\,\mu}{\pi K_c^{3}\,[-g''(0)]}}.$$The coefficient depends on the curvature of $g$ at the peak; for any Gaussian it is $\sqrt{\pi}\approx1.77$.
The assumptions matter. Finite populations fluctuate; other frequency distributions, coupling networks, inertia, noise, or higher harmonics can alter the transition.
For the Lorentzian density
$$g(\omega)=\frac{\gamma/\pi}{\omega^2+\gamma^2},$$with half-width $\gamma$, the integral can be done exactly [[4]] [[5]]: $g(0)=1/(\pi\gamma)$ gives $K_c=2\gamma$ and, on the coherent branch,
$$r=\sqrt{1-\frac{K_c}{K}}. \tag{5}$$Its curvature, $g''(0)=-2/(\pi\gamma^3)$, makes the coefficient above exactly 1, so $r\approx\sqrt{\mu}$ near threshold, for the Lorentzian only.
Fig. 1 sets Eq. (5) against simulations of Eq. (3).
Eq. (5) is useful precisely because its scope is narrow. Sharing a mean-field exponent with other mean-field models does not establish that a musical ensemble and a ferromagnet belong to one experimentally demonstrated universality class.
Rhythmic Applause
Néda and colleagues measured concert-hall applause and found alternation between unsynchronised, fast clapping and synchronised, slower clapping [[6]]. The key change was period doubling: individuals lengthened their clapping period, narrowing the frequency distribution enough for audible synchrony to emerge. In their solo recordings of 73 students, the “synchronised” style had roughly half the frequency of the enthusiastic one and a spread 2.5 times smaller. Since $K_c$ in Eq. (4) scales with the width of $g$, the narrowing lowers the threshold. As people returned to faster, louder clapping, the distribution broadened and synchrony was lost. That account fits a Kuramoto-style competition between frequency spread and coupling.
Applause is still not ensemble performance. Audience members aim at a collective pulse without preserving a score, leader relation, articulation, or expressive microtiming. It is evidence that human rhythmic action can exhibit a synchronisation transition, not a validation of one parameter set for all music.
Delay Changes the Model
A uniform-delay extension of Eq. (1) is
$$\frac{d\theta_i}{dt}=\omega_i+\frac{K}{N}\sum_{j=1}^{N} \sin\!\left[\theta_j(t-\tau)-\theta_i(t)\right]. \tag{6}$$For identical oscillators, $\omega_i=\omega$, Yeung and Strogatz showed that Eq. (6) has in-phase states $\theta_i=\Omega t$ with
$$\Omega=\omega-K\sin(\Omega\tau), \tag{7}$$stable when $\cos(\Omega\tau)>0$ [[7]]. A duo in which each player hears only the other, with gain $K$, is Earl and Strogatz’s case of one delayed input per oscillator [[8]]. Its in-phase states obey the same Eq. (7), and perturbations that pull the two apart decay whenever $K\cos(\Omega\tau)>0$ (my linearisation, confirmed by simulation). Each player corrects towards where the partner was $\tau$ ago, which is behind where the partner is now, so the corrections pull the common tempo down: for small lags $\Omega\approx\omega/(1+K\tau)$. Fig. 2 shows the branches.
The one-way audio latencies in the LoLa and MVTP measurements ranged from 7.5 to 22.5 ms: at 120 BPM, 5.4° to 16.2° of a 500 ms beat. In this model such delays never destroy the in-phase state, since for $\omega\tau<\pi/2$ Eq. (7) always has a stable root with $\Omega\tau<\omega\tau$. They only slow the ensemble, by an amount that depends on the unknown $K$: 6.5 % at 22.5 ms for the $K=\omega/4$ of Fig. 2.
The clapping experiments partly agree. Chafe, Cáceres and Gurevich had pairs clap together from separate rooms with one-way delays of 3 to 78 ms [[9]]. The most stable performances came at “natural” delays, like the travel times of sound within an ensemble. Longer delays brought lag, increasingly severe deceleration and then deterioration; the shortest brought a tendency to accelerate, which the authors attribute to anticipation. With positive $K$ the in-phase tempo of Eq. (7) never exceeds $\omega$, so it has no sweet spot between the regimes. Its locked state survives all their delays (78 ms is at most 0.77 rad at the 86–94 BPM starting tempi of their 2009 preprint), so it predicts no deterioration; and it predicts a fixed lower tempo, where the preprint measures a drift in BPM per second. The model gets the sign of the main effect and misses the anticipating player.
Stability is lost only further out. For $K<\omega/2$ no stable in-phase state exists for $\pi/[2(\omega-K)]<\tau<3\pi/[2(\omega+K)]$ [[7]], the gap in Fig. 2; as $K\to0$ the gap opens at a quarter period, 125 ms at 120 BPM.
For distributed frequencies the threshold itself moves. For a Lorentzian centred at $\bar\omega$, incoherence loses stability at [[7]]
$$K_c(\tau)=\frac{2\gamma}{\cos(\Omega_c\tau)},\qquad \Omega_c=\bar\omega-\gamma\tan(\Omega_c\tau),$$with the root that gives the smallest positive $K_c$. Setting $\Omega_c=\bar\omega$ gives $K_c(\tau)=K_c(0)/\cos(\bar\omega\tau)$, which diverges at a quarter period and seems to promise a collapse delay. The exact threshold stays finite, because the implicit equation always has a root with $\cos(\Omega_c\tau)>0$; with $\bar\omega=10\gamma$ it is $2.65\,K_c(0)$ at a quarter period and peaks at $5.65\,K_c(0)$ at half a period (my evaluation). These are thresholds for an infinite population leaving incoherence, not for a duo. An earlier version of this post read a 125 ms threshold for musicians at 120 BPM off that divergence. I have withdrawn it: the exact threshold has none.
Networked-performance tolerance depends on task, tempo, articulation, feedback mix, direction and variability of delay, strategy, and what counts as acceptable coordination. Faster pulse rates make a fixed delay occupy a larger phase fraction, but “doubling tempo halves the usable network latency” is not a universal law.
The model still earns its place here. It shows why delay cannot be treated as mere lateness: in a feedback system, delay changes the phase of the information used for the next correction.
What EEG Synchrony Does Not Prove
EEG hyperscanning studies have reported statistical phase relations between signals recorded from musicians during coordinated performance. Lindenberger et al. recorded eight pairs of guitarists playing a short melody together [[10]]; Müller et al. analysed intra- and inter-brain network measures in eight pairs improvising on the guitar [[11]].
“Inter-brain synchrony” is a property of an analysis, not direct observation of one brain driving another. Shared sound, score, movement, tempo, sensory input, filtering, reference choice, volume conduction, and analysis parameters can produce or inflate aligned signals. Controls and surrogate analyses determine which interpretations remain plausible.
The findings therefore do not establish that a Kuramoto equation operates at a second neural level, nor whether neural alignment causes behavioural synchrony. Lindenberger and colleagues say so themselves: “Whether between-brain oscillatory couplings play a causal role … needs to be clarified by further research” [[10]]. The studies motivate that question. Demos and Palmer argue that pairwise models of ensemble synchrony miss “shifts in members’ influence” and emergent roles that “differ from individual or pairwise behaviors” [[12]]. Eq. (1), with every pair coupled alike, has no place for such roles.
What the Analogy Is Good For
The model gives rehearsal language a sharper edge without translating it literally.
“Listen more” can mean increasing responsiveness to partners, but responsiveness has gain, direction, delay, and perceptual limits; it is not one scalar $K$. Tempo negotiation can reduce disagreement before performance, but musicians do not possess fixed natural frequencies sampled once from $g(\omega)$.
A click track supplies common external forcing rather than only peer-to-peer coupling. It can stabilise a reference pulse, yet it does not eliminate group dynamics or guarantee identical phase. Expressive timing can coexist with a click, and its absence cannot be inferred from the equation.
Artificial delay can be a revealing supervised exercise when hearing safety, latency calibration, consent, and equipment limits are controlled. The lesson is not to hunt for 125 ms. It is to observe how a delayed feedback channel changes correction strategies and then ask which part a phase model captures.
The clocks on Huygens’s board did not listen. Musicians do. That difference is the boundary of the analogy and the reason the mathematics remains interesting.
Literature and calculations checked through 2026-10-24.
References
[1] Huygens, C. (1665). Letter to Constantijn Huygens, 26 February 1665 (summary), with the extract printed in the Journal des Sçavans, No. XI, 16 March 1665. Letter No. 1335 in D. Bierens de Haan (Ed.), Œuvres complètes de Christiaan Huygens, Vol. 5 (1893), pp. 243–244. DBNL
[2] Huygens, C. (1665). Letter to R. Moray, 6 March 1665. Letter No. 1345 in D. Bierens de Haan (Ed.), Œuvres complètes de Christiaan Huygens, Vol. 5 (1893), pp. 255–256. DBNL
[3] Yang, J., Wang, Y., Yu, Y., Xiao, J., & Wang, X. (2018). Huygens’ synchronization experiment revisited: Luck or skill? European Journal of Physics, 39(5), 055004. DOI: 10.1088/1361-6404/aac98e
[4] Kuramoto, Y. (1975). Self-entrainment of a population of coupled non-linear oscillators. In H. Araki (Ed.), International Symposium on Mathematical Problems in Theoretical Physics, Lecture Notes in Physics 39, 420–422. DOI: 10.1007/BFb0013365
[5] Strogatz, S. H. (2000). From Kuramoto to Crawford: Exploring the onset of synchronization in populations of coupled oscillators. Physica D, 143(1–4), 1–20. DOI: 10.1016/S0167-2789(00)00094-4
[6] Néda, Z., Ravasz, E., Vicsek, T., Brechet, Y., & Barabási, A.-L. (2000). Physics of the rhythmic applause. Physical Review E, 61(6), 6987–6992. DOI: 10.1103/PhysRevE.61.6987
[7] Yeung, M. K. S., & Strogatz, S. H. (1999). Time delay in the Kuramoto model of coupled oscillators. Physical Review Letters, 82(3), 648–651. DOI: 10.1103/PhysRevLett.82.648
[8] Earl, M. G., & Strogatz, S. H. (2003). Synchronization in oscillator networks with delayed coupling: A stability criterion. Physical Review E, 67(3), 036204. DOI: 10.1103/PhysRevE.67.036204
[9] Chafe, C., Cáceres, J.-P., & Gurevich, M. (2010). Effect of temporal separation on synchronization in rhythmic performance. Perception, 39(7), 982–992. DOI: 10.1068/p6465
[10] Lindenberger, U., Li, S.-C., Gruber, W., & Müller, V. (2009). Brains swinging in concert: Cortical phase synchronization while playing guitar. BMC Neuroscience, 10, 22. DOI: 10.1186/1471-2202-10-22
[11] Müller, V., Sänger, J., & Lindenberger, U. (2013). Intra- and inter-brain synchronization during musical improvisation on the guitar. PLOS ONE, 8(9), e73852. DOI: 10.1371/journal.pone.0073852
[12] Demos, A. P., & Palmer, C. (2023). Social and nonlinear dynamics unite: Musical group synchrony. Trends in Cognitive Sciences, 27(11), 1008–1018. DOI: 10.1016/j.tics.2023.05.005
Changelog
- 2026-01-14: Corrected the Demos and Palmer author list and changed an incorrect “period-doubling” edit to “frequency-doubling.”
- 2026-07-11: Restored period doubling in the applause result; bounded the critical-coupling formula; removed a false universal 125 ms latency threshold, ferromagnet universality claim, and causal neural-coupling inference; and made the musician mapping explicitly analogical.
- 2026-10-03: Corrected the Demos and Palmer (2023) DOI to 10.1016/j.tics.2023.05.005 (the previous DOI resolved to an unrelated paper) and fixed a broken
\Omegain the delayed-coupling paragraph. - 2026-10-24: Corrected the delay section: the formula $K_c(\tau)=K_c(0)/\cos(\bar\omega\tau)$, with $\bar\omega$ undefined, is now shown as an approximation to Yeung and Strogatz’s exact Lorentzian threshold, which stays finite where the approximation diverges (the origin of the withdrawn 125 ms); the paragraph in which
\Omegastill rendered as “Omega” after 2026-10-03 was rewritten. Added the delayed in-phase condition and its stability, the NMP latencies as beat phase, Chafe et al.’s clapping duos, the derivation of $K_c$, Figs. 1–2, and sources for the Huygens account and the metronome experiment; narrowed the Demos and Palmer summary to their abstract. Moved here from the body: before 2026-07-11 the post gave $r\approx\sqrt{(K-K_c)/K_c}$ as universal (it holds for the Lorentzian only) and called frequency doubling a nonlinear instability that restores coherence in applause (Néda et al. report period doubling into synchrony).