Contribution
It gave Kelso’s observation of an involuntary switch between bimanual patterns a dynamical model at two levels. At the top, one variable, relative phase, moves in a potential whose shape changes with movement frequency. Beneath it, two coupled limit-cycle oscillators, one per hand, from which that equation is derived. After it, the switch was a bifurcation with testable consequences, and the paper names them (critical slowing, critical fluctuations) for later work.
Key insight
Two patterns of coordination are two minima of one potential in relative phase. As frequency rises and amplitude falls, the anti-phase minimum flattens and vanishes, and the system falls into in-phase. Lowering the frequency restores the minimum but not the occupancy, so the system stays in-phase. Which pattern is seen is decided by the shape of the landscape and where the system started, not by a controller choosing one.
Assumptions
- Phase is the order parameter. Defended on two grounds: it reflects the cooperation of the components, and it stays invariant across scalar changes in rate or force in many motor activities (p. 349). It is also assumed to change much more slowly than the components’ variables, the slaving condition of synergetics.
- Near-harmonic motion. x_j = r_j cos(ωt + φ_j), with amplitudes and phases varying slowly against ω (Eqs. 2.1–2.2).
- Periodicity and symmetry. V(φ + 2π) = V(φ) and V(φ) = V(−φ), the hands playing symmetric roles (Eqs. 2.5–2.6; footnote 5 allows asymmetry).
- Amplitude carries the control. b is tied to the movement amplitude r, which falls as frequency rises, so b/a falls (Eq. 2.18).
- Approximations in §3. The slowly varying amplitude and rotating wave approximations, equal and stable amplitudes r₁ = r₂, and (for convenience) γ ≪ ω.
- Noise. A transition from φ = π to φ = 0 “can be initiated only if fluctuating forces … are present” (p. 353); the source of the noise is left open.
Key results
- The data modelled (§2, p. 348). Kelso’s experiments (1981a, 1984) show “(i) the presence of only two stable phase (or ‘attractor’) states between the hands …; (ii) the abrupt transition from one attractor state to the other at a critical cycling frequency; (iii) beyond the transition, only one mode (symmetrical in-phase) is observed; and (iv) when cycling frequency is reduced, the system stays in the symmetrical mode”. The transition frequency in units of each subject’s preferred frequency was “constant for all subjects”, and friction changed both frequencies but not their ratio (p. 347). Fig. 1 shows one run. These are reported, not shown.
- The order-parameter equation (§2). φ̇ = −∂V/∂φ (2.4) with V = −a cos φ − b cos 2φ (2.7). Extrema at φ = 0, ±π, and at cos φ = −a/4b (2.15), the inner maxima. Bistable while |b| > |a|/4, monostable (in-phase only) when |b| < |a|/4 (2.16–2.17, 3.22). Fig. 5 plots V/a for a series of values of b/a.
- Hysteresis is a consequence. “When we decrease ω starting from high values, the system remains all the time in the φ = 0 state even if ω drops below ω_c. This ‘hysteresis’ phenomenon is well known” (p. 350).
- Component oscillators (§3). Each hand obeys ẍ + ε(ẋ² − ω₀²r₀²)ẋ + ax = 0 (3.6), whose amplitude falls as ω₀r₀/(√3 ω) (3.8). A linear coupling does not give the required potential; a nonlinear coupling of the same symmetry still “does not produce the correct phase relation”; the coupling must involve a time delay (3.11b) or time derivatives (3.25). With the derivative coupling K₁₂ = (ẋ₁ − ẋ₂)(α + β(x₁ − x₂)²) the phase equation is φ̇ = (α + 2βr²) sin φ − βr² sin 2φ (3.26), with critical amplitude r_c² = α/4β (3.27).
- Simulation (§4). Eq. (3.6) with coupling (3.25) and small noise, ω swept from 1.17 to 3.05, switches from out-of-phase to in-phase when ω exceeds a critical value (Fig. 6); “the exact form of the curves depends strongly on the noise level and the rate of changing ω”. Fig. 7 gives both modes’ steady amplitudes against ω, the out-of-phase branch unstable above ω_c.
- What is left for later (p. 355). “In future studies a number of phenomena known to accompany phase transition in synergetic systems (e.g., critical slowing down; critical fluctuations) will also be analyzed.” And: “it remains for further theoretical and experimental research to address how it is that only two stable modes emerge in the first place”.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Bimanual rhythmic movement has two stable phase relations and switches abruptly from anti-phase to in-phase at a critical frequency, with hysteresis | moderate, as reported | Kelso 1981a and 1984, summarized on pp. 347–348; one run shown (Fig. 1); no data in this paper |
| C2 | Relative phase is an order parameter for these patterns | moderate | argued from invariance and cooperativity (p. 349); fits the four features |
| C3 | V = −a cos φ − b cos 2φ is the simplest potential reproducing the features | strong, as mathematics | periodicity and symmetry plus the observed layout; §2 |
| C4 | The φ equation follows from two self-sustained oscillators only if their coupling is nonlinear and involves delay or derivatives | moderate | derivation under the stated approximations, §3; “in all likelihood the choice of f₁ and f₂ is not unique” (p. 351) |
| C5 | The transition occurs at a critical frequency via the amplitudes | weak; a prediction | Eqs. 3.24, 3.27; stated as “now open to further experimental test” (p. 355) |
| C6 | The neural coupling for voluntary bimanual movement may be subcortical | speculation | split-brain patients’ bias toward synchrony or alternation (Tuller and Kelso 1984, a conference paper) |
| C7 | Nonlinearity gives a stable amplitude–frequency coupling and a means of switching, hence “stability and flexibility of motor function” | asserted | hypothesis, p. 355 |
Method
Synergetics’ strategy: identify an order parameter where behaviour changes qualitatively, write the simplest potential consistent with the observed attractor layout and its symmetries, then derive that order-parameter equation from equations for the components and their coupling, and check the whole by numerical simulation (fourth-order Runge–Kutta with small random fluctuations).
Concepts
- Order parameter — a macroscopic variable “created by the cooperation of the individual components” which “in turn governs the behavior of these components”, and which changes much more slowly than they do (pp. 348–349).
- Phase — used in two senses, warned of in footnote 3: a temporal relation between the hands, and a state of aggregation, as in “phase transition”; here the transition is between behavioural modes defined by the first sense.
- Control parameter — movement frequency ω, entering through the coefficient a of the restoring force and, via amplitude, through b/a.
- Hysteresis — staying in the in-phase state when ω is lowered back below ω_c, from the coexistence of two basins of attraction.
Connections
It applies Haken’s synergetics (laser theory, the slaving principle) to Kelso’s bimanual experiments, and compares the switch with gait transitions in the decerebrate cat (Shik et al. 1966) and Grillner’s “switch mechanism”. Its noise term is the opening that Schöner, Haken and Kelso’s 1986 stochastic theory (Biological Cybernetics 53:247, not filed) fills, and the critical fluctuations it promises are reported in Kelso, Scholz and Schöner 1986 (LIT-1851, NOTE-1433). Kelso’s 2021 retrospective (LIT-601, NOTE-467) restates this model as the “top tier” above the coupled components, and Dynamic Patterns (LIT-578) is the book-length treatment.
Bearing on the record
- THEORY-069: partly. The paper is the model’s original statement,
and it confirms THEORY-069’s equations, the b/a = 1/4 threshold, the
direction of the switch (anti-phase to in-phase as frequency rises),
and the derivation from coupled oscillators. It also states hysteresis
as an observed feature of Kelso’s experiments (p. 348) and derives it,
which corrects THEORY-069’s line that “the read text does not report
the hysteresis data”: this text reports it, though it does not show the
data. It does not meet the condition that matters most in the
promote_when: it reports neither critical fluctuations nor critical slowing, naming both as future work (p. 355). The 1985 paper is a theory paper; for first-hand data see NOTE-1433 and NOTE-1449. - THEORY-069’s
promote_whennames Schöner, Haken and Kelso (1986). That is the stochastic theory in Biological Cybernetics 53:247, still not filed and not read. It is a different paper from Kelso, Scholz and Schöner’s Physics Letters A report of the same year (LIT-1851). - LIT-601 and NOTE-467. Kelso’s 2021 account of this paper is faithful on the model. One small difference: he speaks of hybrid oscillators; this paper uses a Van der Pol variant (Eq. 3.6) and a superposition with a Van der Pol term (Eq. 3.9), and notes the choice “is not unique”.
- No instruction for machine-learning practice; nothing for the anthology.
Limitations
- No new data. Every empirical feature is summarized from Kelso’s earlier work, two of whose sources are a one-page abstract (1981a) and an invited talk (1982). How many subjects showed hysteresis, and how it was tested, is not given here.
- Model choice. The potential is the simplest that fits, and the oscillator and coupling forms were chosen among several; the authors say the choice is not unique. The simulation reaches no stationary oscillation, so its curves depend on noise level and sweep rate.
- Amplitude as mediator. Tying b to amplitude predicts that the transition happens at a critical amplitude; the paper offers this as a test, not a result.
- Physiology. The authors say the physiological basis of the nonlinearities is “an open issue”; the subcortical coupling is speculation.
Open questions
- Do critical fluctuations and critical slowing precede the switch, as the synergetic analogy predicts? (Fluctuations: LIT-1851. Relaxation time: Scholz, Kelso and Schöner 1987, not filed.)
- Is the transition reached at a critical amplitude, as Eqs. 3.24 and 3.27 imply?
- Why only two stable modes “emerge in the first place from a wealth of possibilities” (p. 355)?