The study builds directly on Part I, where investigators reported that a gamma friction proportional to theta velocity produces measurable phase amplitude coupling (PAC) and that PAC strength is a closed-form function of a single dimensionless parameter, kappa. The authors asked a concrete dynamical question: when does that coupling make the gamma oscillator lose stability? The friction term acting on gamma is negative during part of each theta cycle. Short bursts of negative friction can inject energy. Yet in ordinary cycles, the energy gained while theta rises is returned while theta falls, and the system's overall growth over a theta period is fixed at the baseline decay by Liouville's formula. This bookkeeping of phase-space volume is central: it explains why negative instantaneous friction does not automatically imply runaway gamma activity.
The paper presents a simple, clear energy argument. For most theta cycles the net work done on the gamma oscillator cancels across the rising and falling segments. In those circumstances, gamma oscillations remain bounded and continue to decay at the baseline rate. The authors emphasized that Liouville's formula fixes the multiplicative growth factor over a theta cycle at that baseline, independent of the coupling—so long as gamma continues to oscillate throughout the cycle. Short, intuitive statement: negative friction during part of a cycle is not sufficient for instability; the timing relative to gamma oscillation matters.
Instability appears only in exceptional conditions: when a large theta cycle produces friction that exceeds the critical damping threshold and thereby stops gamma oscillating for part or all of that cycle. In that case the energy gained during the positive-friction segment is not returned during the falling phase because the oscillator has ceased oscillatory motion and cannot give back the energy in the same way. The authors quantified when such events tip the balance: if those rare large cycles occur often enough—about one in a hundred according to the study—then their net energy contribution can outweigh the decay seen in the far more frequent ordinary cycles. This frequency-of-extreme-events criterion yields an analytic stability boundary written as g* σx = 10, where σx denotes the typical theta amplitude determined via the fluctuation–dissipation theorem, and g* condenses the frictional coupling scale.
The team compared the analytic prediction to direct numerical stability analysis using Lyapunov scans. Data indicated agreement to within 6% for damping ratios in the range 0.0005 to 0.0055. The authors also noted a mean-square boundary that lies substantially lower in parameter space; however, that mean-square threshold is governed by extremely improbable events and therefore does not represent the typical onset of divergence. The paper links the proximity to the analytic boundary back to the same kappa parameter that sets PAC strength—larger kappa produces stronger PAC and places the system closer to instability.
The investigators tested external stimulation of theta at or below its own frequency. They observed that such stimulation increases gamma energy and raises PAC measures, and can drive very large gamma bursts or even divergence in the model. The authors were explicit that these driven responses did not produce self-sustained gamma activity in this simple two-oscillator framework. They wrote that a genuine transition to seizure-like persistent activity would require memory mechanisms or additional circuit elements that are not included in the current model.
The authors made clear several modeling boundaries. This is a minimal two-oscillator system with a velocity-dependent friction law; details beyond that form—specific biophysical substrates, heterogeneity, and explicit memory or plasticity mechanisms—were not modeled and therefore were not reported. The analytic boundary g* σx = 10 and the match to Lyapunov scans apply within the stated damping-ratio window; the source provided those numerical ranges but did not report broader parameter explorations. Open questions remain: how biological memory mechanisms or additional network structure would permit a sustained seizure-like transition is not addressed by the model and therefore remains unresolved. The authors provided code and supplementary material links for replication and further study.
Limitations recorded in the source: this work is a preprint and has not undergone peer review; the simple model lacks memory and other circuit complexities needed to claim relevance to clinical seizures. Future work, as implied by the authors, must add those mechanisms to test whether controlled stimulation can produce self-sustained pathological activity in biologically realistic networks.