CALLIOPE · BCOM

This work presents a pedagogical treatment of linear differential operators as the foundational mathematical machinery underlying neural mass models, arguing that every synaptic filter in such models is formally equivalent to a mass-spring-damper system. Drawing on Appendix J of the Rosetta Stone of Neural Mass Models (arXiv:2512.10982), the exposition develops a unified framework spanning zeroth- through second-order operators, employing complementary Laplace and Green's function perspectives to illuminate both frequency-domain and time-domain behavior. The alpha kernel of the Jansen–Rit model is identified as the impulse response of a critically damped harmonic oscillator, with the "synaptic mass" parameter governing a genuine causal delay that directly controls oscillation frequency. The central question of when excitatory–inhibitory loops produce self-sustained oscillations is resolved through the classical Barkhausen conditions—requiring unity loop gain and 360° total phase accumulation—which are shown to unify Wilson–Cowan (first-order) and Jansen–Rit (second-order) dynamics within a single feedback-theoretic picture. The framework provides a principled, quantitative account of how synaptic time constants, operator order, and axonal conduction delays collectively determine the spectral properties of neural rhythms.

Every synapse in a neural mass model is secretly a mass-spring-damper, and that single insight unlocks a clean, quantitative theory of why brain rhythms oscillate at the frequencies they do.

The paper's central move is to reframe synaptic filtering — the process by which a presynaptic spike train gets smoothed into a postsynaptic potential — as a linear differential operator. That operator has two equivalent descriptions: a transfer function in the frequency domain (poles, gain, phase) and a Green's function in the time domain (how past inputs are weighted and summed). Neither is more fundamental; you use whichever is convenient. The payoff is that the same mathematical object that describes a mass on a spring with damping is the synaptic filter, exactly, not approximately.

The paper then climbs an "order ladder." A zeroth-order operator is just a gain — no memory, instantaneous. A first-order operator is the leaky integrator at the heart of Wilson–Cowan models: one time constant, one real pole, low-pass behavior. A second-order operator adds inertia, and this is where the physics becomes vivid. The critically damped second-order case produces the alpha kernel — the rise-then-decay waveform used in the Jansen–Rit model to represent postsynaptic potentials. The "synaptic mass" parameter $m$ is not metaphorical: it controls a genuine causal delay, pushing the peak of the response later in time as $m$ increases. More inertia means a slower buildup, which means a longer effective loop delay, which means lower oscillation frequency.

The oscillation question itself is resolved by the Barkhausen conditions, borrowed from classical control theory. A feedback loop oscillates when two things are simultaneously true: the total gain around the loop equals exactly one, and the total phase accumulated around the loop equals exactly 360°. The phase condition is the more constraining one. Two first-order (Wilson–Cowan-style) synapses in an excitatory–inhibitory loop can supply at most 180° of phase — not enough to close the budget at any finite frequency. You need either self-excitation, an explicit axonal delay, or an upgrade to second-order synapses. Second-order filters each contribute up to 180°, so a cascade of two (one excitatory, one inhibitory) can reach 360° on its own. This is why Jansen–Rit circuits oscillate without requiring self-excitation, while Wilson–Cowan circuits need extra ingredients.

The practical upshot is a principled handle on brain rhythms. Oscillation frequency is not a free parameter — it is set by the total effective delay around the E–I loop, which includes axonal conduction time plus the group delays introduced by each synaptic filter. Changing synaptic time constants, adding dendritic filtering, or switching from first- to second-order operators all shift that delay budget and therefore shift the oscillation frequency. The paper is explicitly pedagogical, drawn from an appendix of a larger "Rosetta Stone" paper (arXiv:2512.10982), and reads accordingly — it is a tutorial rather than a novel result, but the unification it offers is genuinely clarifying.