This paper studies how small random fluctuations inside interacting populations can create or destroy rhythmic activity. The authors start from a simple, well-known model of two neural populations — one excitatory (X) and one inhibitory (Y) — and add a third fluctuating species Z that stands for a mediator or chemical signal. They show, with math and simulations, that randomness coming from finite population sizes (demographic noise) can seed coherent, sustained oscillations of X and Y. The extra species Z can either boost or shut down those noise‑driven rhythms.

At a basic level the work compares two views of the same system. If all population sizes are taken to be infinitely large, the model becomes deterministic and settles to a steady state: the two neural populations sit at activity 1/2 and Z sits at αz/δz. In that deterministic case small deviations are damped and no sustained oscillations appear. But when population sizes are finite, internal randomness matters. The authors write the exact master equation for the stochastic process and then apply standard approximations (Kramers–Moyal expansion and a linear noise approximation) to derive Langevin equations — equations with explicit noise terms — that describe how fluctuations evolve.

Using that stochastic description they compute power spectra and run numerical simulations. These calculations show a clear peak in frequency space for the X and Y populations: that peak marks quasi‑cycles, sustained oscillations that are driven by noise rather than by an unstable deterministic cycle. The authors obtain an approximate formula for the peak frequency and show how it depends on model parameters such as the interaction strength r, the coupling γ, the birth/death rates αz and δz of the mediator Z, and the relative sizes (volumes) of the populations. They present examples where changing the properties of Z shifts the peak or removes it entirely, which demonstrates that Z can enhance or suppress the noise‑induced rhythms.