Imagine a huge orchestra where every musician plays at their own tempo, yet together they create a single rhythm. Scientists studying the brain or laser networks often face a similar picture: thousands of neurons or oscillators each pulse on their own, but the whole crowd needs to be observed at once. To avoid drowning in details, they look for a way to describe not each instrument but the overall sound — the distribution of who is currently in which phase.

For a long time, one and the same family of smooth bell-shaped curves, the Lorentzian distribution, was used for this purpose. It resembles an ordinary Gaussian bell but with heavier tails: even far from the center, a noticeable probability remains. Why this one? It seemed like luck — a convenient mathematical form that fits neatly into equations. The authors of the new study showed that it is not about convenience but about the deep geometry of the problem.

They reformulated the dynamics of an entire network in the language of projective geometry — the science of what is preserved when we look at an object from different angles. It turned out that the time evolution of such a system is equivalent to transporting a distribution along a circle while preserving symmetry. And it is precisely the Lorentzian curves that are the only ones that do not change their type under such transport; they only shift and stretch. This is like carrying a handful of sand along a complex trough, and only sand grains of a very specific shape would not crumble into dust.

This is the key: the famous Ott–Antonsen and Montbrió–Pazo–Roxin simplifications, which reduce thousands of equations to just two, work not because the authors cleverly chose an approximation. They work because the geometry of the system itself dictates the only possible form of collective behavior: that very Lorentzian shadow of an invisible circle. Gaussian bells do not work here because they lack the required projective invariance: their shape would be destroyed under such transport, just as a handful of flour would scatter.