A Rhythm Nobody Keeps Alone

A forest full of blinking lights poses two different questions. Why do the lights come on together? And why does the whole display come back when it does?

I had been treating those as one question. Give every firefly a little clock, let neighboring clocks adjust to each other, and eventually the woods acquire a beat. It is an appealing picture. For one North American firefly, it misses the most interesting part.

In the recorded conditions, an isolated Photinus carolinus did not reliably repeat whole bursts on the roughly twelve-second schedule of its gathered peers. The group had a regularity the solo performance lacked.

That does not mean the animal has no internal timing. It means we have to be careful about which interval we are explaining.

Two rhythms in one light show

The species produces short trains of flashes: individual flashes roughly half a second apart, then darkness, then another train. The spacing inside a train and the interval from one train's beginning to the next are different measurements.

In their 2021 field study, Raphaël Sarfati, Julie Hayes and Orit Peleg recorded the display in Great Smoky Mountains National Park with pairs of cameras. At low activity, flashes looked scattered in time. During denser activity, they clustered into synchronous flashes within recurring bursts. These were partial views of a much larger swarm, not a census of every insect in the forest. [1, Results and Fig. 1]

The more revealing comparison came from confinement experiments described in the team's later, corrected study. The researchers recorded males in a tent visually screened from the surrounding swarm. Ten solitary animals supplied the individual observations. Each still produced flash trains, but the waits between trains varied from seconds to tens of minutes. With groups of about fifteen to twenty, the collective intervals became much more regular, approaching the roughly twelve-second cycle outside—a period that varies with temperature. [2, Behavioral experiments; Materials and methods; 3, corrected dataset]

The number fifteen is not a universal law of fireflies. It is a result from that setup, with its particular insects, visibility and recording conditions. Nor does the solitary experiment prove an isolated animal could never produce periodic bouts under other conditions.

But the contrast changes the problem. These animals were not merely aligning twelve-second solo performances that continued unchanged when nobody else was there.

Let the first one start

The later paper proposes a simple mechanism. After a burst, each animal waits an irregular amount of time. The first one to start the next burst prompts the others to follow. After that shared burst, everyone begins waiting again.

Consider a deliberately invented example, not measured firefly data. Three animals would, left alone, wait fourteen, twenty-nine and eighty seconds. If the fourteen-second starter recruits the other two, the group does not wait for their average. It starts at fourteen seconds. Next round, a different animal may draw the shortest wait and initiate the burst.

Add more animals and there are more chances for someone to be ready relatively early. If the possible waits have a lower boundary, the group's interval can concentrate near that boundary even though no individual repeatedly keeps that schedule alone. The paper suggests physiological recovery could help set such a boundary; it does not identify a measured neural recovery clock. [2, Proposed principle and Eqs. 1–5]

This is not simply “more insects make more flashes.” Independent animals flashing whenever they happen to be ready would produce a busier jumble. In this model, following and resetting together turn a first arrival into the next shared beginning.

In the ideal model, leadership can change with every burst. No permanent conductor needs to own the tempo. The shared timing is selected through the interaction.

There are substantial assumptions tucked inside that elegant explanation. Each individual draws a new wait independently; the individuals have comparable waiting-time distributions; everybody can see and immediately follow the starter. The analytical calculation takes its input from measured, smoothed solitary-animal intervals. A more flexible computer simulation adds a fitted interaction strength and simplifies a whole burst to one flash. Neither version is an assumption-free reconstruction of the insect's nervous system. [2, Analytical framework; Agent-based simulation]

The forest complicates the answer

The 2021 cameras captured something more spatial than a bank of lights switching everywhere at once. On average, burst activation spread through the reconstructed swarm at about half a meter per second. There was no corresponding bulk movement of flashing insects along that path: the traveling thing was a relay-like pattern of activity, not the same luminous passengers flying across the scene. [1, Fig. 2 and accompanying Results]

The researchers proposed that terrain and vegetation shape a changing network of sight lines. Nearby peers may be hidden behind leaves; a more distant flash may be visible through a gap. The pattern supports that interpretation, but it is not a direct inventory of everything each insect saw. A forest is not an instantly connected tent. [1, Discussion]

The later paper also names an awkward limit of its own explanation. Some recorded solitary intervals were shorter than twelve seconds—down toward five or six. If an enormous group instantly followed the earliest possible starter every time, why wouldn't the forest settle on that faster cycle?

Perhaps some unusually short waits came from alarm or other behavior in confinement. Perhaps incomplete visibility and propagation delays prevent an early flash from immediately resetting everyone. Those are proposed explanations, not an experimentally settled choice between them. [2, Discussion]

The model supplies a mechanism that can generate the observed kind of regularity. It has not uniquely identified every biological step, or made the forest's complications disappear.

A necessary correction

The tent evidence has a publication history worth keeping in view. An October 2025 correction removed an inadvertently included LED-stimulation segment from one twenty-animal recording, replaced a trial with a nonstandard staging sequence, and restored three omitted solitary recordings. It also corrected a computational comparison that had displayed differences in medians instead of the intended statistical test. The authors report repeating the analyses and retaining the central finding. The notice superseded an earlier expression of concern. [3]

I am using that corrected account, not independently reanalyzing its raw recordings. The two papers also belong to the same connected research program; they are not two laboratories independently confirming the whole mechanism.

Those limits matter because the distinction is already strong enough without polishing it into certainty. A flash train can have internal rhythm while its recurrence remains irregular. A group can acquire a repeatable interval through how its members respond, rather than by averaging clocks they were already keeping.

The question I am left with is more specific than “How do fireflies synchronize?” When a first flash reaches some neighbors but not others, who follows—and how much of the forest's twelve-second beat lives in that delay?

Sources and study limits

Sarfati, R., Hayes, J. C. & Peleg, O. (2021). Self-organization in natural swarms of Photinus carolinus synchronous fireflies. Science Advances 7, eabg9259. Complete main text, main captions and methods read.
Sarfati, R., Joshi, K., Martin, O., Hayes, J. C., Iyer-Biswas, S. & Peleg, O. (2023; corrected version). Emergent periodicity in the collective synchronous flashing of fireflies. eLife 12, e78908. Complete main narrative, equations, methods and main captions read in the live corrected version.
Same authors (2025). Correction: Emergent periodicity in the collective synchronous flashing of fireflies. eLife 14, e109449. Main notice, dataset tables and stated figure/text corrections read. The corrected tables include recordings from 2020 and 2021.

No field observation, flash-video viewing, independent figure-image inspection, raw-data reanalysis or replication of the authors' computer simulations is claimed here. Supplementary movies and separately linked supplementary analyses were not studied. The three-wait example is illustrative arithmetic, not a prediction fitted to the animals.