Thought Toys · Cycles & change · Exhibit 122

Why the cicadas wait a prime number of years.

Periodical cicadas spend thirteen or seventeen years underground, then come out all at once, in numbers no predator can eat through. The strange part is the number. A brood on a twelve-year clock, facing a predator whose own numbers peak every four years, walks into that predator at every single emergence it will ever have. Move the brood to thirteen and three emergences in four come up to an empty field. Then shift the predator by one year, and everything you just concluded turns out to be the wrong lesson.

A hundred and forty years, one row per cycle ● the brood emerges   ● predators peak   ● both at once

emergences eaten years between bad years factor shared predator's head start

your turn — drag The brood's life cycle from 12 to 13 and watch the red disappear

What you're seeing

The top strip is time. Amber diamonds are the years this brood comes up; cyan ticks are the years the predator's own population is at its peak. A red ring means both happened in the same year, and that emergence was eaten.

Set the brood to twelve years against a four-year predator. Every emergence is ringed. Not most — every one, for as long as both clocks keep running, because twelve is a whole number of fours and so every multiple of twelve is also a multiple of four. Now nudge the brood to thirteen. Three rings in four vanish. The predator did not change, the brood did not get faster or tougher, and the number of years underground went up by one.

The bar chart underneath is the whole landscape at once. Each bar is a life cycle the brood could have; its height is the share of emergences eaten. The tall bars are the cycles that share a factor with the predator's. And now press Shift the predator a year, which moves the predator's peak years without changing anything else.

Everything you just believed rearranges. The twelve-year brood, doomed a moment ago, is now eaten never — it steps over the predator's years for ever. The thirteen-year brood is still eaten one time in four, exactly as before. Shift again, and again: the prime never moves, and the composites keep flipping between catastrophe and complete safety.

The level cyan rule across the bars is the honest summary, and it is the whole exhibit. It marks each cycle's average over every head start the predator could have had, and it is flat. Every life cycle — prime, composite, 2 or 19 — is eaten exactly one time in p on average. Primes are not eaten less. They are eaten predictably. A brood that shares a factor is making a bet on where the predator happened to start: win it and you are untouchable, lose it and you are eaten at every emergence you will ever have, for ever, with no way back. A coprime brood does not take the bet. It accepts the average and is guaranteed it.

That is why this matters for an animal rather than only for a number. Evolution does not get to choose the predator's head start, and it cannot re-roll a bad one. A strategy whose payoff is the same on average but never catastrophic beats one that is sometimes free and sometimes fatal, because the fatal branch does not get a second turn.

Two more things to try. Press Add three rival predators: the advantage survives and shrinks honestly, from certain death to about 73% eaten, not to nothing — piling up enemies on different clocks is the strongest objection to the whole story. Then push the predator's cycle to 13 while the brood sits at 13. The prime brood is eaten every single time. So the mechanism was never primality. It was coprimality: sharing no factor with whatever you are running from. Being prime is simply the cheapest way to be out of step with everything smaller than you, and it stops working the instant something matches you exactly.

The rule, exactly. A brood emerging every L years and a predator peaking every p years land in the same year only at multiples of their lowest common multiple. So:

when both start in step: share eaten = gcd(Lp) / p      years between eaten emergences = lcm(Lp)

Give the predator a head start of φ years and this changes. Write g = gcd(Lp). The brood's emergence years only ever land on multiples of g in the predator's cycle, so:

eaten g/p of the time on p/g of head starts, and never on the rest   —   average over head starts = 1/p, for every L

An earlier draft of this page stated only the aligned case, and a review pushed on it correctly: stated that way the page implies primes are eaten less on average, which is false. They are not. The average is identical for every life cycle; coprimality buys zero variance, not a lower mean. The correction is now the point of the exhibit rather than a footnote to it.

The predator is treated as a switch that is either at a peak or not, and "eaten" means the emergence meets a peak. "Safest" everywhere on this page means one thing only: the lowest share of emergences meeting a peak, under this binary overlap model, against the predators currently set. The timeline above is a 140-year window; every percentage is computed over the full repeat cycle, which for some pairs is far longer than the window shows. Only the watched predator takes the head start — the three rivals stay aligned — so the button has exactly one meaning. Real periodical cicadas also use predator satiation — emerging in numbers so vast that the local predators are full long before the brood is gone — and the two ideas work together: satiation needs the whole brood to arrive on one day, which needs a shared clock, and a shared clock is what makes the arithmetic here apply.

Verified in node (improve/verify/122-cicada-primes.js, 17 checks, all exact rather than sampled — the first twelve before this page existed, the last five added the same day when the review above forced the correction):

  • The formula matches a year-by-year walk over the full repeat period, for all 253 pairs of L in 2–24 and p in 2–12, to one part in 1012.
  • The gap between eaten emergences is exactly lcm(L, p) across the same 253 pairs.
  • A 12-year brood against a 4-year predator is eaten at 100% of emergences; a 13-year brood at 25%. The gap between bad years goes from 12 years to 52.
  • For every predator cycle from 2 to 12, every life cycle that ties for safest is coprime to it — checked exhaustively, not argued.
  • Negative control one. Against a predator that peaks every single year, all 19 life cycles are eaten 100% of the time. The effect needs the predator to be periodic too; primality on its own buys nothing.
  • Negative control two, the one that corrects the usual telling. Against a 13-year predator, the prime 13-year brood is eaten 100% of the time and the composite 12-year brood only 7.7%. Primality is not the mechanism. Coprimality is.
  • Against a guild of 2, 3, 4 and 5-year predators a 13-year brood is still eaten 73.3% of the time — exactly 11/15 — while every composite cycle from 12 to 20 that shares a factor with the guild is eaten every time. The advantage is real and it is not "safe versus eaten".
  • 7, 11, 13, 17 and 19 all score identically against that guild. The arithmetic does not single out 13 and 17; something else does.
  • The head-start block, and the correction it forced. A coprime brood is eaten exactly 1/p at every possible head start — checked for every coprime pair with L up to 24 and p up to 12.
  • A brood sharing a factor g is eaten g/p on exactly p/g head starts and 0% on all the others. All or nothing.
  • Averaged over every head start, every life cycle from 2 to 24 is eaten exactly 1/p, across all 253 pairs. Identical means.
  • The variance across head starts is zero exactly when the two cycles are coprime — which is the mechanism, stated as narrowly as it is true.
  • Negative control three. With the predator one year out of step, the composite 12-year brood is eaten 0% and the prime 13-year brood 25%. At that head start the prime is the worse bet, which is precisely why this page cannot say primes are eaten less.

Which leaves the honest gap. This page explains why a periodical cicada's cycle should be prime. It does not explain why the surviving broods are 13 and 17 rather than 7 or 11, and neither does the arithmetic — the leading accounts add a minimum development time in a cold climate and, importantly, the avoidance of hybridising with neighbouring broods, since two broods on coprime cycles also rarely meet each other. The prime-cycle hypothesis (Lloyd and Dybas, 1966; Goles, Schulz and Markus, 2001; Yoshimura, 1997) is a live and contested idea, not settled fact, and the part that is not in dispute is the arithmetic above.

Also in Cycles & change: The epidemic threshold →

All 18 in Cycles & change
  1. 04Predator & prey
  2. 102Infinite fuel. Finite speed.
  3. 112The best moment to give up on a good patch
  4. 118The growth stops. The growing doesn't.
  5. 122Why the cicadas wait a prime number of years — you are here
  6. 16The epidemic threshold
  7. 17Compound interest
  8. 53A feedback loop that overshoots
  9. 54Why planets speed up near the star
  10. 55The enzyme that hits a ceiling
  11. 65No spike, no matter how long you wait
  12. 68A perfect engine still throws most of it away
  13. 71Squeeze a reaction and it pushes back
  14. 73The switch that won't switch back
  15. 75Below a threshold, a population can't come back
  16. 80A bigger dose doesn't get there faster
  17. 83Come too close and a moon becomes a ring
  18. 95A system about to collapse gets slow before it goes.

← the cabinet