A hundred and twenty bees, one comb, and a meadow that keeps changing.
Every figure is counted from the run in front of you. Settings are listed in About.
A cut-away of one comb, with the entrance below it and a meadow beyond. Try a hot day, a shower, a new bloom, a hornet. Watch what the bees on the comb above the entrance do when a rich patch opens, and what the edge of the comb does at the lower left.
Comb. Workers of middle age hang in chains, sweat wax and build with it. Cells begin as cylinders round the builders’ bodies and become six-sided as the warm wax flows and the walls pull straight, meeting three at a time at 120° (Pirk, Hepburn, Radloff & Tautz 2004; Karihaloo, Zhang & Wang 2013). Some researchers hold that bees shape flat walls directly (Bauer & Bienefeld 2013), so this is one account, not a settled one.
The dance. A forager back from good flowers runs a short straight line on the comb, waggling. The angle of that line from straight up is the bearing of the flowers from the sun. The time each run lasts grows with the distance, roughly a second a kilometre (von Frisch 1967; Seeley 1995). Better flowers earn more runs. Followers read the dance with some error and many fail to find the place.
Choosing between patches. No bee compares patches. Each forager dances more for a better patch and gives up a poor one sooner, and that is enough for the colony to move its foragers to the best flowers and to follow a change (Seeley, Camazine & Sneyd 1991; Seeley 1995). A forager kept waiting to unload trembles instead of dancing, which calls up more receivers (Seeley 1992). Foragers who fared badly at a patch butt the dancers for it and cut their dances short (Nieh 2010).
Who does what. Work goes roughly with age: nursing, then receiving and building, then guarding and foraging (Seeley 1982; Robinson 1992). In the response-threshold model each bee answers a stimulus, such as hungry brood, when it is strong against her own threshold, and doing the work brings the stimulus down (Bonabeau, Theraulaz & Deneubourg 1996).
Defence. Guards at the entrance release an alarm scent, chiefly isopentyl acetate, which brings out more defenders (Boch, Shearer & Stone 1962).
All references are cited from memory; check before quoting. Pollen colours are approximate. The hum is brown noise filtered near 234 and 602 Hz, taken from the earlier version of this page.
Each bee keeps a count for every kind of work other than her own. The count rises with how much more strongly that work calls her than her present work does, using the Bonabeau response and her own threshold. At a fixed count she switches, and all her counts start again. They also start again whenever she finishes a piece of work. A forager gives up a patch in the same way: she adds up how far each trip fell short, and leaves at a fixed total. A guard counts what she sees of the hornet before she releases scent. The glow on the meadow is the colony’s forage map: a sum of recent waggle runs at the places they advertised, fading over about a minute.
The timing rule here, a count since the last event, a fixed threshold, a restart, and a memory made only of the past, is that of CRR, a research programme under investigation. The bee biology is not CRR’s. Bonabeau’s model already has nearly this shape, though there a bee switches by chance at each moment and here she counts. One candidate value from the programme for the spread of intervals is 0.159; the page shows the spread it measures beside that number and nothing was adjusted towards it.
Wall junctions move along the net pull of their three walls while the hive is warm. The readout gives how far junction angles are from 120° on average. The mean angle itself is always exactly 120°, since three angles round a point make 360°, so it would tell you nothing.
120 bees. Switch threshold 1, rate 0.1 a second. A forager counts shortfall below a profit of 0.8 and leaves at 1.5. Up to 8 waggle runs a dance; a follower needs 3. Dancer scatter 8° and 8%; follower error 10° and 12%. Nectar quality: phacelia 0.9, knapweed 0.8, white clover 0.7, dandelion 0.55; a bloom from the button 0.95. Patches last four to eight minutes. Flight takes 3 seconds plus one for each 200 metres, which is about nine times faster than life. Wax for one deposit takes 20 seconds; a cell takes four deposits. The sun stands still at 22° left of centre.
The queen, drones, eggs, brood growing up, bees ageing, death, swarming, robbing, disease, and night. The sun does not move, though real dancers allow for it. Distances and times are compressed. The hornet takes no bees and no bee stings or dies. Stop signals are given only after a poor trip. A real colony has tens of thousands of bees and many combs; this one has a hundred and twenty and one.
The bets were written down before the model was run. Five random starts each.
| Bet | Result |
|---|---|
| Of two patches, most foragers go to the richer within 5 minutes, and move when the two are swapped | Held, 5 of 5. Share on the richer patch 98 to 100%; six minutes after the swap, 83 to 97%. |
| New junctions settle to within 2° of 120° | Held. 5.3° off when formed, 0.2° two minutes later. With the flow switched off they stayed 6.7° off. |
| Without the dance, at least a fifth less nectar comes in | Held in the final model: 30% less. In my first version the fall was 16% and the bet failed. |
| Alarm at least doubles the guard, and the guard stands down after | Failed, 2 of 5. The hornet was driven off in about 30 seconds every time and the guard stood down, but the guard doubled in only two runs. |
| A hot day brings at least three more fanners and the hive stays under 36.5 °C | Held, 5 of 5. Fanners rose from about 1 to between 8 and 38; the number overshoots in some runs. |
| Spread of task-switch intervals (no pass mark) | CV 0.49 to 0.94 over fifteen minutes, from 26 to 52 intervals a run. The candidate value is 0.159. They do not agree. |
Between the first and final versions I changed the switching rule because whole groups of bees were changing work at once: the response was made to saturate, bees were made to meet stimuli by chance, finished work was made to restart the counts, and a resting state was added. All bets were then run again; the table gives the final run. Every run is kept with the page’s source.