Droplets with no biology in them
In 1992 Yves Couder and Stéphane Douady set up an experiment with no plant in it.
A dish of silicone oil sits in a vertical magnetic field, stronger at the rim than the centre. Drops of ferrofluid fall onto the middle at a steady rate. Each drop is magnetised by the field, so drops repel one another, and the field gradient pushes each one outward from the centre at a steady speed.
The drops arrange themselves into spirals. The parastichy counts are Fibonacci numbers. The divergence angle converges on 137.5°.
What the experiment establishes
It is a strong result and it is worth stating exactly what it does and does not carry.
The pattern does not require biology. No cell, no gene, no hormone and no plant is present, and the pattern appears anyway. Whatever produces spiral phyllotaxis is available to any system with the right character.
The character is what matters. Elements appearing at a steady rate on a boundary, repelling one another, drifting outward. Anything with that structure will do it, and the ferrofluid has it for reasons of magnetism rather than of botany.
The angle is not stored anywhere. Silicone oil holds no constants. If the drops reach 137.5° then the number is a consequence of the dynamics and cannot be anything else.
Those three together dispose of the mystical version of the subject far more decisively than any amount of arguing about the golden ratio. Nothing in a dish of oil admires anything.
What it does not establish
Here is where the reasoning has to be careful, and where a great deal of writing on this subject is not.
The experiment shows that a repulsion-and-drift system produces the pattern. It does not show that plants are repulsion-and-drift systems. Two processes can produce the same pattern for different reasons, and the fact that one of them contains no biology is not evidence about which one the biology uses.
The actual mechanism in plants is now reasonably well understood and it is not repulsion. It is auxin transport. Auxin is a plant hormone; the transporters that move it between cells orient themselves toward cells that already have more of it, so auxin accumulates in peaks. A peak becomes a primordium and drains auxin from its neighbourhood, so the next peak forms where the depletion is least.
That produces the same effective behaviour — an existing primordium inhibits its surroundings, and the next one appears as far from the existing ones as it can — which is why the physical analogue works. But the two systems share a description rather than a mechanism, and the description was arrived at from the physics decades before the biology confirmed it.
The general problem, and why the site keeps returning to it
Reproducing a pattern is weak evidence about a mechanism, and it feels like strong evidence, which is the dangerous combination.
The reason is that patterns are cheap. A great many rules produce spirals; a great many produce branching trees; a great many produce hexagonal tilings. Matching an output constrains the space of possible mechanisms far less than it appears to, and the more visually striking the match, the more convincing it feels and the less it means.
This is the same caution that applies to L-systems, which produce extremely convincing plants out of string rewriting and contain no plant. It is the same caution that applies to rising parastichy counts, where a static geometric effect and a developmental one leave the same trace. And it is why this site draws a firm line: the figures establish what patterns a stated rule produces, and any claim beyond that is labelled as a claim.
What would count as evidence about mechanism
Worth stating positively, because the criticism above is easy and the alternative is not.
Intervention. Blocking auxin transport changes phyllotaxis in specific, predicted ways; applying auxin locally induces a primordium where the model says one should not form. That is evidence about mechanism because it manipulates the proposed cause.
Quantitative prediction beyond the pattern. A mechanism that predicts the timing of primordium initiation, or the response to a change in meristem size, says more than one that reproduces the finished arrangement.
Failure modes. A mechanism that predicts which mutants should show which departures, and is then checked against those mutants, is doing work that pattern-matching cannot.
None of that is available to a dish of oil, and none of it is available to this site either. What both can do is establish the geometry cleanly enough that the biological work has something precise to test against — which is a real contribution and a modest one.
The result worth keeping
Strip the caveats and one thing remains, and it is the reason the experiment is famous.
Before it, the appearance of Fibonacci numbers in plants was a genuine puzzle with a whiff of the mystical about it, and the standard responses were either to invoke selection for optimal packing or to treat the numbers as a curiosity. After it, the pattern is what a certain kind of crowding does, demonstrated in a system where nothing could have evolved anything.
That is a large change in what needs explaining. The question stops being “why do plants use this number” and becomes “does a meristem behave like this”, which is a question about hormones and can be answered.
The apparatus, in enough detail to matter
The setup deserves a careful description, because every part of it maps onto a term in the model and the mapping is the reason the experiment is evidence rather than an analogy.
The dish of silicone oil is the meristem surface. It is horizontal, circular, and the drops sit on it.
The magnetic field, weak at the centre and strong at the rim, supplies the outward drift. A magnetised drop is pulled toward high field, so every drop migrates outward at a rate set by the gradient. That is the growth term: in a plant the primordium does not move, the meristem grows underneath it, but the relative motion is the same.
The ferrofluid drops, magnetised in a common direction, repel one another with a force falling off as the inverse fourth power of distance — a dipole–dipole interaction. The model uses an inverse cube for the potential, which gives that force law, and the exponent turns out not to matter much.
The drip rate is the plastochron. Faster dripping relative to the outward drift means a more crowded ring.
So the experiment’s one tunable — drip rate over drift speed — is exactly the model’s one parameter, and the bifurcation diagram is a prediction about what happens as it is turned. Couder and Douady turned it, and the transitions are where the model puts them.
Why an experiment beats a simulation here
A simulation that produces 137.5° from a repulsion rule is suggestive and no more, because a simulation is a piece of code written by someone who knows the answer. The possibility that the answer is somewhere in the code — in the interaction window, the initial condition, the sampling of the boundary — is not one that inspection reliably rules out. This site has found exactly that kind of bug in its own machinery more than once.
Silicone oil has no author. Whatever the drops do is a consequence of magnetism and geometry, and the number that comes out cannot have been put in by an implementation choice.
That is what makes the ferrofluid experiment the load-bearing evidence for the dynamical account, and the simulation the illustration of it rather than the other way round.
What “does not require biology” leaves open
The negative result is strong and its scope is narrow, so it is worth writing the remainder out.
The experiment shows that a system with the right character produces the pattern. It does not show that a meristem has that character. Those are different claims, and the second one is a question about plant physiology that oil cannot answer.
What has since been established, by people working on plants rather than on oil, is that the meristem does have it — the inhibitory field is auxin, transported by PIN1 proteins that polarise toward local maxima, and the resulting depletion around an existing primordium is the repulsion. The mechanism is chemical rather than magnetic, and the character is the same.
So the full argument runs in two steps, and the experiment is only the first: this character suffices, then separately plants have this character. Popular accounts routinely compress the two into “the golden angle emerges from physics”, which skips the part that is actually about the organism.
The rest of the family
The ferrofluid experiment is the best known of several physical systems that produce phyllotactic order without anything alive, and the list is worth having because a single experiment is easy to dismiss as a curiosity.
Buckled shells. A stiff film on a soft core, compressed, buckles into dimples whose arrangement is spiral with Fibonacci counts. The interaction is elastic rather than repulsive.
Charged particles on a growing disc. A pure electrostatic version, done numerically and in a Paul trap, giving the same branches.
Bubbles and vortices at a growing boundary, which reproduce the whorled regimes.
What the family has in common is not a force but a structure: new elements at a boundary, a repulsive interaction, and a slow outward drift. Any physics that supplies those three arrives at the same place, which is the strongest available statement of why the pattern is so common.
The transitions, and why they are the sharper test
An experiment that reproduces a number is good; an experiment that reproduces a structure is much harder to explain away.
The number 137.5° could in principle come out of many rules, and a single agreeing value is weak evidence for any of them. What the ferrofluid experiment does, and what makes it decisive, is reproduce the whole branch diagram: as the drip rate is turned relative to the drift, the pattern moves through the same regimes in the same order, with the transitions at the parameter values the model predicts.
That is many independent agreements from one tuning, and there is no plausible way for a coincidence to produce it.
It also makes an unusual prediction that the experiment can check and a plant cannot: hysteresis. Because the branches coexist over a range, turning the parameter up and then back down should not retrace the same path — the system stays on whichever branch it is already on until that branch loses stability. The droplets do this. It is the signature of coexisting attractors and it is not something a system with a stored constant could show.
What auxin added, and what it did not
The plant-side mechanism arrived over the following fifteen years and it is worth being precise about what it changed.
Auxin is transported between cells by PIN1 proteins that polarise toward neighbours with more auxin, which concentrates it into peaks. A peak becomes a primordium and depletes the auxin around it, so the region near an existing primordium cannot form another. That depletion zone is the repulsion.
What this settled: the meristem does have the character the experiment showed to be sufficient, and the inhibition is chemical transport rather than mechanical contact.
What it did not settle: the constant. Nothing in the auxin story contains 137.5°, and nobody derived the angle from PIN1 dynamics. The angle still comes from the same place it came from in the oil — the geometry of putting new elements as far as possible from existing ones on a growing boundary.
So the two halves of the argument stayed separate, which is the right outcome. The physics supplies the number; the biology supplies the reason a plant is in the class of systems the physics applies to. Neither one is the explanation on its own, and the arithmetic is a third thing again, explaining why that number rather than another is where such a system ends up.
What the oil cannot do
One asymmetry worth being clear about, since the essay has spent its length on what the experiment establishes.
The droplets settle to an angle, and nothing in the dish measures that angle or reports it. Every number quoted from the experiment comes from photographing the drops and counting, which is exactly the measurement problem this site spends its length on: the counts depend on where in the pattern one counts, and the divergence angle has to be recovered rather than read.
So the experiment’s headline result — “the drops converge on 137.5°” — is itself a recovered quantity with an uncertainty, obtained the same way it would be from a sunflower.
That does not weaken it. It does mean the experiment and the plant are on the same footing methodologically, which is a reason to trust the comparison and not a reason to trust the oil more than it deserves.
What the experiment changed about the argument
It is worth being concrete about the state of the question before and after 1992, because the shift is larger than a single result usually produces.
Before: the observations were a century old and thoroughly catalogued — lattices, parastichy numbers, transitions — and the explanations on offer were either teleological (the angle is optimal, so it was selected) or genetic (the plant encodes it). The first does not survive being measured; the second is untestable in the form it was usually stated and predicts nothing.
After: the pattern is the settled state of a dynamical system, and a system with no genes and no selection produces it. That does not make the biology irrelevant — a meristem still has to be the kind of system the argument applies to — but it moves the explanatory burden from “why does the plant want this angle” to “does the meristem have this character”, which is a question about transport and geometry that can be answered.
It also reframed the golden angle from a value the organism aims at to a value the dynamics lands on, which is what makes the arithmetic account of why that value relevant rather than decorative.