The bee that cannot fly

It is the most repeated claim in this subject and, unusually, it has a traceable origin: a calculation done at a dinner in 1934, in which somebody applied the arithmetic of a fixed wing to an animal that does not have one.

The interesting thing about it is not that it is wrong. It is that doing the calculation properly still falls short, and closing the remaining gap took another sixty years and a mechanism nobody had looked for.

Where the arithmetic goes wrong #

The fixed-wing calculation is short. Take the animal’s weight, its wing area, and the speed at which it flies, and ask for the lift coefficient that would hold it up:

$$C_L = \frac{W}{\tfrac12\rho V^2 S}.$$

For a fruit fly, flying at forty centimetres a second with 3.7 square millimetres of wing, that comes out at 26 — which is preposterous, and is where the story’s conviction comes from.

For a bumblebee it comes out at 1.31, which is not preposterous at all. It is a little above what a thin section at that Reynolds number can give, and a reasonable person looking at it would say the animal is marginal rather than impossible.

For a hawkmoth it comes out at 0.88, which is comfortable.

The spread is the refutation, not any individual value. A calculation whose answer varies by a factor of twenty-nine across four animals that all fly perfectly well is not measuring the animals. It is measuring their forward speeds, which have nothing to do with how any of them makes lift.

The velocity the wings actually see #

A flapping wing sweeps through an arc at Φ radians, f times a second. An element at radius r therefore moves at r times the angular rate, and the lift it makes goes as the square of that. So the span integral carries r², and the whole thing collapses to the second moment of wing area — one number about the shape, and the only thing about the shape that appears.

There is one subtlety that moves the answer by twenty-three per cent and is easy to get wrong. Lift goes as the square of the speed, so what the weight has to be set against is the mean of U² over a stroke — and a wing sweeping harmonically spends most of its time faster than its own average. For $\varphi(t) = (\Phi/2)\cos 2\pi f t$ the mean square angular rate is $(\pi\Phi f)^2/2$ against a mean rate of $2\Phi f$, a factor of π²/8 in the effective dynamic pressure.

The solver checks that factor by quadrature over a stroke against the closed form, to a part in a million. Averaging the speed first and squaring afterwards understates the lift available by a quarter, which is enough to move a species from one side of the argument to the other.

What comes out #

With the right velocity the four requirements are 0.91, 0.97, 1.33 and 1.73 — a spread of 1.9 rather than 29, and every one of them a lift coefficient an aerofoil argues about rather than one it laughs at.

Two of the four need more than a steady aerofoil can give. All four are inside what a revolving wing gives, with the largest at ninety-six per cent of it.

Both ceilings are measurements, made by other people, and neither is computed anywhere on this site — they are drawn in the colour kept for a borrowed quantity and the caption says so. What is computed is the requirement.

That fit is closer than the argument deserves and it is the reason the leading-edge vortex is taken to be the answer rather than one candidate among several. The gap between what the animals need and what a steady aerofoil gives is real, and the one mechanism known to be present closes it — for all four, with no room to spare.

Why the requirement is so nearly the same for all four #

The tightness of the flapping answers is worth a paragraph in its own right, because it is a result rather than a coincidence.

The requirement is a weight divided by a dynamic pressure and an area, and every term in it scales with the animal. Write it out: the weight goes as the cube of a length, the wing area as the square, and the velocity as the wing length times the flapping rate. So

$$C_L \propto \frac{\ell^3}{\ell^2\,(\ell f)^2} = \frac{1}{\ell f^2}.$$

The requirement therefore depends on the product of a length and the square of a frequency, and insects vary those two in opposite directions: small ones beat fast, large ones beat slowly. A fruit fly at 218 hertz and 2.4 millimetres and a hawkmoth at 26 hertz and 48 millimetres differ by a factor of twenty in length and eight in frequency, and the product $\ell f^2$ differs by less than a factor of two.

That is not an accident of the four animals chosen. Wingbeat frequency across flying insects scales roughly as the inverse of wing length, which is the scaling that keeps this group nearly constant — and the reason for that scaling is a resonance in the thorax, which is a question about muscle and elasticity and not about aerodynamics at all.

The solver checks the exponents directly rather than reasoning about them: doubling the frequency, the stroke amplitude or the wing length must each quarter the coefficient required, and each is required to nine decimal places.

The mechanism, named rather than computed #

A wing revolving about its base, at these Reynolds numbers, develops a vortex over its leading edge which does not shed. On a translating wing that vortex grows, detaches, and the wing stalls; on a revolving one it is stabilised — by spanwise flow through the core, by the rotational acceleration gradient, and by the fact that the wing keeps turning into fresh fluid — and it sits there, contributing a strong suction over the upper surface for as long as the stroke lasts.

That is the same trade this site has already met on a delta wing: a separation that would end a conventional wing’s lift, used on purpose because it stays where it is put. The delta’s vortex is held by a sharp swept edge; an insect’s is held by the rotation. The resemblance is not superficial, and the analogy was the route by which insect flight was eventually understood.

Two further mechanisms are named in the literature and neither is computed here. Rotational circulation: the wing flips over at each end of the stroke, and a rotating aerofoil generates circulation by the same argument that gives a spinning cylinder its lift. Wake capture: the wing reverses into the fluid it just set in motion and recovers some of the momentum it put there.

This site computes none of the three. It solves steady fields on coarse grids and cannot resolve a revolving wing’s vortex, and pretending otherwise would break the invariant this whole site is built on. What it can do is compute the size of the hole and name what fills it, which is a different and honest claim.

How the vortex was actually seen #

Naming a mechanism is cheap and the evidence for this one is worth having, because of how it was obtained: the leading-edge vortex was not found on an insect. It was found on a machine built to be the same flow as an insect, and the reasoning behind that machine is the whole content of similarity.

The difficulty with the animal is plain. A fruit fly’s wing is two millimetres long and beats two hundred times a second, so seeing the flow round it means resolving structures of a fraction of a millimetre in a fraction of a millisecond, and measuring the force on it means instrumenting a wing that weighs micrograms. Neither was available.

The escape is that neither the size nor the speed matters — only the two dimensionless groups do. Build a wing a hundred times larger, flap it in a fluid a hundred times more viscous, and slow it down until the Reynolds number and the reduced frequency match, and the flow round the model is the flow round the animal. The numbers are startling in the direction opposite to the usual one: a wing a quarter of a metre long, immersed in mineral oil, sweeping through its stroke in several seconds, reproduces a fruit fly at Reynolds number 130.

That is a wind tunnel’s problem inverted. The standard complaint about scale models is that matching the Reynolds number needs a speed or a pressure nobody has, because models are smaller than the real thing. Here the real thing is tiny, so the model is larger and slower, and the group comes out right without any facility at all — a tank of oil and a stepper motor.

What the arrangement bought was two things the animal could not give. The flow could be seeded and photographed, which is how the vortex over the leading edge was seen to form and, crucially, not to shed for the length of the stroke; and the forces could be measured directly on the wing, cycle by cycle, so the lift a revolving wing actually produces could be compared with what an animal requires rather than inferred from it. The ceiling near 1.8 that this essay draws is a reading from apparatus of that kind.

It is also why the mechanism is not a special property of insects. A revolving wing at these Reynolds numbers holds the vortex whatever is turning it, so the same structure has since been identified on bats, on small birds’ wings in slow flight, and on the autorotating seed of a maple — which descends more slowly than its shape alone would predict, and does so by spinning a stable vortex over its own leading edge all the way down.

The number that disqualifies the calculation #

There is one more measurement and it is the one that makes the whole essay’s claims narrow.

The reduced frequency compares the wing’s own chord to the distance it travels in a beat. Below about 0.05 the flow settles and a quasi-steady calculation is honest; above about 0.1 it does not. All four animals are between 0.3 and 0.6.

So the quasi-steady requirement is not a prediction of what the animal does. It is a bound on what steady aerodynamics would have to supply if steady aerodynamics were the whole story, and the point of computing it is precisely that it is too large — that is the measurement of the gap.

That is why the agreement with the revolving-wing ceiling is the interesting number and the disagreement with the steady one is the interesting number, and why an exact match to either would have been suspicious rather than satisfying.

What was actually claimed, and by whom #

The origin is usually given as a dinner in 1934 at which the entomologist August Magnan, or his assistant André Sainte-Laguë, did an envelope calculation and found the insect’s wings insufficient. Magnan’s book Le Vol des Insectes of that year records something close to it, with a remark that the result had led him to think aviation’s laws did not apply to insects.

The calculation was not stupid and it was not a joke. It is the fixed-wing calculation above, done before anybody had the blade-element apparatus for flapping flight, at a moment when the aerodynamics of a steady wing had just been put on a firm footing and the natural move was to apply it.

What turned it into folklore is a separate process. It travelled, lost its qualifications, acquired the word proves, and has been used ever since as a story about the arrogance of theory — which is almost exactly backwards. The theory was fine. The velocity was wrong, and the version of the claim that is actually true is the hovering one: a fixed wing at zero forward speed has no answer at all, not a large one.

The solver states that as a refusal. Ask it for the fixed-wing requirement at zero speed and it declines, with the reason: with no forward speed there is no dynamic pressure and no lift coefficient however large will do — that is not a large number, it is the absence of one.

Three things worth taking away #

A model is disqualified by a parameter, not by an answer. The reduced frequency says the quasi-steady calculation cannot be a description of insect flight before any number is compared with any measurement. Checking the applicability parameter first is cheaper than checking the answer, and it is the step the 1934 calculation missed.

A spread across cases is more diagnostic than a value in one case. The fixed-wing calculation gives a defensible answer for a bumblebee and an absurd one for a fruit fly, and the absurdity is the information. One case can always be explained away.

And a gap that survives a correct calculation is a finding. Getting the velocity right did not make the problem go away; it turned an absurdity into a factor of about one and a half, which is exactly the size of gap that identifies a missing mechanism rather than an arithmetic error. That is how the leading-edge vortex was looked for.

The other version of the story #

There is a second form of the claim that is worth separating from the first, because it is heard almost as often and fails differently.

Insects fly at Reynolds numbers where aerofoils do not work. That one is closer to being true and is still misleading. At Re around a hundred — a fruit fly — a thin section really does behave badly: its maximum lift coefficient is low, its drag is high, and the ratio of the two is a fraction of what a full-scale wing achieves. A fruit fly’s lift-to-drag ratio is around three or four, where an airliner’s is nearly twenty.

But badly is not not at all, and the arithmetic here is what makes the difference precise: at Re ≈ 100 a thin plate at high incidence still gives a lift coefficient above one, and one is enough. What is expensive is the drag, and the consequence is a power requirement rather than an impossibility.

The low-Reynolds-number claim is about efficiency and gets stated as being about feasibility, which is the same category error as the original with a different quantity in it. An insect is a spectacularly inefficient flying machine and a completely functional one.

What the model does not contain #

Every morphological number is a measurement. Mass, wing length, wing area, stroke amplitude, wingbeat frequency and the second-moment radius are all taken from the literature. Nothing here predicts an insect; it prices one.

Both ceilings are measurements too, and they carry the largest uncertainty in the whole argument. The steady value depends on the section and the Reynolds number; the revolving value depends on the aspect ratio and the stroke kinematics. Quoting them to two significant figures is generous.

No unsteady aerodynamics at all. Delayed stall, rotational circulation, added mass and wake capture are named and none is computed.

No spanwise or chordwise variation of angle. A real insect wing twists substantially along its span during a stroke, and the blade-element calculation here uses one mean coefficient for the whole wing over the whole cycle.

No power, no muscle and no efficiency. The essay computes what lift is needed and says nothing about whether the animal can produce the power to make it — which is a genuinely separate question and one where the margins are also tight.

And no drag. A flapping wing pays a profile drag that this arithmetic ignores entirely, and in a hover that drag is what sets the power requirement.

Where the ladder goes next #

This phase opened with the shape of a camber line and closes with an animal, and everything in between has been the same question: what does a surface have to do to the air to hold something up, and what is the number that decides it. The ladder above each of these anchors goes on — a panel method wants a boundary layer coupled to it, a wake wants a roll-up computed rather than bounded, and a flapping wing wants a solver this site does not have.