A calculation with no memory in it

The bee that cannot fly traces the story to its origin, finds the mistake — a fixed-wing calculation applied to an animal that has no fixed wing — and does the blade-element sum properly for four insects. It ends with an admission: the proper sum still asks for more lift than a steady aerofoil at that Reynolds number produces, and it names the mechanisms that close the gap without computing any of them.

This essay computes them, and the reason they were named rather than computed is worth stating first. None of them is a property of the flow at an instant. Every one is a statement about what the wing has been doing, which is exactly the class of quantity a quasi-steady sum has no slot for.

What “quasi-steady” claims #

A blade-element calculation takes the wing at an instant, reads off its local velocity and incidence, looks up the lift coefficient a wing of that section at that incidence produces in steady flow, and integrates along the span.

The phrase doing the work is in steady flow. A steady lift coefficient is measured on a wing that has been travelling at that incidence for a long time — long enough that its starting vortex is far downstream and its circulation has settled. The measurement is a limit, and the calculation assumes the limit has been reached.

For an aeroplane in cruise that is unimpeachable: the wing has travelled millions of chord lengths since anything changed. For an insect it is the thing to check, and checking it is one line of arithmetic.

How far an insect wing gets #

The blade-element argument collapses a revolving wing onto the radius of its second moment of area, so that is where the travel should be measured. Multiply the stroke amplitude by that radius to get the distance the representative element covers in a half-stroke, and divide by the mean chord.

A hawkmoth gets 2.7 chords. A bumblebee gets 3.8. A hoverfly gets 3.5. A fruit fly gets 4.7.

Four animals whose masses span a factor of 1,646, all between 2.66 and 4.73 chords. That is not something the calculation was set up to produce; it comes out of measured morphologies, and it is the first indication that something is being selected for rather than accidentally arrived at.

Marked on the same axis is the distance over which a leading-edge vortex on a translating plate has been measured to stay attached before it sheds — four chords. The four animals sit at 2.66, 3.45, 3.78 and 4.73, which brackets it: three below and one just above, on morphologies spanning three orders of magnitude in mass.

What Wagner’s function says about it #

A wing that starts impulsively does not have its steady circulation. It has to shed a starting vortex, and until that vortex is well away the bound circulation is short. Wagner’s function is the fraction built as a function of distance travelled: about half at the first instant, three quarters after four chords, and asymptotically one. Averaged over each animal’s own half-stroke it reads 0.686 to 0.755 — so between 24.5 and 31.4 per cent of the steady lift is never built at all, and the shortfall is largest for the animal with the shortest stroke.

Average it over each insect’s own half-stroke and the answer is the fraction of the steady circulation the wing actually carries, averaged over the stroke.

Hawkmoth 0.686. Hoverfly 0.717. Bumblebee 0.728. Fruit fly 0.755.

The quasi-steady sum uses one. So before anything else happens it has given the wing between a quarter and a third more circulation than an unsteady wing of the same geometry doing the same motion would ever build.

The same curve says how much longer the stroke would have to be for the assumption to hold. Fifty chords gets within five per cent. Two hundred gets within two. An insect gets three or four, and no amount of care with the blade-element integral changes that, because the error is not in the integral.

Where that curve comes from #

Wagner’s function is used above as a given, and this collection derives it elsewhere, so it is worth saying briefly what it is a consequence of rather than treating it as a fitted curve.

The lift that arrives late is the essay that computes it: when a wing starts, Kelvin’s theorem requires the circulation it acquires to be matched by an equal and opposite starting vortex, and while that vortex is still nearby it induces a downwash at the wing that cancels part of the incidence. The deficiency is therefore the wake’s own influence on the wing that made it — a memory in the most literal sense, since the shed vorticity is a physical record of every earlier instant of the motion, sitting downstream and still being felt.

The consequence for this essay is that the shortfall is not a modelling choice. It is required by a conservation law, and any wing that changes its circulation pays it. The only question is how much of it has been paid off by the time the stroke ends, and the answer for an insect is about three quarters.

That also explains why the number depends on distance travelled rather than on elapsed time: what matters is how far the starting vortex has got, in chord lengths, and it convects with the flow.

The air the animal has already moved #

The second omission is larger and is easier to see.

A hovering animal holds itself up by throwing air downwards. That air is still moving when the next stroke arrives, and the wing therefore does not fly through still air; it flies through the wake of everything it has done for the last several strokes.

Momentum theory prices it without any modelling: the induced velocity is the square root of the weight over twice the density times the swept area.

For the bumblebee that is 1.39 metres a second against a stroke speed of 4.70 — thirty per cent. For the hawkmoth it is 45 per cent. Converted into incidence, the wing sees between 16 and 24 degrees less than the geometric angle the quasi-steady sum uses.

And it goes the same way as Wagner’s shortfall. Both make the real wing produce less than the calculation assumes. The two errors do not cancel; they compound, and the honest version of the quasi-steady deficit is therefore larger than the factor of two the previous essay reports.

So what actually holds the animal up #

Having made the hole bigger, the essay owes an account of what fills it.

The mechanism is the leading-edge vortex, and its significance is precisely that it is a memory. A wing at 40 degrees of incidence separates at the leading edge. On a translating wing the separated vortex grows, detaches after a few chords, and the wing stalls. On an insect wing the stroke ends first.

The animal is therefore living in the interval between separating and stalling — an interval that is several chord lengths long — and it is doing so on every stroke, deliberately. The vortex sits over the wing, its low core pressure adds lift, and the wing reverses before the bill arrives.

That is why the stroke lengths cluster where they do. A longer stroke would shed; a much shorter one would waste most of the cycle in reversal. The band the four animals occupy is bounded above by a memory.

Two refinements are worth naming because they change the picture from “just about survives” to “works robustly”. On a revolving wing rather than a translating one the vortex is stabilised by spanwise flow along the wing and does not shed at four chords at all — which is what Ellington’s work established and what makes the mechanism reliable rather than marginal. And the wing’s rapid rotation at each stroke reversal generates circulation of its own, in the same way a rotating aerofoil does, adding lift exactly where the translational contribution is passing through zero.

What the solver computed, and how it was checked #

The morphologies are measurements and are marked as such. Everything else is arithmetic on them.

The masses span a factor of 1,646 and every half-stroke lands between two and five chords.

Wagner’s function is the standard two-lag fit, averaged over each stroke in closed form. It is checked to increase monotonically with stroke length and to reach 0.988 at two hundred chords, which is the check that the average is being taken over the right variable — an average over time rather than travel would not have that limit.

The stroke travel is checked to lie between two and five chords for all four animals, which is the essay’s central observation and is the assertion that would fire if a morphology were mistyped.

The mass spread is checked to exceed a thousand, because the observation is only interesting if the animals are genuinely different sizes. It is 1,646.

And the induced velocity is checked to exceed a tenth of the stroke speed for every animal, which is the threshold below which it could be neglected. It is three to four times that.

The same wing, in a wind tunnel #

The clearest confirmation that these are memory effects and not modelling errors comes from a case where the same wing is driven at two different speeds.

Two lifts at one incidence computes the dynamic-stall loop: a wing pitched up and down through the stall carries more lift going up than coming down, and the gap is 0.22 in lift coefficient at twelve degrees. The mechanism is that separation takes time to develop, so a wing that is moving quickly reaches a higher incidence before the separation catches up.

That is the same statement as the leading-edge vortex above, made about a wing on a shaft rather than an animal. It also produces a number the insect case can use: the overshoot above the static maximum rises monotonically to 1.4 times, at reduced frequencies comparable to an insect’s.

So the insect’s mechanism is not exotic. It is dynamic stall, used deliberately and continuously, by an animal whose stroke is short enough to stay inside it.

The same lag, on an aeroplane wing rather than an insect’s, is what the lag that makes flutter possible is about — there it is a hazard rather than a resource, and dropping it from the model moves a computed flutter speed by sixty per cent. It is the same physical quantity in all three essays: the wake’s influence on the surface that shed it, delayed by how long the wake takes to get away.

And it is the reason the fixed-wing intuition fails here in a way that is not about arithmetic. An angle, not a speed records that a wing stalls at an angle rather than a speed — which is true of a wing that has always been going, and is exactly the statement that stops being true once the incidence is changing fast compared with the wake’s convection.

Where the numbers put this essay #

The general form is every memory number is one time over another, and the insect’s group is the reduced frequency — the time for the wake to convect a semi-chord over the period of the motion. Insects run at reduced frequencies of order 0.2 to 0.4, which is squarely in the band that essay identifies as the one where the history is part of the answer and no simple limit applies.

That is worth stating because it removes the mystery. The quasi-steady sum is the small-reduced- frequency limit of the correct calculation, and it is being applied at a reduced frequency of a third. It is not that insects are strange; it is that the calculation was evaluated outside its own domain, and the domain has a number attached to it that was available all along.

What the misconception actually costs #

The bee story is usually treated as harmless, and the version corrected in the previous essay — that somebody used the wrong velocity — is indeed harmless once corrected.

The version this essay is about is not harmless, because it is still live. A quasi-steady blade-element calculation is the standard first-cut tool for small flapping vehicles, and it is used because it is cheap and because its errors are believed to be modest. The errors computed here are not modest, they are systematic, and they are in the direction that makes a design look worse than it is — so the practical consequence is a vehicle sized with margin against a deficit that does not exist, or a mechanism dismissed as insufficient when it was being evaluated wrongly.

The correction is not to abandon the method. It is to carry two corrections it can accommodate: a Wagner factor on the translational term, and an induced velocity from momentum theory subtracted from the stroke velocity. Both are single lines and both are standard in the flapping-flight literature; neither is in the popular account.

What the picture cannot show #

The stroke lengths are computed at the radius of the second moment of area, which is one representative station. The tip travels further and the root travels less, so the wing is not at one point on Wagner’s curve — it is smeared across a range, and the outboard part of the wing is closer to steady than the inboard.

The induced velocity is a mean over the swept disc. The real induced field is unsteady, concentrated in the wake structures the animal makes, and a wing passing back through a vortex ring it shed sees something much larger than the mean for part of the stroke. That is wake capture, and it is a memory this essay prices only in the average.

And Wagner’s function is derived for a thin aerofoil starting impulsively in inviscid flow at small incidence. An insect wing is thick relative to its Reynolds number, starts smoothly, and operates at forty degrees. The function is used here as an order-of-magnitude statement about how long circulation takes to build, which is what it is good for, and not as a quantitative prediction.

Where the model stops #

Four animals. They are the four with well-measured morphology in this collection’s own table, and the clustering of stroke length is an observation about those four rather than a survey.

Hovering only. The induced-velocity calculation assumes hover. In fast forward flight the wing meets mostly free stream and the correction shrinks, which is one reason forward flight is easier for the animal than hovering.

No wing flexibility. Insect wings deform substantially under load, and the deformation changes both the camber and the effective stroke plane. Nothing here has a flexible wing in it.

No impulsive limit. The added-mass force at stroke reversal is a genuine unsteady term and it is not a memory: it depends on the acceleration now and on nothing earlier, which is the same path-independence that everything about the start, except one vector computes for an impulsively started body. It is left out here because this essay is about the terms that carry a history, and it should be carried in a real force budget.

And no energetics. This essay is entirely about whether the lift is there. Whether the animal can supply the power is a separate calculation with its own assumptions, and it is where the muscle physiology enters.

Who found it, and when #

Wagner’s function is 1925. Osborne applied quasi-steady blade elements to insects in 1951 and found the gap. Weis-Fogh’s clap-and-fling of 1973 was the first mechanism proposed to close it, and it applies to a minority of insects. Ellington’s group established the leading-edge vortex on a revolving wing in 1996, and Dickinson’s Robofly measurements of 1999 separated the contributions of delayed stall, rotational circulation and wake capture experimentally.

The chronology is the interesting part. The quasi-steady calculation was known to fall short for forty-five years before the mechanism was found, and during that time the honest position was that something was missing — which is not the same as the position the story attributes to aerodynamicists.

The contrast with a genuinely wrong theory is worth drawing. The theory that forbade flight is about a model that gives an answer thirty-six times too small at five degrees and is wrong because its physics is wrong. The quasi-steady insect sum is a correct model evaluated outside its domain, and it is off by a factor of two in the other direction. Those are different failures and only the second is repaired by a correction term.

Limits recorded rather than smoothed over #

The shedding length is borrowed. Four chords for a translating plate is a measurement from the literature, not a computation here, and the revolving-wing case does not shed at all. It is drawn as a band on the axis because that is the honest way to show a number of that provenance.

Wagner and the induced velocity are not additive. They are computed separately and their sizes compared, and combining them properly requires an unsteady model that carries both, which this collection does not have.

The lift coefficients required in the ledger are the previous essay’s numbers, reproduced here for comparison rather than recomputed.

And nothing here predicts an insect. It prices the terms a standard calculation drops. Whether the animal’s actual mechanism supplies exactly that much is a measurement, and the measurements exist and are cited above.