The gradient that does both

The boundary-layer momentum equation, evaluated at a solid wall, collapses to one term on each side. Both velocity components vanish there, so every convective term goes with them, and what remains is

$$\mu \left.\frac{\partial^2 u}{\partial y^2}\right|_{0} = \frac{\mathrm{d}p}{\mathrm{d}x}$$

This is the most consequential line in the subject that is never given a name. It says the curvature of the velocity profile at the wall is fixed by the pressure gradient — exactly, within the boundary-layer approximation, with no further assumption.

Two entirely different things follow from it, and this essay is about the distance between them.

The first consequence: an inflection point, immediately #

At the outer edge of a boundary layer the profile flattens onto the free stream, so its curvature there is negative. At the wall the curvature has the sign of dp/dx.

In a favourable gradient both are negative, and the profile can run from one to the other without its curvature ever changing sign. There is no inflection point inside the layer.

In an adverse gradient the wall curvature is positive and the edge curvature is negative, so the curvature must pass through zero somewhere between. There is an inflection point, necessarily, from the very first station at which the gradient turns adverse.

That matters because of Rayleigh’s criterion: an inviscid parallel flow cannot be unstable without an inflection point, and an inflectional profile — where it is unstable at all — is unstable through a mechanism one to two orders of magnitude more vigorous than the viscous one that destabilises a flat-plate layer.

So the moment the pressure begins to rise, the layer’s stability problem changes character. Not its thickness, not its shear, not its drag: its mechanism.

The second consequence: separation, eventually #

Separation is a different statement about the same profiles. It happens when the wall shear reaches zero — when the fluid nearest the surface has been brought to a standstill and the next increment of adverse pressure sends it backwards.

The Falkner–Skan family parameterises the whole question with one number, and the site bisects on it rather than quoting it: laminar separation sits at β = −0.198838, and past that value the similarity solution does not exist at all.

Both facts come from the same equation and the same sign. But they do not arrive together.

The inflection point appears at β = 0⁻, the instant the gradient turns adverse.

Separation waits until β = −0.198838, which on a real surface is a great deal further along.

Measuring the distance between them #

The gap can be put in terms an aerofoil designer would recognise, and the arithmetic is worth doing because the answer is not marginal.

On a conventional section at moderate incidence, the pressure minimum sits somewhere between 10% and 40% of chord depending on the design, and separation — if it happens at all before the trailing edge — is well aft of 60%. The adverse region is therefore typically a third to a half of the chord long.

Over the whole of that distance the layer is inflectional, and therefore inviscidly unstable, and therefore amplifying disturbances at rates that make transition likely well before it reaches the separation point.

This is why, on most aerofoils at most Reynolds numbers, transition happens shortly after the pressure minimum. Not because there is anything special about the minimum for the transition mechanism directly, but because the minimum is where the profile acquires the inflection point that switches on the fast instability.

The design consequence follows immediately and is the whole of natural-laminar-flow practice: to keep a layer laminar, keep the flow accelerating. Every laminar section is shaped to hold its pressure minimum as far aft as it can, because the laminar run ends shortly after it, and the drag difference between the two states is a factor of several.

The shape factor, as one number for both #

There is a single quantity that tracks both consequences at once, and it is worth knowing because it is what a practical boundary-layer method actually carries: the shape factor H, the ratio of the displacement thickness to the momentum thickness.

For a Blasius profile H = 2.59. For a Falkner–Skan profile at the separation value it is 4.03. For a turbulent profile at the same free-stream conditions it runs between about 1.3 and 1.4, and separation for a turbulent layer is usually taken at H between 2.4 and 2.9 depending on whose correlation is used.

The number is doing two jobs. It measures how hollowed out the profile is — how much momentum has been lost from the region near the wall relative to the mass displaced — which is exactly the quantity that decides both how close the layer is to separating and how strong its inflection is. And it is computable from an integral method that does not resolve the profile at all, which is why it is the variable every panel-code boundary-layer module marches.

The three numbers above also make the momentum argument quantitative. A turbulent layer’s H of 1.35 against a laminar layer’s 2.59 is the statement that the turbulent profile has lost far less momentum near the wall for the same displacement, and it is the whole of why it survives a steeper hill.

Why the order of events is fortunate #

There is a lucky consequence of instability arriving first, and it is responsible for a great deal of practical aerodynamics working at all.

Transition changes the profile. A turbulent boundary layer mixes momentum vertically, dragging fast fluid from the outer part of the layer down towards the wall, and the resulting profile is far fuller — much more momentum in the lowest tenth of the layer than a laminar profile of the same thickness carries.

A profile with more momentum near the wall can climb a steeper pressure hill before running out. Where a laminar layer separates at β = −0.1988, a turbulent one survives adverse gradients several times steeper.

So the sequence on a real aerofoil is: the pressure minimum puts an inflection point in the profile; the inflection point makes the layer violently unstable; the layer transitions; the turbulent profile is far more resistant to the very gradient that caused the instability; and separation is postponed, often to the trailing edge.

Instability is what saves the flow from separating. That is not the relation the words suggest, and it is the correct one.

The refutation, stated plainly #

The belief that a turbulent layer separates sooner is common and is worth taking apart rather than merely contradicting, because the intuition behind it is not stupid.

The intuition is that separation is a kind of disorder, that turbulence is disorder, and that more of the second should produce more of the first. Both halves of that are category errors. Separation is a statement about the sign of a single derivative at the wall; turbulence is a statement about the spectrum of the velocity field. They are not the same quantity and there is no reason for one to imply the other.

The mechanism that actually connects them runs the other way and is about momentum. Turbulent mixing transports momentum toward the wall. Momentum near the wall is exactly what resists separation. Therefore turbulence resists separation.

The practice built on this is unambiguous. Turbulator strips, boundary-layer trips, vortex generators and dimpled golf balls all exist to make a layer turbulent on purpose, and they are fitted precisely where separation is the problem. Nobody would spend drag on tripping a layer if the belief were true.

The cost side of the bargain #

Tripping a layer is not free, and the reason it is nevertheless usually right is a comparison between two drag contributions rather than a single number.

Skin friction goes up. A turbulent layer’s wall shear is several times a laminar layer’s at the same Reynolds number, and the ratio rises with Reynolds number. On a flat plate at Re = 5·10⁶ the turbulent layer costs about eleven times the laminar one, and there is no way to have the resistance to separation without paying it.

Pressure drag goes down, and usually by more. A separated flow leaves a wide low-pressure wake, and the pressure difference between the front and back of a bluff body is a much larger force than the friction on either. On a sphere the transition from laminar to turbulent separation cuts the drag coefficient from about 0.5 to about 0.1 — a factor of five reduction in total drag, bought with an increase in the smaller of the two contributions.

The trade-off therefore depends on which contribution dominates, which is a question about the shape. On a streamlined body at small incidence, friction is most of the drag and keeping the layer laminar wins. On a bluff body, or a streamlined one near stall, pressure drag is most of it and tripping wins. The two drags a wing pays is where that split is measured rather than asserted.

Four ways to buy the same quantity #

Once separation is understood as a shortage of momentum near the wall, every device fitted to prevent it can be read as a way of supplying some — and there are exactly four routes, distinguished by where the momentum comes from and what each costs.

Make the layer turbulent. A trip strip, a rough band, a wire: the mixing then transports momentum inwards by itself, and the profile fills out. This is the mechanism the essay has been about, and its price is the several-fold rise in skin friction over everything downstream of the trip.

Bring outer momentum in by advection instead. A vortex generator is a small vane set at incidence, of a height comparable with the layer’s own thickness, that sheds a streamwise vortex. That vortex carries fast fluid from the outer part of the layer down to the surface and slow fluid outward — the same exchange turbulence performs, but organised, directed, and far more effective per unit of drag paid. It is why an array of vanes a centimetre tall can keep a wing attached where tripping alone would not: they are not making the layer turbulent, which on a transport wing it already is; they are replacing random mixing with a pump.

The sizing follows directly. A vane much shorter than the layer reaches only fluid that is already slow and has nothing to fetch; one much taller sticks out into the free stream and pays form drag for nothing. Sub-boundary-layer devices — a fraction of $\delta$ tall — work because the momentum worth fetching is not at the very top of the layer, and they cost far less when the aircraft is not near separation.

Add momentum from outside the flow. A slot in a multi-element aerofoil discharges fresh fast fluid along the upper surface of the element behind it, starting a new boundary layer where the old one was exhausted; blown flaps do the same with engine air. This buys the most and costs plumbing.

Or remove the fluid that has none. Suction through the surface takes the slowest fluid out of the layer altogether, which raises the mean momentum by deleting the part that was dragging it down rather than by adding anything.

All four move the same number. Each raises the momentum in the lowest tenth of the layer, each lowers the shape factor, and each therefore lets the layer climb a steeper hill. What separates them is entirely the bill: friction everywhere, form drag when not needed, engine bleed, or a pump — and which one an aircraft carries is a decision about that bill rather than about the aerodynamics.

The bubble, which is what happens when both occur #

At low Reynolds numbers the two events can happen in the other order, and the result is one of the more troublesome features in aerodynamics.

If the layer reaches the separation point while still laminar — which happens on model aircraft, on sailplane sections at low speed, and on turbine blades at part load — it detaches. The detached shear layer is then a free shear layer, inflectional and violently unstable, and it transitions within a short distance. The now-turbulent layer entrains momentum, and reattaches to the surface behind the separation point.

What is left is a laminar separation bubble: a closed region of slow recirculating fluid, some per cent of chord long, with a laminar layer arriving at the front of it and a turbulent one leaving the back.

Bubbles are unwelcome for three reasons. They add drag, because the pressure distribution over the bubble is flat and the recovery afterwards is abrupt. They are hysteretic, so the incidence at which a bubble forms is not the incidence at which it clears. And they can burst — fail to reattach — which converts a small drag penalty into a full stall with very little warning, and this is the mechanism behind the abrupt low-Reynolds-number stall that model sections are notorious for.

The remedy is again to trip the layer early, before it reaches separation, so that the sequence returns to the fortunate one.

Where the model stops #

Three limits, each of which the figures above are inside.

Falkner–Skan is a similarity family, not a geometry. Its β describes a free stream varying as a power of x, which no aerofoil has. It is a way of parameterising how adverse a gradient is, and the separation value it yields is a benchmark rather than a prediction for a particular section.

The inflection criterion is inviscid. It is being applied to a viscous layer, on the argument that where the inviscid instability exists it dominates the viscous one. That argument is sound and it is an argument.

Nothing here computes transition. The essay says the layer becomes strongly unstable at the pressure minimum and that transition usually follows shortly. How shortly depends on the disturbance environment, which is not a property of the fluid and is not computed anywhere on this site.

What the site’s own solver can and cannot show of this #

The vorticity–streamfunction stepper resolves separation and does not resolve transition, and the distinction is worth stating because both appear in the figures above.

Separation is a mean-flow event at a Reynolds number the grid can hold. The stepper marches the field at Re = 400, the wall shear changes sign at a definite angle, and the recirculating region behind the body is a computed object with the conservation checks applied to it. When a figure on this site says the flow separated at a particular place, a solver found the place.

Transition is not resolvable at any Reynolds number this site’s solver can reach. It is three-dimensional, it depends on disturbances the solver does not contain, and the grid it would need is beyond a build step by many orders of magnitude. Every statement in this essay about when the layer transitions is therefore a statement about the stability mechanism plus an appeal to experiment, and the figures mark it as such.

The honest form of the whole argument is then: the site computes both ends — the profile that becomes inflectional and the profile that separates — and asserts, on the strength of the stability theory plus the measured behaviour of real aerofoils, that transition usually falls between them.

Who found it, and when #

The wall relation comes with the boundary-layer equations themselves — Prandtl, 1904 — and the inflection consequence was noted almost immediately, though it took until Tollmien and Schlichting’s work in 1929–33 for the stability side to be quantitative.

Falkner and Skan published the similarity family in 1931, and Hartree computed the solutions in 1937, including the separation value.

The practice of deliberate tripping is older than the explanation. Golf balls were dimpled by the 1900s on empirical grounds; Prandtl’s own 1914 experiment with a tripwire on a sphere is what demonstrated the mechanism, and it is the same experiment that established the drag crisis.

Where the ladder goes next #

This anchor’s earlier rungs establish what separation is, the number at which a laminar layer gives up, and what a separated flow does to the forces. This rung is the one that connects them to the other consequence of the same sign.

From here the argument continues on the turbulent side: the cost of going turbulent puts the two skin-friction laws side by side, and is where the trade-off in tripping a layer deliberately gets its numbers.