Ideal flow

The exact theory says nothing has any drag

Solve the flow past a body in a fluid with no viscosity and the answer is beautiful, closed-form, and predicts that a cyclist needs no legs and an airliner no engines. This is not a small error, and it is the most useful failure in the subject.

Take a cylinder. Put it in a steady stream of fluid with no viscosity. Solve the flow exactly — and it can be solved exactly, in closed form, in about three lines.

Then work out the force along the stream.

It is zero.

d'Alembert's paradox, measuredSurface pressure round a cylinder in ideal flow, plotted against angle. The distribution is symmetric front to back, so every push on the front is matched by an equal push on the back, and the total force along the stream is exactly zero.frontback+1−30angle round the bodycomputed drag-1.1e-16not “small” — zeroideal flow — inviscid, irrotational, steadyany Reynolds number
Fig. 1 The pressure round the surface of a cylinder in ideal flow. Front and back are mirror images, so every push on the front is cancelled by an equal push on the back, and the total comes to about one part in ten thousand million million — which is arithmetic noise.

Not small. Not “negligible at low speeds”. Exactly zero, for any shape, at any speed, in any inviscid fluid. That is d’Alembert’s paradox, and taking it seriously is the beginning of understanding this subject.

What the flow looks like

The solution is genuinely lovely, which is part of the trouble.

Ideal flow past a cylinderA uniform stream past a circular cylinder in a fluid with no viscosity. The solution is exact and closed-form: streamlines part at a stagnation point, run round the surface and close up perfectly behind, and the pressure recovers to exactly what it was in front.ideal flow — inviscid, irrotational, steadyno circulation
Fig. 2 Ideal flow past a cylinder. The stream parts at a stagnation point on the nose, accelerates round the shoulders, decelerates again towards the rear, and closes up behind in perfect symmetry.

The air arrives, slows to a complete stop at the front, speeds up round the sides to twice the free-stream velocity, slows down again towards the back, and comes to a second complete stop at the rear stagnation point before departing. Front and back are mirror images.

Every streamline that opened ahead of the body closes behind it. There is no wake. Nothing is left behind at all.

Why the drag vanishes

The symmetry is the whole argument, and it can be seen without any calculation.

Pressure and speed trade off against each other along a streamline: where the flow is fast the pressure is low, and where it is slow the pressure is high. At the front stagnation point the flow has stopped, so the pressure is at its maximum. Round the shoulders the flow is fastest, so the pressure is at its minimum. Towards the rear the flow slows again and the pressure climbs back up — and by the rear stagnation point it has recovered to exactly the value it had at the front.

So the high pressure pushing backwards on the nose is matched, point for point, by high pressure pushing forwards on the tail. Integrate round the surface and the streamwise components cancel identically.

pnxds=0\oint p \, n_x \, ds = 0

The fluid gives the body a hard shove at the front and an equally hard shove at the back, and the body feels nothing.

The pressure recovers, and that is the false step

Everything above is correct arithmetic. The physical claim buried in it is the phrase the pressure recovers, and that is what real fluids refuse to do.

In a real fluid the air immediately next to the surface is not sliding past. It is stuck — the no-slip condition — and between the stuck layer and the free stream there is a thin region where the velocity climbs from zero to full speed. That region is the boundary layer, and the air inside it has lost energy to friction.

Fluid in the boundary layer therefore arrives at the rear half of the body with less to spend, and the rear half is exactly where it is being asked to climb back up a pressure gradient. Some of it cannot. It stops, reverses, and the flow lifts away from the surface entirely.

Once that happens the pressure never recovers. The rear of the body sits in a region of low, roughly constant pressure, the front is still being pushed hard, the cancellation is destroyed, and what is left over is drag.

What the ideal theory predicts, and what happensThe same cylinder, the same free stream. On the left the exact inviscid solution, closing up behind the body and exerting no drag at all. On the right the real flow at the same conditions, separated, with a wake and therefore with drag.ideal flow — closes up, no dragreal flow at Re 100 — separatedleft: exact closed form · right: solved on a gridRe = 100
Fig. 3 The same cylinder in the same stream. On the left the exact inviscid solution, closing up behind and exerting no force at all. On the right, a real fluid at the same conditions.

Every shape, not just this one

The cylinder is convenient, not special. The result holds for any closed body in an inviscid, irrotational, steady flow: a sphere, an aerofoil, a car, a cathedral.

The general argument does not need the symmetry that made the cylinder so easy. It runs through the fact that in an irrotational flow the far field decays fast enough that no momentum leaves the region containing the body, and if no momentum is leaving, no force is being exerted. The symmetry of the cylinder’s pressure distribution is one visible instance of a much more general accounting.

Which means the failure is general too. Every bluff body in a real fluid has separation and drag, and the ideal theory predicts none for all of them equally.

Ideal flow past a cylinderA uniform stream past a circular cylinder in a fluid with no viscosity. The solution is exact and closed-form: streamlines part at a stagnation point, run round the surface and close up perfectly behind, and the pressure recovers to exactly what it was in front.ideal flow — inviscid, irrotational, steadyno circulation
Fig. 4 The pressure field round the cylinder, contoured. The high-pressure regions fore and aft are the same colour because they are the same value — that identity is the whole of the paradox.

The one place the symmetry is visible

There is a spot on the figure where the argument becomes physical rather than algebraic: the rear stagnation point.

At the front, the flow stops. That is unsurprising — something has to give way for the body. At the back, in ideal flow, the flow stops again, having accelerated to twice the free-stream speed round the shoulders and then given all of it back.

Giving it all back is the suspicious part. It requires the fluid arriving at the rear to have exactly as much to spend as the fluid that set out from the front, which means nothing was lost on the way. In a fluid with no viscosity that is true by construction. In air it is not, and the deficit is precisely what the boundary layer has taken.

So the rear stagnation point is where the ideal theory makes its one physically unreasonable demand, and it is also the point where a real flow first refuses. Separation begins near there and works forward as conditions get harder.

What the solver computed

The number on the first figure is not quoted from a book. lib/flow.js builds the exact potential flow past a cylinder, walks round its surface in 720 steps, computes the local velocity, converts it to a pressure by Bernoulli, resolves against the outward normal and sums.

The result is 1.1×1016-1.1 \times 10^{-16}. Double-precision arithmetic carries about sixteen significant figures, so that is not a small drag — it is the largest number that can honestly be called zero on this machine.

The build asserts it. If the computed drag ever exceeded 10910^{-9} the figure would refuse to render, on the grounds that a “small” d’Alembert drag would mean the solver was wrong rather than that the paradox had softened.

The same routine, applied to a cylinder with circulation, returns a lift of 2.994 against the Kutta–Joukowski prediction of 3 — so the machinery is not simply returning zero for everything. It finds the force that is there and reports the absence of the one that is not.

A test the figure has to pass

The zero drag is asserted, but an assertion that only ever sees the correct answer proves nothing about the assertion. So the same machinery is pointed at a case where the answer is not zero.

A cylinder with circulation round it lifts. Kutta–Joukowski says the lift should be ρUΓ\rho U \Gamma; for a circulation of three in unit conditions, that is three exactly. The surface-pressure integral — the very same routine that returns 101610^{-16} for the drag — returns 2.994.

So the routine is not simply reporting zero for whatever it is given. It finds a force of three when there is a force of three, and reports nothing when there is nothing. That is what makes the zero worth believing.

Lift from a spinning cylinderA circular cylinder with circulation round it. There is no aerofoil section, no camber and no sharp trailing edge, and it lifts — which rules out shape as the explanation and leaves circulation as the thing that matters.lift 3.393= ρUΓ = 3.400drag 1.5e-16 — still zeroideal flow — no shape, only circulationΓ = -3.4
Fig. 5 The same integral applied to a lifting case. The lift is found and the drag is still zero, which is the combination the ideal theory always produces.

The paradox is not a paradox

The name is unfortunate. Nothing is self-contradictory here: an inviscid fluid genuinely would exert no drag, and the theory is right about the fluid it describes. What is wrong is the assumption that such a fluid is a good approximation to air.

And the assumption is nearly right, which is what makes it dangerous. Air has very little viscosity. Over almost the entire flow field the inviscid solution is accurate to a few percent, and the lift it predicts is close to correct. It fails only in a layer that may be a millimetre thick — and that layer decides drag entirely.

This is the thing worth carrying away from the whole subject: a model can be excellent nearly everywhere and useless for the one quantity that matters, because that quantity is decided in the small region the model got wrong.

The size of the error

It is worth being concrete about how bad the prediction is, because “zero” invites the response that the theory is only a bit optimistic.

A cyclist at thirty kilometres an hour is fighting perhaps eighty percent aerodynamic drag. Ideal flow says there is none, so it is not out by ten percent or a factor of two. It is out by all of it.

For a modern airliner, drag at cruise is what sets fuel burn, range, and whether the aircraft is economically possible at all. Ideal flow predicts the engines are unnecessary.

Yet the same theory, applied to the same aircraft, predicts the lift within a few percent. One theory, two quantities, one nearly right and one entirely absent. That combination is strange enough to be worth sitting with.

Shapes, and why streamlining works

If ideal flow gives no drag for any shape, it can say nothing about which shapes are good. Real flow says a great deal, and the difference between the two is entirely about where the flow separates.

A bluff body — a cylinder, a flat plate face-on, a lorry — makes the flow climb a very steep pressure hill immediately behind its widest point. The boundary layer gives up early, the separated region is wide, and the drag is large. Most of it is pressure drag: the low-pressure wake pulling backwards.

A streamlined body spreads that pressure recovery over a long tapering tail, so the hill is gentle enough for the boundary layer to climb. Separation is delayed almost to the trailing edge, the wake is thin, and the pressure drag nearly vanishes. What is left is skin friction over a rather large surface area.

That trade — a great deal less pressure drag for somewhat more friction — is what streamlining is, and it can be worth a factor of twenty. It is also entirely invisible to the ideal theory, which assigns both shapes the same drag: none.

Why it took so long to resolve

Jean le Rond d’Alembert published the result in 1752 and was candid that he could not explain it. It stood for over a century and a half, and it was not a minor curiosity — it was the reason a good deal of nineteenth-century hydrodynamics was regarded by engineers as elegant and useless.

Ludwig Prandtl resolved it in 1904, in a paper of eight pages, by pointing out that viscosity matters only in a thin layer near the surface and that the layer can separate. The rest of the flow could keep the inviscid treatment. That single idea rescued the theory by drawing a boundary round the part of it that worked, and it is why the twentieth century has aeroplanes in it.

The gap between the paradox and the resolution is a century and a half. It is worth remembering when a modern textbook presents the boundary layer as an obvious refinement.

What it cost the nineteenth century

The practical consequence of an unresolved paradox is worth recording, because it is easy to treat this as a curiosity that theorists worried about while engineers got on with it.

They did not get on with it. For most of the nineteenth century, hydrodynamics and hydraulics were two separate trades that barely spoke. Hydrodynamics was rigorous, elegant and predicted no drag. Hydraulics was empirical, unlovely, and worked. Engineers designing ships and pipes used tables and correction factors; the mathematics of the subject was, in the phrase of the time, the study of fluids that do not exist.

That split lasted until the boundary layer explained why both sides were right about their own half of the flow. It is the clearest case in physics of a theory being discarded by practitioners not because it was wrong but because nobody could say where it was wrong.

Reading a flow picture sceptically

One habit this essay should leave behind.

The ideal-flow figure is the most persuasive picture in this whole site. It is smooth, symmetric, manifestly the product of a real calculation, and it describes a situation that does not occur. If a figure that good can be that misleading, no flow picture should be trusted on the strength of looking right.

What can be trusted is a picture with its checks attached: this field conserves mass to one part in 10710^7, the velocity is tangent to the surface to fifteen decimal places, the force integral was computed two independent ways and they agreed. Those statements survive the picture being beautiful.

That is why every figure on this site carries a note saying which model produced it and at what regime, and why the essays keep returning to what the picture cannot show. A flow figure is worth exactly as much as the conservation laws it was checked against, and no more.

The two halves of the subject

It is worth placing this result in the shape of the whole field, because the paradox is not a curiosity at the edge of the theory — it is the hinge the subject turns on.

Everything upstream of it is ideal flow: exact, closed-form, linear, and correct about the bulk of any flow to a few percent. It gives lift, properly derived, and it gives the pressure field that everything else is built on.

Everything downstream of it is viscous flow: approximate, nonlinear, and correct about the thin region the ideal theory cannot see. It gives drag, separation and stall.

The paradox is what forces the division. Without it there would be no reason to suspect that a theory agreeing with experiment nearly everywhere was missing something essential, and no reason to look in a film a millimetre thick.

Flow past a cylinder at Re 40A real fluid past a circular cylinder. At low Reynolds number the flow closes up behind the body much as the ideal theory says; as it rises the flow separates and a region of reversed flow appears behind, which is where drag comes from.separatedrecirculation 0.56 Dviscous flow, solved on a coarse grid — the bubble is under-resolvedRe = 40
Fig. 6 The half of the subject the paradox opens. Here the flow does not close up, the pressure does not recover, and there is drag — none of which the exact solution can represent.
Reynolds number: one number, four different flowsReynolds number is inertia ÷ viscosity. It is not a property of the fluid or of the shape but of the combination, and crossing a threshold changes the physics rather than the magnitude.creepingattachedseparated, sheddingturbulentbacterium swimmingshedding begins, Re ≈ 47a thrown ballan airliner winga whaleReynolds numberinertia ÷ viscositylog₁₀ Rethe ratio decides the regime, not the size or the speed alone
Fig. 7 And the number that says which half applies. Ideal flow is a good description at the right-hand end of this axis and a poor one at the left, and the ratio is what decides.

Where the model stops

The paradox needs the flow to be steady, inviscid, incompressible and irrotational. Relax any of them and it does not apply.

It says nothing about lift. Circulation gives lift perfectly well in an inviscid fluid, which is why the ideal theory is still the right starting point for a wing.

“Zero” is exact only in the mathematics. The computed value here is 101610^{-16} because the arithmetic is finite, and a figure claiming a mathematically exact zero from a numerical calculation would be overstating what was done.

The comparison figure is coarse. The real flow beside it is solved on a build-time grid that resolves separation and does not resolve turbulence, and the wake it shows is correspondingly approximate.

The ladder from here

Next rungs: the thin layer that resolves the paradox, and why it is where all the friction lives; what separation is, and the adverse pressure gradient that causes it; and the comparison of ideal against real worked through at several Reynolds numbers.

Further out: pressure drag against skin friction, and which shapes are dominated by which; why a streamlined body has so much less drag than a bluff one of the same frontal area; and the reason a golf ball is dimpled, which is a deliberate manipulation of exactly the mechanism this essay is about.