The theory that solves everything

The full equations of fluid motion have been known since the 1840s and cannot be solved. Not “are hard to solve” — there is a million-dollar prize outstanding for showing that solutions even exist and stay smooth.

Drop two terms and everything changes. The equations become linear, solutions can be written in closed form, and problems that are otherwise intractable reduce to a few lines of algebra.

That is ideal flow, and it is worth understanding both why it is so powerful and what it cost.

The two assumptions #

Inviscid. The fluid has no internal friction. The viscous term in the momentum equation vanishes, and with it the highest derivative.

Irrotational. No fluid element is spinning about its own axis. This one sounds like an extra restriction and is closer to a consequence: a flow that starts from rest in an inviscid fluid stays irrotational for ever, by Kelvin’s theorem. Vorticity has no way to get in.

Together they are transformative. Irrotational flow can be written as the gradient of a single scalar potential; substituting that into the continuity equation gives Laplace’s equation:

$$\nabla^2 \phi = 0$$

which is one of the most thoroughly understood equations in mathematics.

What that buys #

Four things, and each is large.

Linearity. Solutions add. That gives the whole construction kit of elementary flows — streams, sources, doublets, vortices — and the ability to build complicated flows from simple pieces.

Closed form. The flow past a cylinder is three terms. The flow past an aerofoil is those three terms mapped through a conformal transformation. No iteration, no grid, no convergence to worry about.

A century of mathematics for free. Laplace’s equation appears in electrostatics, heat conduction, gravitation and complex analysis. Every result about it transfers, and the theory of complex variables in particular becomes a tool for solving fluid problems.

One unknown instead of three. A scalar potential replaces a vector velocity field, and in two dimensions a stream function does the same job with the added property that its contours are the streamlines.

For a subject whose full equations resist analysis entirely, that is an enormous amount to gain from dropping two terms.

What it costs #

One thing, and it is the subject of a separate essay: the theory predicts that nothing has any drag.

Not approximately. Exactly zero, for every shape, at every speed. A cyclist needs no effort, an airliner needs no engines, and a parachute does not work.

That is not a small discrepancy to be corrected later. It is the complete absence of the quantity that most of the practical subject is about.

The stream function, and why two dimensions are special #

One consequence of the assumptions deserves its own note, because it is why so much of classical aerodynamics is two-dimensional.

In two dimensions, an incompressible flow can be described by a single scalar $\psi$ whose derivatives give the velocity components. Its contours are the streamlines — so plotting $\psi$ is drawing the flow.

Two things fall out of that at once. Mass conservation is satisfied automatically, since the divergence of a curl vanishes identically, so it does not have to be imposed. And the whole velocity field is one function of two variables rather than two.

Add irrotationality and there is a second scalar, the velocity potential, whose contours are everywhere perpendicular to the streamlines. The pair of them are the real and imaginary parts of a single analytic function of a complex variable, which is why complex analysis is the natural language for this subject and why conformal mapping works at all.

None of this survives in three dimensions, where the stream function has no straightforward analogue. That is a large part of why the classical theory is a theory of sections rather than of wings, and why finite wings needed a separate and later development.

What “irrotational” really excludes #

The assumption is easy to state and its content is easy to miss.

Irrotational does not mean the flow is straight, or that it does not go round corners, or that there are no vortices in the everyday sense. Fluid can travel in perfect circles and be irrotational, provided the speed falls off as one over the radius — which is exactly what a point vortex does.

What it excludes is local shear: a fluid element being deformed such that it acquires spin. A paddle wheel dropped into an irrotational flow orbits without turning about its own axis.

The reason that matters is that shear is precisely what a wall produces. No-slip means the fluid at the surface is stationary and the fluid just above is not, and that is shear, and shear is vorticity.

So the irrotational assumption is not an extra idealisation on top of inviscidness. The two are the same restriction seen from different sides: no viscosity means no mechanism to generate vorticity, and no vorticity means the potential formulation works.

Why the failure is so specific #

The interesting question is how a theory can be so accurate about the velocity field and so wrong about the force, and the answer is a matter of where the assumptions fail.

Viscosity is genuinely negligible over almost the whole flow. Air’s viscosity is tiny, and away from surfaces the velocity gradients are gentle, so the viscous term really is small there. The inviscid solution is a good description of most of the flow.

Next to a surface it is different. The no-slip condition forces a steep velocity gradient in a very thin region, and a small coefficient multiplying a large derivative is not small. That layer is where viscosity lives, and it is where the irrotational assumption fails too, since shear is vorticity.

So the theory is excellent in the region that occupies almost all the volume and wrong in the region that determines the force. Volume and importance are not the same thing, which is the lesson.

A singular failure, not a small one #

There is a technical reason the failure is qualitative rather than proportional, and it is worth having because it explains why nobody found the fix by refining the theory.

Setting the viscosity to zero does not merely make a term small. It removes the highest derivative from the equation, which lowers its order — and an equation of lower order can satisfy fewer boundary conditions.

The full equations can enforce both no flow through a wall and no flow along it. The inviscid equation can enforce only the first. So the no-slip condition is precisely the one that gets dropped, and every consequence of no-slip disappears with it.

This is a singular perturbation: a small parameter whose removal changes the problem’s character rather than its answer. Such problems cannot be fixed by adding correction terms, which is why a century and a half of trying produced nothing and why the resolution had to come from a different direction entirely.

Two dimensions, and the price of them #

Nearly everything in this essay is two-dimensional, and that restriction is worth making explicit.

A section is an infinite wing. It has no tips, so the circulation bound to it never has to end, and Helmholtz’s theorem that a vortex line cannot terminate in the fluid is satisfied trivially.

A real wing has ends. The bound circulation must turn and trail downstream from each tip, and those trailing vortices induce a downwash over the wing that tilts the oncoming flow. The lift vector tilts back with it, and the backward component is induced drag — a drag that exists in perfectly inviscid flow, and which the two-dimensional theory therefore cannot see.

So ideal flow does predict a drag after all, but only in three dimensions and only for finite wings. That is a genuine prediction and a good one, and it sits oddly beside the same theory’s insistence that a two-dimensional body has none.

The resolution is that the two drags have different origins. Induced drag is the cost of leaving momentum in the air; the drag the theory misses is the cost of the boundary layer giving up. One is inviscid and one is not, and a real wing has both.

What the solver computed #

Every ideal-flow figure on this site is a closed-form expression evaluated directly. There is no grid, no timestep and no convergence criterion — the velocity at a point is computed from a formula.

That makes the accuracy checks unusually sharp. Mass conservation comes out at $3.3 \times 10^{-8}$, which is entirely the finite-difference error in the check rather than in the field, since the analytic solution is divergence-free identically. Surface tangency comes out at $7 \times 10^{-15}$, which is machine precision.

Those numbers are worth contrasting with the viscous solver’s, where the field is a numerical approximation and the errors are real. Here the only error is in the measurement.

It also means the failures are not numerical. When this theory gives zero drag, that is the theory speaking and not the arithmetic — which is why the site asserts the drag is below $10^{-9}$ rather than merely observing that it looks small.

The theory that engineers ignored #

The historical consequence is worth recording, because it is unusual for a correct theory.

Through the nineteenth century, hydrodynamics and hydraulics were separate trades. Hydrodynamics was mathematical, elegant, and predicted no drag. Hydraulics was empirical, unlovely, full of correction factors, and worked.

Engineers building ships, pipes and turbines used the tables. The mathematics of the subject was described — the phrase is contemporary — as the study of fluids that do not exist.

That split lasted until somebody explained where each side was right. It is the clearest case in physics of a theory being abandoned by practitioners not because it was wrong, but because nobody could say where it was wrong — and the moment that could be said, it became indispensable again.

What it predicts correctly #

Since this essay is largely about a failure, the successes are worth listing so the picture is fair.

The pressure over the front of any body. Good to a few percent for attached flow, and this is what a boundary-layer calculation takes as its input.

Lift. Circulation gives it correctly, and the lift-curve slope near $2\pi$ is a prediction rather than a fit.

The far field. Away from surfaces, the real flow is very nearly what the theory says.

Added mass. When a body accelerates through a fluid it has to accelerate some of the fluid with it, and the effective extra mass is calculable exactly from the potential solution. This one is often forgotten and matters for airships, submarines and anything manoeuvring.

Wave-making, in the free-surface version. Ship wave resistance is predicted by potential theory reasonably well, because it is an inviscid phenomenon.

That is not the record of a discarded theory. It is the record of one whose domain is known.

What replaced it, and what did not #

A closing note on the theory’s status today, since it is a century past its resolution.

It was not replaced. Numerical solution of the full equations is now routine, and for a final design that is what gets used — but ideal flow remains the first tool reached for, because it is instant and because its answers can be reasoned about.

More importantly, its concepts were not replaced at all. Circulation, the Kutta condition, the velocity potential, added mass, induced drag: every one of them comes from this theory, and every one is used to interpret the output of a modern calculation. A numerical result is a table of numbers until somebody says “the circulation has dropped, so it has stalled”.

So the theory survives in two roles: as a fast approximation, and as the vocabulary in which results are discussed. The second is the more durable, and it is the reason a subject with far better tools still teaches a model that predicts an aeroplane needs no engines.

Where the model stops #

No drag, which is most of the practical subject.

No separation, no stall, no wake. All require vorticity, which the theory excludes by construction.

Irrotationality fails downstream of anything. Once a body has shed vorticity, the flow behind it is not irrotational and the theory does not describe it.

It gives no Reynolds number. The equations have no viscosity in them, so they predict the same flow at every speed and in every fluid — and the fact that reality does not behave that way is visible as a prediction that stays put while experiments move.

The moral, which is not about fluids #

The shape of this theory’s success and failure is worth extracting, because it recurs far outside the subject.

A model was simplified by removing a term that was genuinely small. The simplification made the problem solvable and the solution was accurate over almost the entire domain. And the one quantity the model was wanted for was decided entirely by the term that had been removed.

Three lessons follow.

Small does not mean unimportant. A term with a small coefficient dominates wherever its derivatives are large, and that can be a region of vanishing size and total importance.

Accuracy over most of a domain says nothing about accuracy of an integral. The velocity field was right nearly everywhere and the force was wrong by all of it, because the force is decided where the field is wrong.

A model that fails should be bounded, not patched. A century of adding correction terms achieved nothing; drawing a boundary round the region where the assumptions held, and solving a different problem inside it, achieved everything.

The third is the one worth carrying. When a good model fails somewhere, the productive question is usually “where exactly, and what is different there” rather than “what term is missing”.

Reading the figures on this site #

A practical note, since roughly half the figures here come from this theory and half do not.

Ideal-flow figures carry no Reynolds number, because the equations contain none. If a figure on this site states a Reynolds number, it came from the viscous solver; if it does not, it is an exact solution and describes every speed equally — which is both its strength and its defect.

Both kinds carry a model note stating which assumptions produced them. That note is not decoration: an ideal-flow figure showing beautiful attached flow round a bluff body is a picture of something that does not happen, and the note is what says so.

The habit that goes with it is to read the note before the picture. A smooth streamline pattern proves nothing at all, and the assumptions are the only thing that says what a figure is a figure of.

Who built it, and when #

Euler wrote the inviscid equations in 1757, and they were the first general equations of fluid motion. Laplace’s equation and the potential formulation follow in the decades after. Helmholtz’s vortex theorems arrive in 1858, Kelvin’s circulation theorem in 1869.

By the 1870s the theory was essentially complete: a beautiful, closed, self-consistent account of the motion of a fluid that has no friction. It took another thirty years for anybody to explain why it was useless for calculating drag, and the explanation turned out to be eight pages long.

The ladder from here #

Next rungs: the construction kit and how bodies are built from singularities; the velocity potential and stream function; the paradox in detail; and Kelvin’s theorem, which is what guarantees the irrotationality the whole thing rests on.

Then across to the layer where it fails, and to the comparison of the two theories side by side.