What a flow is
Air is molecules. At room temperature and pressure there are about of them in a cubic metre, each moving at several hundred metres a second in a direction unrelated to its neighbours’, colliding every tenth of a nanosecond or so.
None of the rest of this site mentions them. Every figure here treats air as a continuous substance with a definite velocity at every point, and that treatment is not a rough approximation — it is accurate to many decimal places.
What a velocity at a point means
It cannot mean the velocity of the fluid at that point, because at a point there is either one molecule doing something erratic or nothing at all.
What it means is an average. Take a small volume round the point, add up the momentum of all the molecules inside it, divide by their total mass. That is the fluid velocity there.
The trick is that “small” has to be interpreted twice, in opposite directions. The volume must be small compared with the scale over which the flow varies — otherwise the average smears out the very structure being described. And it must be large enough to contain a great many molecules — otherwise the average is noisy and depends on which molecules happened to be inside.
Why both can be satisfied at once
The continuum assumption works because those two requirements are separated by an enormous margin.
A cube one micrometre on a side contains about 25 million air molecules — enough that the statistical noise in their average momentum is around one part in five thousand. And a micrometre is very much smaller than any scale over which the flow round an aircraft, a bird or a pipe varies.
So there is a window: volumes large enough to be statistically quiet and small enough to resolve the flow. Any size in that window gives the same answer, which is precisely what makes the field well-defined rather than an artefact of the averaging.
That window spans several orders of magnitude in most practical flows. When it does, the fluid can be treated as a continuum and the molecular picture can be forgotten entirely.
The number that closes the window
The window closes when the molecular scale approaches the flow scale, and there is a ratio that says when.
The Knudsen number: the molecular mean free path — the average distance between collisions — divided by a length characteristic of the flow. In air at sea level the mean free path is about 68 nanometres.
For a wing with a chord of a metre, , and the continuum treatment is beyond question. For flow in a channel a micrometre wide it is about 0.07, and the assumption is starting to fail. Above about 0.1 it has failed and something else is needed.
This is the same style of argument as the Reynolds number: form a ratio, find that a threshold in it separates regimes, and note that neither ingredient on its own says anything.
Where it breaks
Three regimes where the continuum description stops being available.
Very small. Micro-channels, MEMS devices, the gaps in porous media. The mean free path becomes comparable to the geometry, and even the no-slip condition fails — the gas slips at the wall.
Very thin. The upper atmosphere. At 100 kilometres the mean free path is centimetres; at 200 kilometres it is hundreds of metres. A re-entering vehicle passes from free-molecular flow, where molecules strike it individually and never interact with each other, through a transitional regime, to continuum flow — and needs different physics at each stage.
Very fast, locally. The interior of a shock wave is a few mean free paths thick. Continuum equations describe what is on either side of a shock correctly and cannot describe its internal structure at all, which is why shocks are treated as discontinuities rather than resolved.
Five numbers, and that is all
Once the continuum assumption is granted, the entire state of a fluid at a point is five numbers.
Three components of velocity, a pressure, and a density. For a compressible flow a temperature joins them, related to the others by an equation of state. That is the whole description.
It is worth pausing on how strong that claim is. Every property of the fluid at a point that has any consequence for its motion — the momentum of molecules, their distribution of speeds, the frequency of their collisions — is captured by five numbers, because everything else has averaged away.
The molecular details do not vanish entirely; they reappear as material properties. Viscosity is a number that summarises how effectively molecular motion transports momentum sideways. The speed of sound summarises how quickly a pressure disturbance propagates. Both come from the molecular picture and both enter the continuum equations as constants to be measured.
So the continuum description is not ignorant of molecules. It has compressed them into two or three coefficients, and it is that compression that makes the subject tractable.
What viscosity actually is
Since it is the property this site cares about most, it is worth saying where it comes from.
Consider two adjacent layers of gas sliding past each other at different speeds. Molecules cross between them constantly in random thermal motion. A molecule crossing from the fast layer to the slow one carries extra momentum with it and speeds the slow layer up; one crossing the other way slows the fast layer down.
The net effect is a transfer of momentum across the flow direction, which is exactly what a shear stress is. Viscosity is the coefficient relating that transfer to the velocity gradient.
Two consequences follow that are hard to guess otherwise. Gas viscosity increases with temperature, because hotter molecules cross between layers faster — the opposite of a liquid, where viscosity falls as it warms. And gas viscosity is nearly independent of pressure, because doubling the density doubles the number of carriers and halves the distance each one travels between collisions.
Both are experimental facts and both fall out of the molecular picture immediately, which is a good demonstration that the two descriptions are consistent rather than rival.
What the solver computed
Everything on this site assumes a continuum, and the assumption is so far from failing in these flows that it is never checked.
That is worth being explicit about, because the site’s habit is to check things. The Knudsen number for every figure here is of order or smaller. There is no meaningful sense in which the continuum assumption is marginal, and an assertion testing it would always pass and prove nothing.
The assumptions that are checked are the ones that could fail: mass conservation, tangency at a wall, and the agreement of two independent routes to the lift. Those are checked because the code could get them wrong, which is a different question from whether the physics is applicable.
Knowing which assumptions are at risk and which are not is part of using a model, and testing an assumption that cannot fail is a way of looking careful rather than being careful.
What a field buys
The payoff for the continuum assumption is worth naming, because it is the reason the subject exists in its current form.
Differential equations become available. A field can be differentiated; a cloud of molecules cannot. Everything from continuity to Navier–Stokes is written in terms of derivatives of fields, and none of it could be written otherwise.
The number of variables collapses. Tracking molecules is not merely hard, it is meaningless — the answer would be a list nobody could read. Five fields describe the same physics for every purpose anybody has.
Averages become exact. Pressure, temperature and density are statistical quantities, and in the continuum limit they behave as though they were properties of a substance.
That last one is worth sitting with. Pressure is molecular bombardment, and in this limit it can be treated as a smooth function with a gradient — which is what makes Bernoulli’s equation and every force calculation on this site possible.
Fluid particles, and a word of caution
The phrase “fluid particle” appears throughout this subject, including elsewhere on this site, and it needs a note.
A fluid particle is not a molecule. It is a small parcel of the continuum — large enough to contain many molecules, small enough to be treated as a point — and it is followed as though it retained its identity while being carried, stretched and deformed by the flow.
That is a useful fiction and it is a fiction. Molecules diffuse in and out of any such parcel constantly, so a fluid particle does not consist of the same molecules from one moment to the next. What it retains is not its matter but its place in the continuum description.
For most purposes the fiction is harmless: the pathlines and streaklines drawn on this site follow fluid particles in exactly this sense, and they are meaningful curves.
It matters where diffusion matters. A dye filament in a flow spreads as well as being carried, and over long times the spread can dominate. A parcel of fluid tracked for long enough stops being a parcel at all, which is why streakline visualisations blur downstream and why the useful description of a turbulent flow is statistical rather than particle-based.
Density, pressure, temperature
Three quantities that this description treats as properties of a substance, and all three are statistical.
Density is the easiest: mass in a small volume divided by that volume. The averaging window is the same one discussed above, and the noise falls as the square root of the number of molecules inside.
Pressure is momentum transfer. Molecules strike a surface and rebound, each delivering an impulse; the pressure is the average rate of momentum delivery per unit area. There is no continuous push, only a very rapid succession of impacts — about 0^{27}$ per square metre per second at atmospheric pressure, which is why it averages so smoothly.
Temperature is the mean kinetic energy of molecular motion, once the bulk flow has been subtracted. That subtraction matters: a fast-moving parcel of air is not hot on account of moving. Temperature is about the random part of the motion, and the bulk part is the velocity field.
The separation of molecular motion into a bulk average and a random remainder is what makes the whole description work, and it is the same separation that gives the speed of sound its role: the random motion is what carries a pressure signal, so its speed sets how fast news travels.
Where the model stops
The window can close, and the Knudsen number says when.
No-slip is not fundamental. It is an experimental fact about continuum flow at a surface, and it fails in rarefied conditions where a gas genuinely slips.
Shocks are not resolved. The continuum equations give the states either side and not the structure between.
Nothing here is molecular. Temperature, viscosity and the speed of sound all have molecular origins this site takes as given.
What a picture of a flow is a picture of
A consequence for reading every figure on this site, and it follows directly.
A streamline is not a thing in the air. It is a curve constructed from a field, which is itself constructed from an average over molecules. There is no filament of anything running along it, and nothing is travelling down it in the way the picture suggests.
The same is true of a contour of pressure, a separation point, and a vortex. Each is a feature of a mathematical object that describes the average behaviour of a very large number of molecules — real in the sense that it has measurable consequences, and not real in the sense that it could be pointed at.
That is not a reason to distrust the pictures. It is a reason to be clear about what they claim: they are statements about a field, checked against the conservation laws that field must satisfy, and their authority comes from those checks rather than from resembling anything.
It also explains why a smooth picture proves nothing. Smoothness is a property of the mathematical object, and every field — right or wrong — has it, because the averaging that produced it removed everything rough.
The hierarchy of descriptions
Worth setting out, because this essay sits at the bottom of it and the rest of the site sits above.
Molecular. Individual particles, collisions, distribution functions. Exact, and useless for anything larger than a micrometre.
Kinetic. The statistical distribution of molecular velocities, evolved by the Boltzmann equation. Necessary in rarefied flow, and the origin of the transport coefficients used below.
Continuum. Fields of velocity, pressure and density obeying Navier–Stokes. This is where nearly all fluid mechanics lives, including everything on this site.
Inviscid continuum. The same, with viscosity dropped. Solvable in closed form, and wrong about drag.
Each level is derived from the one above it by averaging, and each buys tractability with a loss of detail. The skill of the subject is largely knowing which level a question needs — and the answer is almost always the third, which is why the third is what gets taught.
Who established it, and when
The continuum treatment is older than the molecular one, which is the wrong way round and explains a good deal. Euler wrote the equations of fluid motion in the 1750s; the kinetic theory that justifies treating a gas as a continuum is Maxwell’s and Boltzmann’s, a century later.
So for a hundred years the equations of fluid mechanics were used successfully by people who did not know whether matter was made of atoms, and the argument about atoms was still live into the 1900s.
The Knudsen number is Martin Knudsen’s, from work on rarefied gases around 1909 — which is to say the limit of the assumption was quantified only after the assumption had been in productive use for a hundred and fifty years.
Why this is the first essay and not the last
A note on ordering, since the continuum assumption is logically prior to everything and is rarely taught first.
The reason is that it is invisible when it works, and it works essentially always. A student meeting fluid mechanics needs streamlines, pressure and forces long before needing to know that all three are averages, and introducing the averaging first makes the subject look harder than it is.
The cost of that ordering is a vague sense that the continuum is an idealisation and therefore suspicious. It is not suspicious. At the scales in question it is accurate to a part in a thousand million, which is far better than the accuracy of anything else in the calculation — the geometry, the material properties, or the models of viscosity and separation layered on top of it.
So the honest framing is that the continuum assumption is the least questionable thing on this site. The approximations worth worrying about are the ones made afterwards: dropping viscosity, assuming steadiness, treating density as constant. Those are the ones with visible consequences, and every essay here names them.
The ladder from here
Nearby: the kinetic origin of viscosity and pressure; slip flow and the transition regime; free-molecular flow and re-entry; and the internal structure of a shock.
Then across to the curves drawn through a field and what the field must satisfy.