The gas that leaves is not the gas inside

A still room contains molecules at every speed, and the distribution of those speeds is the same everywhere in it. This essay is about a place where a sample of that gas has a different distribution — and about how much follows from one factor of $v$.

The factor of v #

The distribution of speeds in a gas is Maxwell’s, proportional to $v^2 e^{-mv^2/2kT}$: the $v^2$ counts the directions available and the exponential is the Boltzmann factor.

Now ask which molecules reach a small hole in the wall in the next second. A molecule moving at $2v$ covers twice the distance in that second, so twice as many of them are close enough to arrive. The flux through the hole therefore carries an extra factor of $v$, and the beam’s distribution goes as

$$v^3 e^{-mv^2/2kT}$$

That is the whole of the physics. Everything below is a consequence.

Pressure is the everyday picture of what a gas does to a wall, and it is not what a small hole samples. A hole samples the flux — arrivals per second — and a fast molecule arrives more often than a slow one at the same density. That extra factor of $v$ is the whole essay: the escaping population is weighted toward the fast end, so what comes out is not a sample of what is inside.

The consequences of the extra factor are exact and checkable. The mean speed in the beam is $3\pi/8$ times the mean speed in the gas — 1.1781, reproduced by the numerical integration to four decimals. The mean kinetic energy in the beam is exactly $2kT$ against $\frac{3}{2}kT$ inside.

So an effusive beam is a third hotter than its source, in the only sense in which a beam has a temperature, and nothing has heated it. A slow leak also cools the gas that stays behind by more than an adiabatic expansion would, because the molecules that leave carry away more than their share.

The arrival rate at a wall is a quarter of the number density times the mean speed times the area. That standard result is the same quantity which becomes the effusion rate when the wall is given a hole, and the quarter is worth noticing because it is where the geometry lives — it is an average over directions and contains no dynamics at all.

The hole has to be small #

Everything above assumes molecules arrive at the hole independently, each carrying its own speed from wherever it last collided. That requires the hole to be smaller than the mean free path.

How far a molecule gets between collisions is 68 nanometres in air at atmospheric pressure and a metre or more in a good vacuum. The hole has to be small compared with that, and the reason is what distinguishes effusion from a leak: if molecules collide on their way through, the gas flows as a fluid and carries everything with it at the same speed, and the selection this essay is about disappears entirely.

Above that size the gas near the hole collides with itself, sets up a flow, and leaves as a fluid — with a bulk velocity, a pressure gradient and no speed bias at all. The dimensionless ratio is the Knudsen number, and effusion is the large-Knudsen limit.

At atmospheric pressure the mean free path is 68 nanometres, so effusion needs a hole smaller than that; at a millionth of an atmosphere the free path is centimetres and an ordinary aperture will do. That is why effusion is a vacuum phenomenon in the laboratory and a molecular-scale phenomenon at ordinary pressure — helium leaking through a balloon’s rubber is effusion through pores; air escaping a puncture is not.

How the rate is derived, and the quarter in front of it #

The escape rate is quoted everywhere as $\tfrac{1}{4}n\bar{v}A$, and the quarter is worth deriving because it is where the geometry lives and because it is the same quarter that appears in radiation and in reaction rates.

Molecules approach the hole from every direction in a hemisphere. A molecule travelling at angle $\theta$ to the normal contributes $v\cos\theta$ to the flux through the hole, and averaging $\cos\theta$ over a hemisphere weighted by solid angle gives $\tfrac{1}{2}$. Averaging over directions rather than over the full sphere gives another factor of $\tfrac{1}{2}$, since only the outward half can leave. The product is the quarter.

That decomposition is worth keeping because the same two factors appear whenever a flux is computed from an isotropic population: the emission from a blackbody surface, the rate at which molecules strike a catalyst, and the arrival rate of neutrons at a detector all carry a quarter for the same reason.

The quarter in the effusion rate is a count over directions and nothing else — no dynamics, no interaction, no property of the gas. Counting arrangements is what an angular average is, and it is worth saying so because the factor looks like it ought to encode something physical and does not. It would be the same number for any gas of any molecule at any temperature.

The number that results is large. A hole one micrometre across in a vessel of nitrogen at atmospheric pressure passes about 10¹⁴ molecules a second, which is why a vacuum system’s leak rate is quoted in pressure-volume units per second and why a leak of any visible size is fatal to a high vacuum.

The same quarter, for photons #

The claim that the quarter recurs elsewhere is worth making concrete, because the clearest instance has no gas in it at all.

A cavity at temperature $T$ contains radiation with an energy density $u$, travelling in every direction at $c$. Open a small hole in the wall and ask how much energy leaves per unit area per second. The calculation is the one above with the molecules replaced by photons and the speed distribution replaced by a single speed, and the answer is

$$\text{emissive power} = \tfrac14 u c,$$

with the same quarter and for the same two reasons — half the directions point outward, and the average of the cosine over those is a half.

That expression is what turns the energy density of a photon gas into the emissive power of a surface, and substituting the density gives the Stefan–Boltzmann law with its constant. Every calculation of how much a hot object radiates rests on it.

The parallel is exact enough to be worth stating as a definition. A blackbody is a hole in a cavity, and the reason a hole in a cavity is the standard realisation of one is precisely this argument: what leaves is a fair sample of what is inside, weighted by the same geometry as an effusing gas. The only difference is that photons have one speed rather than a distribution, so there is no $v^3$ bias — the beam and the cavity have the same spectrum, which is why the radiation from a small hole is the standard against which every other emitter is compared.

Which sharpens what the effusion bias actually is. It is not a property of holes; it is what happens when the population being sampled has a spread of speeds. Give every member the same speed and the bias disappears and the quarter remains.

Graham’s law, and what it cost #

The rate of escape is a quarter of the number density times the mean speed times the area, and the mean speed goes as $1/\sqrt{m}$. So light gases escape faster, in the ratio of the inverse square roots of their masses — Graham’s law, measured in 1848 before there was a kinetic theory to explain it.

That last number is the historically consequential one. Uranium-235 and uranium-238, as hexafluorides, have molar masses 349 and 352, and the separation factor per ideal stage is $\sqrt{352/349} = 1.00429$.

The arithmetic of a cascade then decides everything. Each stage multiplies the abundance ratio by 1.00429, so the number of stages needed is the logarithm of the required ratio change divided by the logarithm of 1.00429 — about 1,665 for weapons-grade material and a few hundred for reactor fuel. The K-25 plant at Oak Ridge, built for exactly this, enclosed 170,000 square metres under one roof and consumed a substantial fraction of the electricity generated in the United States.

Centrifuges replaced it because they separate by the mass difference itself rather than by the square root of the mass ratio, giving a per-stage factor of a few per cent instead of a few tenths of a per cent — which reduces a cascade of thousands to a cascade of tens.

A centrifuge is separation by mass with an effective gravity of a hundred thousand times the Earth’s. The mechanism is the one that makes an atmosphere’s scale height depend on molecular mass, and the enhancement is what makes it practical — but the separation per stage is still small, which is why enrichment plants have cascades rather than machines.

A pressure difference that does not go away #

The strangest consequence of the flux picture is a statement about equilibrium.

Join two vessels by a hole smaller than the mean free path and hold them at different temperatures. What comes to equilibrium is not the pressure — nothing is pushing on anything — but the flux each way. Equal fluxes means $n_1\bar{v}_1 = n_2\bar{v}_2$, and since $\bar{v} \propto \sqrt{T}$ and $P = nkT$, the steady state is

$$\frac{P_1}{\sqrt{T_1}} = \frac{P_2}{\sqrt{T_2}}$$

So the hotter vessel sits permanently at the higher pressure. It looks like a violation of the second law and is not: no cycle is available, nothing circulates, and the state is a true equilibrium of two subsystems that communicate only by molecular arrivals. The moment the hole is made larger than the mean free path the gas flows as a fluid and the pressures equalise.

The effect is called thermal transpiration, and it is a routine nuisance in vacuum work. A gauge at room temperature measuring a cryogenic vessel reads the wrong pressure by exactly $\sqrt{T_{\text{gauge}}/T_{\text{vessel}}}$, and the correction is applied as a matter of course in pressure metrology and in the calibration of the pressure standards that everything else is traced to.

Thermal transpiration maintains a pressure difference from a temperature difference and yields no work at all. That is worth setting against the efficiency ceiling: it is not an engine running below Carnot, it is a steady state with no cycle in it, and asking what work it produces is a category error rather than a disappointing answer.

Where the same bias appears #

In any measurement made by counting arrivals. A detector that samples a flux rather than a population is biased toward whatever arrives faster, and the correction is always one factor of the transport speed. Cosmic-ray detectors, neutron flux monitors and molecular-beam sources all carry it.

In the escape of planetary atmospheres. The molecules that leave the top of an atmosphere are the fast ones from the tail of the distribution, and the escape rate is dominated by them — which is why hydrogen and helium are absent from the Earth’s atmosphere and argon is not — the same competition between a thermal speed and an escape speed that thins the air with height decides it, and why the calculation is a flux-weighted integral over the tail rather than a comparison of mean speeds.

And in evaporation. A liquid loses its fastest molecules first, which is why evaporation cools and why a boiling point is a pressure rather than a property of the liquid alone — and the cooling is a flux-weighted effect of exactly this kind, though the sticking probability at a liquid surface makes the arithmetic messier than a hole in a wall.

Escape from a gravity well or from a liquid is an exponential in a barrier multiplied by a flux weighting — the same bias this essay is about, applied to a population already cut off at the low end. The second factor is the smaller of the two by a long way, which is why evaporation and atmospheric escape are usually treated with the exponential alone and why doing so slightly underestimates both.

The instrument the bias became #

An effusive source is not merely a leak; it is the standard way of making a beam of atoms or molecules, and its properties follow from the distribution drawn at the top of this essay.

Knudsen introduced the arrangement in 1909: an oven with a small aperture, whose emerging flux is calculable from the vapour pressure inside without any calibration. That makes it a primary source — a Knudsen cell measures vapour pressures by weighing what leaves, which is how the vapour pressures of refractory materials are known at all.

The same cell is the source in molecular-beam epitaxy, where the beam’s calculable flux is what allows a semiconductor to be grown one atomic layer at a time. Its angular distribution is a cosine law, which is what makes the deposit uniform over a wafer, and the cosine has the same origin as the quarter above.

Stern and Gerlach’s beam was effusive, and so was every atomic-beam experiment for the following half-century. That matters because the beam’s velocity distribution is not the gas’s: it is weighted by $v$, so the mean speed in the beam exceeds the mean speed in the oven, and any measurement whose answer depends on transit time inherits the bias. The instrument’s selection is part of its calibration.

There is a historical detail worth recording. Early atomic-beam experiments frequently used the wrong distribution, treating the beam as a sample of the gas, and the resulting velocity averages were systematically wrong by the factor derived here. The corrections appear in the literature of the 1930s as a series of increasingly exasperated notes.

Which gases a planet keeps #

The remark about atmospheric escape deserves its arithmetic, because it explains a composition rather than a rate.

At the top of an atmosphere — the height above which a molecule moving outward is unlikely to collide again — a molecule escapes if its speed exceeds the escape speed. Almost none does: the escape speed is several times the typical thermal speed, so the escaping molecules come from the far tail of the distribution, and the flux is dominated by the exponential there rather than by anything typical.

The quantity that decides everything is therefore a ratio of two energies: the gravitational binding of a molecule at that height against $kT$. Written as speeds, it is the escape speed against the most probable thermal speed, and the exponential makes the dependence brutal. A ratio of two and a half loses an atmosphere in a geological instant; a ratio of six retains it for the age of the solar system; and the range between those two covers a factor of well over a hundred in mass at a given temperature.

For the Earth, with an escape speed of 11.2 kilometres a second and an upper atmosphere near a thousand kelvin, hydrogen’s most probable speed is about 4 kilometres a second and the ratio is under three — so hydrogen leaves, and has. Nitrogen’s is under a kilometre a second and the ratio is nearly fifteen, so nitrogen stays. The line between them falls between helium and carbon, which is why the Earth has kept everything heavier and lost everything lighter.

The same arithmetic run on other bodies explains the pattern of the solar system without any further ingredient. Jupiter’s escape speed is five times the Earth’s and its upper atmosphere is colder, so it has kept its hydrogen. The Moon’s is a fifth of the Earth’s, so it has kept nothing. Titan is cold enough to keep nitrogen despite an escape speed lower than the Moon’s, which is the one case where the temperature rather than the mass is decisive.

And the escaping flux is the flux-weighted tail integral this essay is about, not a comparison of mean speeds — which matters, because the extra factor of $v$ weights the fast tail further and the answer would be wrong by a substantial factor without it.

Why leak detectors use helium #

One consequence is worth recording because it is the reason a particular gas cylinder stands beside every vacuum system in the world.

Finding a leak means finding a hole, and the useful signal is whatever comes through it. Two properties decide which gas to use. The rate through a small hole goes as $1/\sqrt{m}$, so a light gas gives the largest signal. And the gas has to be rare enough in the atmosphere that its arrival is unambiguous.

Helium wins on both. It is the second lightest gas, so it effuses about 2.7 times faster than nitrogen; it is chemically inert, so it does not react with anything or adsorb onto surfaces and confuse the reading; and it makes up five parts per million of the air, so a background is easy to distinguish from a signal. Hydrogen is lighter still and is used occasionally, and it has the drawbacks of being reactive, of being produced by surfaces themselves, and of being flammable.

The instrument on the other end is a mass spectrometer tuned permanently to mass four, so it responds to helium and to nothing else. Spray helium over the outside of an evacuated system, and the moment the jet crosses a leak the detector reads. The method finds leaks passing $10^{-11}$ pressure-volume units a second, which is a hole a few atoms across, and it is direct in a way no pressure-rise measurement is: it says not only that there is a leak but where.

Everything about that technique is one factor of the square root of the mass, put to work.

What the pictures cannot show #

The hole has no thickness. A real aperture is a short tube, and molecules that strike its walls are re-emitted in random directions, which changes both the rate and the beam’s angular distribution. Correcting for it is the Clausing factor, and it is a substantial correction for anything but a knife-edge.

The beam is not thermal in the transverse direction. Only the component of velocity along the hole gets the extra factor; a beam collimated by a second aperture is a quite different object from the full effusive distribution, and molecular-beam experiments quote which one they have.

The container is assumed to stay in equilibrium. If gas leaves faster than collisions can restore the distribution, the fast tail is depleted and the beam is cooler than the calculation says. That is the regime of a strongly effusing source, and it is why beam intensities cannot simply be scaled up by making the hole bigger.

And the isotope arithmetic assumes ideal stages. A real stage does not achieve the full separation factor, cascades need reflux, and the number of stages in a real plant exceeds the ideal count by a substantial factor. The 1,665 is a lower bound, and the fact that it is a lower bound is what made the enterprise so large.

The separation that runs the other way #

One more consequence is worth drawing out, because it inverts the usual reading of the same equations.

Everything above treats the bias as a fact to be corrected for or exploited. But run it in reverse: a gas mixture repeatedly effusing through a membrane full of small holes ends up separated, and the separation costs work — which must be at least the free energy of unmixing the two components.

That connects effusion to the thermodynamic floor on any separation. The gaseous-diffusion cascade at Oak Ridge consumed vastly more energy than that floor, not because the physics was wrong but because a process with a per-stage factor of 1.004 must recirculate its material enormously: nearly everything that enters a stage is returned to the one below, so the work of pumping dominates the work of separating by orders of magnitude.

A diffusion cascade is a population spreading by random steps with a slight bias per step, and the ratio of the bias to the spread is what decides how many stages are needed. With a separation factor of 1.004 per stage, enriching uranium takes thousands of them in series — which is the honest reason the technology is hard, and it is arithmetic rather than physics.

The general lesson is one about efficiency rather than about gases. When a separation factor per stage is close to one, the ideal work is unchanged and the practical work scales with the reciprocal of the separation factor minus one — so a process’s cost is decided by how far from unity its per-step advantage is, and not by the thermodynamics of the endpoints.

The ladder from here #

Later rungs on this anchor: the Knudsen number as the organising parameter, and the transition regime between effusion and viscous flow; Clausing’s calculation for a tube of finite length; the angular distribution of the beam, which is a cosine law and is what makes molecular-beam epitaxy uniform; and evaporation as effusion with a condensation coefficient in front of it.

The neighbouring ladders are the speeds in a still room, whose distribution is being reweighted here, how far a molecule gets, which decides whether any of this applies, and pressure as a rate of arrival, which is the same flux counted against a wall rather than through a hole.