A coefficient that belongs to the surface, not the satellite

The density of the air at four hundred kilometres is the quantity that decides a low orbit’s future, and it is not computed. It is fitted, to decades of observed satellite drag, through the drag equation

$$a_{\rm drag} = \frac{1}{2}\,\rho\,v^2\,\frac{C_D A}{m}.$$

The measurement is of the acceleration. Everything else in that expression has to be supplied before a density comes out, and three of the four are known: the speed from the orbit, the area from the design, the mass from the launch manifest. The fourth is $C_D$, and it is not a property of the satellite at all.

It is a property of the collision — of what a molecule does after it hits. And the conventional value everybody uses, 2.2, is a fitted average for satellite shapes rather than a measured coefficient for any particular one.

Why the flow is not a flow #

At sea level a molecule travels seventy nanometres between collisions and a body moving through air pushes a continuous fluid: the molecules near the surface talk to each other, a boundary layer forms, and the drag coefficient is a matter of shape and Reynolds number.

At four hundred kilometres the mean free path is of order a kilometre. A satellite is metres. So a molecule striking the surface has no idea that any other molecule is doing the same, and nothing it does afterwards affects what the next one does — the flow is free molecular, and the drag is a sum over independent collisions rather than a property of a fluid.

That simplification is enormous. There is no boundary layer, no separation, no wake in the aerodynamic sense. The drag is computed by integrating the momentum each molecule delivers over the incident flux, and for simple shapes the integral has a closed form.

What it does not remove is the dependence on the collision itself, and that is the whole subject.

Two ways to leave, and they are not equivalent #

A molecule arrives with momentum. What it takes away depends on how it goes.

Specular reflection is a mirror bounce: the molecule keeps its speed, reverses the component normal to the surface, and keeps the tangential one. The momentum transferred is twice the normal component. For a flat plate face-on in a hypersonic flow, every molecule delivers twice its own momentum, which is where the number two in $C_D \approx 2$ comes from.

Diffuse reflection is the opposite: the molecule adsorbs, thermalises with the surface, forgets where it came from, and leaves in a cosine distribution at the surface’s own temperature. It transfers all of its incident momentum, and then takes away some more in a direction unrelated to the flow — which adds to the drag, because the re-emitted flux has its own momentum flux and the surface has to push against it.

Real surfaces do both, and the fraction that thermalises is the accommodation coefficient. It is a number between zero and one, it depends on the surface material and on what is adsorbed onto it, and it is the quantity this essay is about.

For a sphere the two limits have closed forms — Schaaf and Chambre’s, in terms of the speed ratio $s = V/\sqrt{2RT_\infty}$ — and the figures here interpolate between them.

What makes the surface diffuse #

The reason the accommodation coefficient is not a material constant is that the surface is not the material it was launched as.

The thermosphere at four hundred kilometres is dominated by atomic oxygen, produced by the photodissociation of molecular oxygen higher up and prevented from recombining by the low density. Atomic oxygen is chemically aggressive. It erodes polymers, oxidises metals, and — most importantly here — adsorbs onto every exposed surface within hours of launch.

An oxygen-covered surface is close to fully accommodating: the arriving atoms are chemically similar to the ones already there, they stick, they thermalise, and they leave diffusely. So a satellite that has been in orbit for a day has a higher drag coefficient than the same satellite on the pad, and the difference is not small.

The coverage depends on the local oxygen density, which depends on altitude and on solar activity. So the accommodation coefficient varies with the thing being measured, and a drag-derived density carries a coefficient that was itself a function of the density.

That circularity is not vicious, but it does mean the systematic is not a constant offset. Above about five hundred kilometres the oxygen density is low enough that the coverage is partial, the accommodation falls, and the drag coefficient falls with it — by tens of per cent over the altitude range where most satellites are.

Where the number two comes from #

It is worth deriving the leading term once, because the familiar $C_D \approx 2$ is often quoted as though it were empirical and it is not.

Take a flat plate facing the flow at hypersonic speed, so that the molecules’ thermal motion is negligible against the bulk velocity and every one of them arrives along the same direction. Each carries momentum $mV$ toward the plate.

Under specular reflection each leaves with momentum $mV$ away from it. The change is $2mV$ — twice the incident momentum, delivered per molecule.

The drag force is that change times the flux, which is $nV$ per unit area, so the force per unit area is $2nmV^2 = 2\rho V^2$. Comparing with the definition $F = \frac{1}{2}\rho V^2 C_D A$ gives $C_D = 4$ for a plate.

For a sphere the same calculation carries a geometric factor, because most molecules strike at an angle and reflect at that angle rather than straight back, and integrating over the hemisphere gives exactly 2.

So the two is a geometric integral, not a measurement, and it applies to a mirror-reflecting sphere in a flow with no thermal spread. Everything in this essay is the corrections to those two idealisations: the thermal spread, through the speed ratio, and the reflection, through the accommodation.

That is why the departures are modest — tens of per cent rather than factors — and why they matter anyway. A modest systematic shared by every measurement of a quantity is worse than a large random one.

The consequence, which is a factor on everything #

The propagation is direct. The drag equation is measured for the acceleration and solved for the density, so

$$\rho = \frac{2\,m\,a_{\rm drag}}{C_D A v^2},$$

and an error in $C_D$ is the same fractional error in $\rho$, with the opposite sign.

Every density in every empirical thermospheric model inherits it, because the models are fitted to drag-derived densities. When the community’s adopted value moved from 2.2 to something altitude- and solar-activity-dependent — closer to 2.1 at four hundred kilometres and rising to 2.4 in the lower thermosphere — every historical density shifted by several per cent, in a way that varied with altitude.

Two further consequences are worth separating because they behave differently.

For a single object’s orbit prediction, the coefficient does not matter much, because the fitted ballistic coefficient absorbs it. Tracking determines the product of the density and the inverse ballistic coefficient, and whatever $C_D$ was assumed is folded into the fitted value.

For a model of the atmosphere, it matters completely. The model is a statement about the density, and getting there from a drag measurement requires the coefficient explicitly. A systematic in $C_D$ shifts the whole model, and because it is shared by every object it does not average down with the number of them.

What has actually been measured #

Three kinds of experiment bear on the coefficient and none of them is conclusive on its own.

Laboratory beams. A beam of atomic oxygen at orbital energy is directed at a sample and the scattered distribution is measured. This is the direct approach and it is hard: producing a beam at five electronvolts with the right flux is a specialised business, the samples are small, and reproducing the contamination state of a real surface after months in orbit is not possible.

Spherical satellites. A sphere has no attitude and a known area, which removes two of the unknowns at once. Several have been flown precisely for this — smooth, dense, passive spheres tracked for decades, which are also how the Earth’s own shape was read off a node — and their drag histories are the cleanest constraint on the product of density and coefficient there is. What they cannot do is separate the two.

Paired objects. Two objects of very different shape at the same altitude and time experience the same density, so the ratio of their fitted ballistic coefficients constrains the ratio of their drag coefficients. That is a genuine separation and it is weak, because the shapes differ in more ways than one.

The modern practice is a fourth thing: a physical model of the interaction, with accommodation computed from an adsorption model driven by the local atomic-oxygen flux, validated against the spheres. It reproduces the spheres well and it contains a coverage model nobody has measured in orbit.

Why the constellations change the arithmetic #

Something has changed in the last few years that bears directly on this, and it is not a measurement technique.

Large constellations in low orbit consist of thousands of identical satellites, whose masses and attitudes the operator knows exactly, carrying navigation receivers that determine their positions continuously to centimetres, which is a position measured from a frequency. Each one is a drag sensor with a known area and a known ballistic coefficient — everything in the drag equation except the density and the coefficient.

Thousands of them at different altitudes and local times, all of the same design, is a dataset of a kind the field has never had. The shape systematic is shared and constant, so the variation between altitude and local time is a clean measurement of the density variation; and the absolute calibration can be tied to the spheres.

The obstacle is not technical. The data are proprietary, and the question of whether operational drag data from commercial constellations becomes a scientific resource is a matter of arrangements rather than of instruments.

The one place it was measured in orbit #

There is a single class of experiment that separated the coefficient from the density in flight, and it is worth describing because it shows what such a measurement has to do.

A satellite carrying a sensitive accelerometer measures the drag acceleration directly, without going through an orbit fit. Combined with a known mass and area, that gives the product $\rho C_D$ at every point of the orbit at high spatial resolution — which is the same product a drag-derived density gives, so on its own it separates nothing.

What separates them is a second body with a different shape at the same place and time. Several missions have flown with a spherical calibration element, or in formation with a differently shaped companion, so that the ratio of the two accelerations depends on the ratio of their coefficients and not on the density at all.

The results confirm that a real satellite’s coefficient is higher than a sphere’s, that it varies with altitude in the direction the adsorption model predicts, and that the variation is tens of per cent. They do not pin the absolute value, because the ratio of two coefficients is not either of them.

A ratio measurement removes the shared unknown and cannot supply an absolute scale, which is the same limitation every differential technique in this collection carries, and the reason the absolute calibration still rests on the passive spheres and on a model.

Where the picture stops #

A sphere is not a satellite. Every curve here is for a sphere, which is the one shape with a closed form. A real satellite has flat panels, and a flat plate’s coefficient depends on the angle it presents to the flow — so a satellite’s coefficient depends on its attitude, and a tumbling object’s depends on the time average of its attitude, which nobody knows.

Accommodation is one number standing for two. The momentum a molecule takes away depends on how well it thermalised in speed and on how completely it forgot its direction, and those are separate quantities with separate coefficients. The single accommodation coefficient used everywhere is a convenient collapse of both, and it is not clear the collapse is safe at the per-cent level the models now claim.

And the flow is not quite free molecular below about two hundred kilometres. The mean free path there falls toward the size of a spacecraft, and the transitional regime between free molecular and continuum flow has no closed form at all — which is exactly the altitude range a re-entering object spends its last orbits in.

What the disagreements between techniques amount to #

The three ways of measuring thermospheric density do not agree, and the drag coefficient is the largest single reason.

Drag-derived densities carry the coefficient explicitly, so any error in it shifts them all by the same factor. This is the enormous historical dataset and it is the one the models are fitted to.

Accelerometer densities carry the same coefficient, because they still divide by it — what they remove is the orbit-fit uncertainty rather than the aerodynamic one. They are better data about the same product.

Mass spectrometer densities carry a completely different systematic: the instrument’s own calibration, the effect of the spacecraft’s surface on the gas entering the aperture, and the correction for the ram enhancement at orbital speed. They are absolute in a way the others are not.

The three differ at the ten to twenty per cent level, and the pattern of the difference with altitude is consistent with an altitude-dependent drag coefficient rather than with a constant offset. Adopting a physically modelled coefficient in place of 2.2 reduces the disagreement without removing it.

That residual is the state of the art, and it is the number quoted as the accuracy of an empirical thermospheric model — not because the models fit their own data badly, but because the data disagree with each other by about that much.

The shape of the problem #

The structure here recurs and is worth naming, because it is the reason this kind of error is hard to find.

A measurement of $X$ is made by observing $Y$ and dividing by $K$. If $K$ is a property of the individual object, errors in it scatter and the population average is fine. If $K$ is a property of the process — shared by every object because the physics is the same everywhere — then errors in it are a systematic that survives any amount of averaging.

The drag coefficient is the second kind, and so is the gas-to-dust conversion in a molecular cloud mass, and so is the drag coefficient’s counterpart in every field where a bulk property is inferred through a microscopic one.

The test for which kind a quantity is: would two independent groups measuring two different objects get the same answer, and would they be wrong in the same direction? If the second answer is yes, no amount of agreement between them is evidence.

The practical response is the one the field has taken, and it has two halves. Where the shared factor can be modelled from physics rather than fitted, model it — which is what an adsorption-based accommodation model does, and it converts a free parameter into a prediction that can be wrong in a detectable way. And where it cannot, find an observation whose dependence on it is different: a mass spectrometer does not divide by a drag coefficient, so its disagreement with a drag measurement is a measurement of the coefficient rather than noise.

Still open: what the next question is #

The density and the coefficient together set what a drag calculation can say about the past. What they do not address is the future, and the future is what operations need.

A model fitted to decades of drag is a climatology, and its error does not shrink with more data because the limitation is the functional form and the proxies driving it. The alternative is to update it continuously from the drag observed on objects in orbit right now — which is the manoeuvre a weather forecast makes, applied to an atmosphere whose observations are orbits. How much that buys, and for how long, is a question about how fast the thermosphere forgets.