Mass decides everything, by a power of three and a half

Give a star a mass and a composition, and everything else follows. Its luminosity, its surface temperature, its radius, its structure, how long it lasts and how it ends are all determined — not statistically, but as the solution to a set of equations with no remaining freedom.

That statement is called the Vogt–Russell theorem, and it is more of a working principle than a theorem. It is nevertheless very nearly true, and it is the reason stellar astronomy is tractable at all: one parameter to explain a population.

The claim is worth stating carefully, because it is stronger than it sounds and weaker than it is usually taken to be. What it says is that the equations of stellar structure — four differential equations in the mass coordinate, plus the constitutive relations for pressure, opacity and energy generation — have a unique solution once the total mass and the composition are given. There is no residual freedom to be fixed by how the star was born, how fast it collapsed, or what it was made of before the last mixing. A star forgets its history. Two stars of one solar mass and solar composition are the same star to the precision anything can be measured, whether one formed in a quiet filament and the other in the wreckage of a nearby supernova.

The steepness, and where it comes from #

Across the middle of the range,

$$L \propto M^{3.5},$$

which is a very steep relation. Doubling the mass multiplies the luminosity by $2^{3.5} \approx 11$. A star of ten solar masses is three thousand times brighter than the Sun; one of a tenth is about three thousandths.

The exponent comes out of two requirements a star has to satisfy at once.

Hydrostatic equilibrium. Every layer is squeezed by the weight above and held up by the pressure below, and the two must balance exactly or the star moves. Balancing them for a self-gravitating gas ball gives a central temperature scaling roughly as $T_c \propto M/R$.

Energy transport. The luminosity is fixed by how fast energy leaks out through the overlying material, not by how fast fusion could in principle run. For a radiative interior the leakage rate works out at $L \propto M^3$ with the composition and opacity supplying the rest.

The observed 3.5 is those two constraints together with the way opacity varies. The exponent is not universal: it is nearer 4 for stars around a solar mass, closer to 3 for the massive ones, and around 2.3 at the bottom of the range where the interiors are convective rather than radiative. The kinks in the curve are those regime changes, and they are physics rather than fitting noise.

The essential point is what fixes $L$. It is not the fusion rate. Fusion adjusts itself to whatever is needed — if the core produced too much energy it would expand and cool, throttling itself back; too little and it would contract and heat. A star is a thermostat, and the setpoint is imposed from outside by how fast the envelope can get rid of energy.

That inversion is worth dwelling on, because the intuitive story runs the other way. A candle’s brightness is set by how fast the wax burns. A star’s is not: it is set by how badly the overlying material obstructs radiation, and the core then burns at whatever rate maintains it. The Sun’s core produces about 276 watts per cubic metre, which is roughly the output of a compost heap and less than a human body per unit volume. The Sun is bright because it is enormous, not because its fusion is fierce. Nothing about the mass–luminosity relation survives forgetting that.

The scaling can be assembled in three lines. Hydrostatic balance gives a central temperature $T_c \propto M/R$. Radiative diffusion carries a flux proportional to $T^3/(\kappa\rho)$ times the temperature gradient, which for a star of mass $M$ and radius $R$ integrates to $L \propto M^3/\kappa$. If the opacity $\kappa$ were constant the exponent would be exactly 3. It is not: through the middle of the main sequence, bound-free and free-free absorption give Kramers’ opacity $\kappa \propto \rho T^{-3.5}$, and feeding that back steepens the relation past 3 and toward 4. The observed 3.5 is the average of a curve that is never quite any single power.

What follows from it #

Because the relation is so steep, everything else about a star inherits its extremity from the mass.

Temperature. Combining $L \propto M^{3.5}$ with $R \propto M^{0.6}$ and $L = 4\pi R^2\sigma T^4$ gives $T \propto M^{0.5}$ or thereabouts. Surface temperature varies far less than luminosity does — a factor of ten in temperature across the whole main sequence against a factor of $10^{10}$ in output.

Radius. Roughly $M^{0.6}$, so a 20 solar-mass star is only about six times the Sun’s radius. Mass changes brightness enormously and size hardly at all.

Lifetime. Fuel goes as $M$ and consumption as $M^{3.5}$, so lifetime goes as $M^{-2.5}$. This is the most consequential inheritance of all.

Fate. Below about 8 solar masses a star ends as a white dwarf; above it, as a supernova leaving a neutron star or a black hole. One number decides.

The compression is worth stating numerically. Main-sequence masses run from about 0.08 to about 100 solar masses — a factor of a thousand. Luminosities over the same range run from $10^{-4}$ to $10^{6}$ solar — a factor of $10^{10}$. A modest range of one quantity produces a range of another that spans ten orders of magnitude, and the diagram is a picture of that amplification.

How a mass is obtained #

Everything above depends on knowing stellar masses, and there is only one direct way to get one: watch something orbit.

Kepler’s third law with Newton’s correction gives the total mass of a binary from its period and separation. Splitting the total into two individual masses needs the mass ratio, which comes from the two stars’ orbits about their common centre — the barycentre argument, where the orbit sizes are in inverse proportion to the masses.

Both quantities are only available together in particular configurations, and the best of them is the eclipsing binary. When the orbit is seen edge-on, the stars pass in front of each other and the light curve gives the inclination exactly, removing the $\sin i$ ambiguity that spoils every other spectroscopic measurement. The eclipse timings also give the radii, in units of the orbital separation. So an eclipsing double-lined spectroscopic binary yields two masses and two radii, to a few percent, with no model of stellar structure assumed.

There are a few hundred such systems measured well. That is the entire empirical basis of the mass–luminosity relation, and every stellar mass quoted for a single star is an inference from those few hundred calibrators.

What was actually measured #

The number is worth pinning down, because a relation drawn as a smooth curve across three decades of mass invites the assumption that it was sampled across three decades of mass. It was not.

The working catalogue of detached eclipsing binaries with masses and radii good to better than 2% contains a few hundred systems. Their masses cluster where the stars are common and bright enough to have been noticed and followed: between roughly 0.2 and 25 solar masses, with the great majority between 0.5 and 5. Below a tenth of a solar mass there are a handful of systems, and above about 40 there are fewer than a dozen anywhere, all in crowded fields at kiloparsec distances where the light of one star is hard to separate from its neighbours’.

What each system actually yields is worth being precise about, because none of it is a mass directly. The spectra give two velocity curves, from which the amplitude ratio $K_1/K_2$ is the inverse mass ratio, and from which $M\sin^3 i$ falls out for each star. The eclipse light curve gives the inclination $i$ from the shape and duration of the flat-bottomed portion, and the fractional radii from the eclipse durations. Only when both are in hand does the $\sin^3 i$ divide out and leave two masses in kilograms — or, more honestly, in solar masses, since $GM_\odot$ is known to nine figures while $G$ alone is known to five and $M_\odot$ therefore no better.

That last point is a small scandal worth stating plainly. The product $GM$ for the Sun is one of the best-measured quantities in physics, because it is what planetary orbits actually depend on and radar ranging measures orbits superbly. The gravitational constant on its own is the worst-measured fundamental constant there is, at about one part in $10^4$. Every stellar mass in this essay is therefore quoted in solar units not out of convenience but out of necessity: converting to kilograms would throw away four digits of precision that were never in doubt.

The relation is fitted through those calibrators and then applied, by extrapolation, to everything else. A quoted mass for a single field star is not a measurement. It is the luminosity, run backwards through a curve calibrated on a few hundred pairs of stars, with the composition assumed.

Both ends of the range #

The relation stops at both ends, and each stopping point is a piece of physics.

Below about 0.08 solar masses, the core never reaches the ten million kelvin needed to fuse hydrogen. The object contracts until electron degeneracy halts it and then cools forever: a brown dwarf. The boundary is sharp because the ignition condition is a threshold, and it was crossed observationally only in 1995 — brown dwarfs are dim, cool and were long suspected before any was confirmed.

Above about 100 solar masses, radiation pressure becomes a problem. The outward push of the star’s own light grows in proportion to $L$, and since $L$ grows as $M^{3.5}$ while gravity grows only as $M$, there is a mass above which the light blows the star apart. That is the Eddington limit, and stars near it shed mass in violent winds — Eta Carinae lost perhaps 20 solar masses in a single eruption in the 1840s.

Between those bounds the relation holds, and outside them there are no stars.

The consequence that outlives the star #

Of everything mass fixes, the one with the longest reach is how long the star lasts. The steepness of that curve is a direct inheritance from the mass–luminosity relation, and it is the reason the galaxy’s chemistry is dominated by a tiny and short-lived minority of stars. Reading those two figures together gives the compression that makes stellar astronomy work. One parameter — mass — spans three decades. Everything else it fixes spans between one and ten. That amplification is why the main sequence is a narrow band rather than a cloud, and why a single number is enough to identify a star.

A steep exponent is a machine for making rarity matter #

The generalisation is not about stars. It is about what a steep power law does to a population, and it is the reason a relation this simple has consequences this far-reaching.

Take any quantity distributed over a range, and pass it through $y = x^{3.5}$. The output is dominated by the largest inputs, however rare they are. Stars are born with a mass distribution that falls off steeply — roughly $dN/dM \propto M^{-2.35}$, so for every star above ten solar masses there are about three hundred below one. Multiply that scarcity by the luminosity relation and the arithmetic reverses: those three hundred faint stars together produce less light than the one massive star does. A cluster’s total output, its colour, its ultraviolet flux and the appearance of its spectrum are set almost entirely by stars that make up well under one percent of its membership. The surprising direction is what this does to time. Because lifetime falls as $M^{-2.5}$ while luminosity rises as $M^{3.5}$, the total energy a star emits over its whole life goes as $M^{3.5} \times M^{-2.5} = M$ — simply proportional to the mass. Every kilogram of hydrogen ever assembled into a main-sequence star releases about the same total energy regardless of which star it ended up in. The mass–luminosity relation, which appears to say that big stars are extravagant, says nothing of the kind about the lifetime budget. It says only that they spend it faster.

That is the connection worth carrying away, because it explains a fact that looks unrelated: the galaxy’s chemical enrichment is set by which stars formed, not by how much gas there was. Massive stars return processed material and low-mass stars lock it up for longer than the universe has existed, and the split is decided by an exponent.

The relation before it was a relation #

Eddington derived $L \propto M^{3.5}$ in 1924 from radiative transfer, before anyone knew what powered a star. That is the part usually left out. He had no fusion, no nuclear physics, no idea where the energy came from — and it did not matter, because the derivation never uses the energy source. It uses only the requirement that the energy get out.

The prediction landed in a field that had a small amount of data and a strong prior against it. Masses were known for a few dozen binaries, and the trend was there. What briefly troubled Eddington was that the giants appeared to obey the relation too, which his derivation said they should not, and which turns out to be a near-coincidence: a giant’s luminosity is fixed by its core mass, and core mass happens to correlate loosely enough with total mass that a sparse sample cannot tell the two stories apart.

The episode is a good example of a theory being right for a reason more general than the one that made it seem plausible. Radiative diffusion does not care what is heating the gas. Any star, powered by anything at all, that is in hydrostatic equilibrium and transporting energy radiatively must obey a relation of this shape.

Where the model stops #

Main sequence only. Giants and white dwarfs obey nothing like this relation; a red giant’s luminosity is set by its core mass rather than its total.

Composition matters. The Vogt–Russell statement includes composition, and metal-poor stars of the same mass are hotter and brighter. That shift is why the main sequences of globular clusters are offset from the local one.

Single stars. Close binaries transfer mass and become stars that no single-star relation describes. The Algol paradox — the less massive star being the more evolved — was unintelligible until mass transfer was recognised.

Rotation and magnetism. Rapid rotation flattens a star and changes its transport, and both effects show up as a shift in the measured colour; strong fields suppress convection. Both are second-order and both are measurable.

There is also a limitation in how the relation is usually presented. It is drawn as a line, which suggests a star’s mass can be read off its luminosity. Running it in that direction is far less safe than running it forwards: the scatter is a few tenths of a magnitude, the relation depends on composition and age, and inferring a mass from a brightness is an estimate rather than a measurement. Every mass in astronomy that is quoted with real precision came from an orbit.

The ladder from here #

Later rungs: hydrostatic equilibrium and the equations of stellar structure. Opacity, and why it decides the exponent. Radiative and convective transport, and the boundary between them. Eclipsing binaries in detail. The mass–radius relation. The initial mass function, which says how many stars are born at each mass. The hydrogen-burning limit and brown dwarfs. The Eddington limit and mass loss. Mass transfer and the Algol paradox. And the upper mass limit, which is still argued over.

Eddington derived the mass–luminosity relation in 1924 from theory, and then found it in the data. He had expected it to apply only to main-sequence stars and was briefly troubled that giants seemed to obey it too — a coincidence that dissolved when their structure was understood.