The main sequence is a place stars sit, not a track they travel

A diagonal band on a plot invites one reading above all others: that things move along it. Nine out of ten stars sit on the main sequence, the band runs from hot and bright to cool and faint, and the natural story is that a star begins at the top and slides down as it exhausts itself.

Nothing about that story is right. A star does not move along the main sequence in either direction. It arrives at one point on the band, determined almost entirely by the mass it was born with, stays within a short distance of that point for ninety percent of its existence, and then leaves — not along the band but away from it, to the right.

The band is a mass sequence. Reading it as an evolutionary track is the commonest error in stellar astronomy, and correcting it turns the diagram from a picture of a process into a census of a population.

What fixes a star’s position #

The chain that places a star has no free parameters in it, which is why the band is a line rather than a region.

Mass fixes the luminosity, through the mass–luminosity relation. Mass fixes the radius, through a separate and shallower power law. Luminosity and radius together fix the effective temperature through $L = 4\pi R^2\sigma T^4$. So a star’s coordinates on the diagram — luminosity and temperature — are both functions of the single number it was born with.

Two independent-looking observables that are both functions of one hidden variable cannot help but fall on a curve. That is the whole reason the main sequence exists as a narrow feature, and it is a statement about a correlation rather than about a trajectory.

The band’s residual width is real and is worth accounting for, because it is what the diagram measures once the main relation is taken out. Composition contributes: a metal-poor star of the same mass is hotter and bluer, so globular-cluster main sequences sit to the left of the local one. Age contributes a little, as discussed below. Rotation contributes. And unresolved binaries contribute a distinct feature — a second sequence about 0.75 magnitudes above the first, being pairs of near-identical stars whose combined light is twice one star’s.

The motion that does occur #

Saying stars do not move along the band is a simplification, and the honest version is more interesting.

A star does move on the diagram while burning core hydrogen, and it moves in a direction almost perpendicular to the band. As hydrogen becomes helium the mean molecular weight of the core rises, so at a given pressure fewer particles are available and the core must be hotter and denser to hold itself up. A hotter core means more fusion, so the luminosity climbs.

The Sun is about 30% more luminous now than when it settled onto the main sequence 4.6 billion years ago, and its surface temperature has changed rather little. That is a motion almost straight upward on the diagram, of about 0.3 in the logarithm — small on a plot spanning ten decades, and not zero.

Two useful definitions follow. The zero-age main sequence is the locus of stars that have just begun burning hydrogen, and it is the lower edge of the band. The main sequence as observed is a strip a few tenths of a magnitude thick above it, populated by stars at every stage of their hydrogen-burning lives.

So the correct statement is: a star arrives at the ZAMS at its mass’s position, drifts a short way upward and slightly to the right over its whole main-sequence life, and then leaves. The drift is perpendicular to the band, small, and it is not what produces the band.

Leaving, which happens sideways #

What terminates the main-sequence phase is the exhaustion of hydrogen in the core, not in the star. About 10% of a solar-mass star’s hydrogen is in the region hot enough to burn, and when that is gone the star has 90% of its fuel left and no way to reach it.

The core, without a source of energy, contracts. Contracting releases gravitational energy and heats a shell of hydrogen around it, which ignites. Shell burning is more luminous than core burning was, because the shell sits hotter and denser than the core ever did, and the envelope responds to the increased flux by expanding enormously — which cools it.

The star therefore moves right on the diagram, toward lower temperature, while moving up, toward higher luminosity. It leaves the band at an angle close to perpendicular to it and crosses to the giant branch. Nothing about the transit resembles sliding down the main sequence, and the crossing is quick — a few percent of the star’s life — which is why the region between the band and the giants is sparsely populated.

The turn-off, which turns the diagram into a clock #

The census reading has a consequence that makes the diagram the primary dating tool for anything older than a radioactive clock can reach.

Plot every star in a cluster. They all formed at about the same time from the same material, so they differ only in mass. The most massive have already exhausted their cores and left; the rest are still on the band. The point where the sequence bends away — the turn-off — is at the mass whose main-sequence lifetime equals the cluster’s age.

The method’s sensitivity comes from the steepness of that curve. A turn-off at 0.85 solar masses gives about 13 billion years; at 1.5, about 2 billion; at 4, about 200 million. A modest shift in the turn-off is an enormous shift in age.

The object actually fitted is an isochrone — the locus, on the diagram, of a population of a single age and composition across all masses. It is not the main sequence and it is not an evolutionary track; it is a slice across a whole family of tracks at one instant. The distinction matters because all three are curves on the same plot and confusing them is easy: a track follows one star through time, an isochrone follows many stars at one time, and the main sequence is where the isochrone happens to lie for the masses still burning hydrogen.

The band as a census of what was born #

If the main sequence is a population rather than a process, the obvious next question is what sets the population — and the answer is the other half of the diagram’s information.

How many stars occupy each point of the band is fixed by two things multiplied together: how many were born at each mass, and how long each of them stays. The first is the initial mass function, which falls steeply — roughly $dN/dM \propto M^{-2.35}$ above a solar mass, so for every star born above 10 solar masses about three hundred are born below one. The second is the lifetime relation, which falls as $M^{-2.5}$.

Multiplying them gives the observed density along the band, and it is startlingly lopsided. The number of stars currently on the main sequence at each mass goes roughly as $M^{-4.85}$ — the birth rate’s steepness compounded by the survival time’s. A hundredfold range in mass produces something like a $10^{10}$ range in how many are sitting there.

That is why the top of the main sequence is nearly empty in any diagram drawn from a real sample, and why the bottom is so crowded that the lowest-mass stars are usually omitted for legibility. The most common star in the galaxy is an M dwarf at the bottom right of the band, and not one of them is visible to the naked eye.

Running the argument backwards is how the initial mass function is measured. Count stars along the band in a young cluster — young enough that nothing has left yet — and the counts are the birth distribution, with no correction for lifetimes required. That measurement, made in a handful of nearby clusters, underpins essentially every estimate of a galaxy’s stellar mass, since what is observed is light and what is wanted is mass, and the conversion between them is an integral over the mass function.

The two ends, where the band stops for different reasons #

A mass sequence has to end where the masses do, and the two ends terminate for reasons that have nothing to do with each other.

The bottom end is a threshold in the physics. Below about 0.08 solar masses a contracting object never reaches the central temperature that sustains hydrogen fusion, because electron degeneracy halts the contraction first — the same support that holds up a white dwarf, arriving early enough to prevent the star from ever igniting. Such an object cools for ever and is a brown dwarf. The main sequence therefore has a genuine lower edge, and it is set by the crossing of two curves: the temperature contraction can deliver, and the temperature fusion requires.

There is a second feature not far above it, and it is a change of internal structure rather than a threshold. Below about 0.35 solar masses a star is convective throughout, with no radiative core, so its entire hydrogen supply is continuously mixed into the burning region. Such a star has no core to exhaust separately from itself, and the account of leaving the main sequence given above simply does not apply to it: it burns nearly all its hydrogen, takes something over a trillion years to do so, and none has ever finished. The bottom of the band is populated entirely by stars that have not yet done anything, which is a different kind of stasis from the one this essay is about.

The top end is not a threshold at all. It is a limit on stability, and where exactly it sits is not agreed. Above roughly 100 solar masses radiation pressure dominates the support, the star sits close to the Eddington luminosity, and the mass loss becomes violent enough that the object sheds itself faster than the diagram can meaningfully place it. The most massive stars known are around 200 solar masses at formation and considerably less now, and the uncertainty in that figure is a factor of two.

So one end of the band is sharp and calculable and the other is fuzzy and argued about, and the diagram shows neither: the top is empty because massive stars are rare and short-lived, and the bottom is usually cropped because M dwarfs are too faint to appear in a sample selected by brightness. The band’s drawn extent is a property of the sample, and its physical extent is a property of the stars, and no figure distinguishes them.

What was actually measured #

The reading above is a theoretical claim about motion, and the data that support it are almost entirely statistical, because no star has been watched moving.

The timescales rule out direct observation completely. A solar-mass star spends $10^{10}$ years on the main sequence and a few times $10^{8}$ years crossing to the giant branch. The whole of telescopic astronomy is four centuries. Nothing has been seen to move on this diagram, and nothing will be.

What is measured instead is the distribution of stars over the plane, and the argument runs through occupancy. A region of the diagram is densely populated if stars spend a long time there and sparsely populated if they pass through quickly. The main sequence holds 90% of stars because 90% of a star’s life is spent on it; the gap between the band and the giant branch — the Hertzsprung gap — is nearly empty because the crossing takes a few percent of the time. Sparse regions are fast regions, and the density of points is a direct readout of how long each phase lasts.

That inference is checked where a check is available. Cluster ages from turn-offs agree with ages from white-dwarf cooling and from radioactive dating to within about a billion years, and those three methods share almost no assumptions. Asteroseismology gives ages for individual field stars, from their oscillation frequencies, without using the diagram at all — and where both are available they agree.

There is one place where motion on the diagram has genuinely been observed, and it is worth having as the exception. Stars in the instability strip pulsate, and the period of a Cepheid changes measurably over decades as the star’s radius evolves. Polaris’s period has lengthened by about eight seconds per century, which corresponds to a real and observed evolution across the strip. It is a motion of a few parts in $10^{5}$ per year, on a diagram spanning ten decades, and it is the only such measurement there is.

The generalisation: a locus is not a path #

The error the essay corrects is not specific to astronomy, and naming the general form makes it easier to spot.

A plot of two quantities across a population produces a locus. A plot of two quantities for one object over time produces a path. They look identical on paper and mean entirely different things, and the confusion between them recurs wherever a scatter plot has structure.

The diagnostic is whether the axis variable can be shown to change for an individual. On the main sequence it cannot: mass does not change appreciably during hydrogen burning, so no star can slide along a band whose coordinate is mass. The same test resolves the confusion elsewhere — in a plot of galaxy luminosity against rotation speed, in a plot of species’ body mass against metabolic rate, and in every economic scatter of one quantity against another across a set of countries.

Where the distinction is most valuable is that it identifies what is moving. On the HR diagram, what moves is the boundary: the turn-off migrates down the band as a cluster ages, not because any star travels down it but because stars keep leaving from the top. A feature can move across a population while no member of the population moves at all, and that is the diagram’s actual dynamics.

Where the model stops #

Single stars. A close binary transfers mass, so its components’ masses change and they really do move along the band — in the wrong direction, since the receiver gains mass and climbs. The Algol paradox, in which the less massive star of a pair is the more evolved, was unintelligible until mass transfer was recognised.

Constant mass. Massive stars lose a substantial fraction of themselves in winds, and above about 40 solar masses this changes the evolution qualitatively.

Solar composition. The band’s position depends on metallicity, so an isochrone fitted with the wrong composition returns the wrong age. Composition and age are partially degenerate in a cluster diagram, and separating them requires spectroscopy.

Hydrogen burning. Everything above the turn-off and everything off the band belongs to different physics, and the pre-main-sequence phase — the Hayashi track, where a contracting protostar approaches the band from the right — is a genuine path rather than a locus.

The figures share the limitation that makes the misreading tempting in the first place. They are static. A diagram that shows where stars are cannot show where they were, and the human tendency to read a band as a trajectory is not corrected by drawing the band more carefully. The only figure that would settle it is one showing a single star’s position over ten billion years, and it would consist of a dot that barely moves and then jumps.

A final note on terminology, since it encodes the misreading the essay corrects. Hot stars are still routinely called “early type” and cool ones “late type”, which are the names Russell used when he believed stars began hot and cooled down the band. The physics was abandoned a century ago and the vocabulary was not, so a phrase asserting a direction of travel survives in a subject that has established there is none.

The band and the isochrones through it are the two halves of the argument, and both are worth reading once more.

The ladder from here #

Later rungs on this anchor: the zero-age main sequence and its thickness. Evolutionary tracks, computed. Isochrones, and how one is fitted to a cluster. The turn-off as an age indicator, done properly. The Hertzsprung gap and the occupancy argument. The subgiant and red giant branches. The helium flash. The horizontal branch and the red clump. The instability strip, and the pulsators in it. The Hayashi track and pre-main-sequence contraction. Mass transfer and the Algol paradox. And the main sequences of other galaxies, where the same structure appears with different chemistry and the isochrones have to be recomputed.

Hertzsprung’s and Russell’s diagram is 110 years old and its central feature was misread as an evolutionary track for the first twenty of them — by both of them. Russell’s early interpretation had stars forming as red giants, contracting up the band and then sliding down it, which is why the two ends of the main sequence are still occasionally called “early” and “late” types.