A convective boundary with no theory to fix it

The Schwarzschild criterion says where convection happens: wherever the radiative temperature gradient a star would need exceeds the adiabatic one. It is an unambiguous local statement, it requires no free parameters, and it decides that the Sun’s outer third boils and its core does not.

What it does not say is where the convection stops. At the boundary the buoyant acceleration is zero, so a blob arriving there has stopped accelerating — and it is still moving. It penetrates into the stable region beyond, decelerating, and while it does so it mixes.

The parameter is called overshoot, or penetration, or convective boundary mixing depending on who is writing, and the multiplicity of names is a fair indication that nobody is confident about what it describes. Its sibling at the other boundary of the star has the same character and the same history.

Why the criterion is not the boundary #

The confusion is worth being precise about, because the criterion is often described as locating the convective region and it locates something narrower.

The Schwarzschild criterion is a statement about stability: a displaced parcel either accelerates away or returns. It says where convection can start, which is where the acceleration is outward.

Motion, however, is not determined by acceleration alone. A parcel that has been accelerating through the convective interior arrives at the boundary with a velocity, and it needs a distance to stop. That distance is not given by any local criterion, because it depends on the kinetic energy the parcel accumulated on the way — a non-local quantity that a local theory cannot express.

So the convective region and the mixed region are different regions, and the difference is a length that mixing-length theory has no way to compute. It is parameterised as a fraction of the pressure scale height, usually between zero and about a third, and the value is chosen to fit observations.

There is a second, more physical way to state the same thing, and it makes the size of the effect plausible. Convective velocities in a stellar core are of order tens of metres a second, which is slow — but the deceleration available at the boundary is also small, because the buoyancy just outside is only slightly restoring. A parcel with a modest velocity entering a region with a weak restoring force travels a long way, and “a long way” in units of the pressure scale height is what the parameter measures. The reason the answer is of order a tenth rather than of order $10^{-6}$ is that the stable region just outside a convective core is only marginally stable.

What extra mixing does #

The effect on a star’s evolution is entirely through the fuel supply, and it is large because the core is small.

A star of a few solar masses burns hydrogen in a convective core containing perhaps a tenth of its mass. Everything in that core is mixed and available; everything outside it is not. Extending the mixed region even slightly brings in fresh hydrogen from a region where the density is still high, and the fuel added is a substantial fraction of what was there.

More fuel means a longer main-sequence lifetime and a more massive, more luminous core at the end of it. Both consequences are observable, and they show up in three ways:

A brighter, cooler turn-off. The point at which a cluster’s stars leave the main sequence sits at higher luminosity for a given age, so an age read from a turn-off is younger with overshoot than without.

A longer main-sequence hook. The final contraction as the core exhausts its hydrogen produces a characteristic feature in the track, and its extent depends directly on the core mass.

A different mass at every subsequent stage. The helium core mass entering the giant branch is set by the hydrogen core mass leaving the main sequence, so the parameter propagates into the whole later evolution — into the star that swells because its centre shrank and everything after it.

And there is a fourth consequence that is easy to overlook and is where several published discrepancies live: the shape of the composition profile left behind. Instantaneous full mixing over an extended region leaves a sharp step in composition at its edge; a gradually decaying mixing efficiency leaves a smooth gradient. The two give the same fuel supply and completely different profiles, and a composition gradient is what supports or suppresses the waves that seismology measures. So the two prescriptions are indistinguishable in every classical observable and distinguishable in exactly one modern one.

One more thing depends on it and it is the most consequential single item: the mass boundary between stars that end as white dwarfs and stars that end as supernovae. That boundary is set by whether the carbon–oxygen core left after helium burning exceeds the Chandrasekhar mass, and the core mass is set by the mixed region at every prior stage. Shifting the overshoot parameter across its plausible range moves the boundary by a solar mass or more, which changes the number of supernovae a population produces by tens of per cent — and therefore the metals it makes, the energy it injects, and the rate at which its gas is used up. A parameter about a few hundred kilometres of a stellar core propagates all the way into galaxy formation.

What can measure it #

Three observations constrain the parameter, and they are all comparisons between a model and something measured without one.

Eclipsing binaries. A detached binary with two stars of known mass and radius, both on the main sequence, has to be fitted by a single isochrone: the two stars are the same age. Requiring that constrains the overshoot, because a larger overshoot changes the radius at a given mass and age. The best systems give the parameter to about 0.05 in units of the scale height, and the results indicate that it increases with mass below about two solar masses and is roughly constant above.

The width of a cluster’s main sequence. More overshoot extends the main sequence to higher luminosities, so the observed extent of a cluster’s upper main sequence constrains it. The measurement is model-dependent in a way the binary one is not, but it applies to many more stars.

Asteroseismology of intermediate-mass stars. This is the newest and most direct. A star with a convective core has a sharp composition gradient at its edge, and that gradient reflects waves — producing a periodic signature in the frequencies whose period depends on the location of the discontinuity. Measuring it locates the boundary of the mixed region directly, without an evolutionary model.

It is worth saying why the eclipsing-binary constraint is so much stronger than it looks. A detached binary supplies two masses and two radii, measured geometrically, with no model anywhere. A single star of known mass and radius constrains a one-parameter family of models — any age can be matched by adjusting the overshoot. Two stars of different masses at the same age cannot: the two constraints have to be satisfied simultaneously by one age and one parameter, and for a well-chosen pair the solution is unique. The information comes from the difference in mass rather than from the precision of either measurement, which is why the best systems are those with the most unequal components, and why a binary with two nearly identical stars — which is easier to find and easier to measure — is nearly useless for this.

The parameter is not one number #

The evidence now says fairly clearly that a single value does not work, and the pattern of the failure is informative.

The binaries indicate a mass dependence: below about 1.2 solar masses the convective core is small or absent and there is nothing to overshoot from; between 1.2 and about two the required parameter rises steeply; above two it plateaus. That is what a physical picture would predict — a small core has a small velocity at its boundary and less momentum to carry — and it is not what a constant parameter provides.

There is also evidence that the extra mixing is not a simple extension of the fully mixed region. A parameterisation in which the mixing efficiency decays exponentially beyond the boundary, rather than stopping abruptly, fits the seismic signatures better. The two prescriptions give the same fuel supply for different values of their parameters, so they are indistinguishable from lifetimes alone and distinguishable from frequencies.

Why the ages are the thing at stake #

The reason this parameter attracts attention out of proportion to its physical interest is that stellar ages are the currency of a great deal of astrophysics, and they come almost entirely from turn-off luminosities.

The age of a globular cluster constrains the age of the universe, and an age read off a bend is how it is obtained. The ages of stars in the Galactic disc constrain when it formed. The age of a planet host constrains how long its system has been evolving. All of them are read from a turn-off, all of them depend on the core’s fuel supply, and the fuel supply depends on a number nobody derives.

For the oldest clusters the effect is small, because their turn-off stars are below a solar mass and have radiative cores with nothing to overshoot from. For intermediate ages it is at its worst: an age of a billion years, read from a turn-off near two solar masses, is uncertain by a quarter from this parameter alone.

There is a fourth constraint worth mentioning because it is independent of everything above: the surface abundances of stars that have finished the main sequence. Extra mixing at the core boundary brings processed material closer to the region that will later be dredged up by the convective envelope, so the surface abundances of carbon, nitrogen and oxygen after the first dredge-up depend on how far the mixing reached. Measuring those abundances is an ordinary spectroscopic analysis with all its own systematics, and the constraint is correspondingly weaker than the seismic one — but it is sensitive to a different aspect of the same physics, which is what makes it worth having.

Where the picture stops #

Three limits stand out, and the second is the one that will resolve it.

Overshoot is not the only extra mixing. Rotation drives circulation and shear instabilities that also transport material into the core, with a different dependence on mass and rotation rate. The two are degenerate in their effect on lifetimes, so a parameter fitted to lifetimes measures their sum, and separating them requires observables sensitive to the shape of the composition profile rather than to its extent.

The seismic measurement is the way out and it applies to few stars. The frequency signature of the composition gradient is small and requires high signal-to-noise, so it is available for tens of stars rather than thousands — and those stars are intermediate-mass, which is the regime that matters most. Extending it is a matter of observing time rather than of new physics.

And three-dimensional simulations do not yet settle it. Simulations of core convection with penetration exist, they show the exponentially decaying mixing that the seismology prefers, and they are run at Reynolds numbers many orders of magnitude below a real star’s. Whether the penetration depth they find scales to stellar conditions is exactly the question, and it is the same question the mixing-length calculations face at the other boundary.

Why a boundary is harder than an interior #

The recurring shape here is one worth extracting, because it is the same in several places in this collection.

Interiors are easier than boundaries. The interior of a convection zone is nearly adiabatic and needs no parameter; the interior of a radiative zone is set by the opacity and needs none either. Everything uncertain in stellar structure lives at the interfaces — the top of the convective envelope, the edge of the convective core, the surface — and the reason is the same in each case: an interface is where a local criterion stops applying and a non-local quantity takes over.

That is also true away from stars. The edge of a planetary ring is a balance rather than a boundary; the boundary of a magnetosphere is where two pressures meet; the edge of a convective core is where momentum runs out. In each case the bulk is computable and the interface is parameterised.

The practical consequence for a reader of stellar results is short. A quantity derived from a star’s bulk properties — a mean density, a luminosity, an effective temperature — is on firm ground. A quantity that depends on how far a mixed region extends, which is every age and every core mass, carries a fitted parameter whose value is known to about a factor of two.

A final observation on what makes this parameter different from the mixing length, since the two are often mentioned together. The mixing length affects a star’s radius, which is measurable directly for many stars by interferometry and by eclipsing binaries, so the parameter has a large and growing body of external calibration. Overshoot affects a star’s lifetime, and lifetimes are not measurable at all: there is no star whose age is known independently to the precision required. The best that can be done is internal consistency — two stars in a binary having the same age, a cluster’s stars having one age — and consistency constrains the parameter only through the differences between stars rather than through an absolute value. That difference in what is measurable, rather than any difference in the physics, is why one parameter is close to settled and the other is not. Two scaling relations calibrated on one star faces the same asymmetry between the quantity it delivers well and the one that is wanted.

It is worth naming the one place the parameter can be avoided altogether: for stars below about 1.1 solar masses there is no convective core at all, so there is nothing to overshoot from and no parameter to choose. That is the mass range of the oldest clusters’ turn-offs, and it is why globular-cluster ages are less exposed to this particular uncertainty than intermediate-age ages are. It is also a reminder that the difficulty is not distributed evenly across stellar astrophysics: a technique’s vulnerability to a fudge factor depends on which stars it happens to use, and two age determinations of comparable formal precision can differ enormously in how much of a model they depend on.

Be explicit about how the parameter is usually chosen in practice, because the procedure is not what a reader would assume. A grid of stellar models is not fitted star by star with the mixing extent free; it is computed once, at a single value calibrated on a set of well-observed clusters, and then applied to everything. So a published age for an individual star inherits a parameter fitted to a population that star may not resemble — a different mass, a different metallicity, a different rotation rate — and carries no uncertainty from it at all, because the grid has no such uncertainty in it. The quoted age error is the error of the fit to the grid, not the error of the grid, and the two differ by a factor of several. That is a general hazard of any result derived by interpolating in a precomputed model set, and it is worth asking of any age, mass or radius that arrives that way.

Where the ladder goes next #

The immediate next rung is the seismic signature itself: how a sharp composition gradient imprints a periodicity on a frequency spectrum, and what its amplitude says about how sharp the gradient is. The one above that is rotational mixing — the second transport process acting on the same boundary, degenerate with this one in its effect on lifetimes and separable by its effect on surface abundances.