A star that tells its distance by how slowly it blinks

Most measurements in astronomy trade one difficulty for another. This one trades a difficulty for something that is not difficult at all.

A Cepheid variable brightens and fades on a period of days to months, and its period can be measured by anyone with a telescope, a detector and patience. Nothing about that measurement depends on the distance to the star, on the dust in the way, on the star’s motion, or on any calibration whatever. Counting the days between two maxima is as close to a raw observation as astronomy offers.

And the period is correlated with the star’s intrinsic luminosity. A Cepheid with a thirty-day period is about eight times more luminous than one with a three-day period, and the relation is tight enough to give distances good to a few per cent.

A clock has become a ruler.

Why a pulsation period says anything about a luminosity #

The correlation is not a coincidence, and the chain that produces it is short.

A pulsating star is a standing sound wave, and its period is set by the time a pressure disturbance takes to cross it. To an order of magnitude that is the dynamical timescale,

$$P \sim \frac{1}{\sqrt{G\bar{\rho}}},$$

so the period depends on the mean density and on nothing else. Big diffuse stars pulsate slowly; small dense ones pulsate fast. The relation $P\sqrt{\bar{\rho}} = Q$, with $Q$ nearly constant, is the pulsation constant, and it is the first link.

The second link is that Cepheids all sit in a narrow, nearly vertical band on the HR diagram called the instability strip. Within that band the surface temperature is confined to a narrow range, so the Stefan–Boltzmann relation $L = 4\pi R^2\sigma T^4$ makes luminosity a function of radius alone — the same relation that makes colour a thermometer.

Put the two together. Low density means large radius; large radius at fixed temperature means high luminosity; and low density means long period. Therefore long period means high luminosity, and the relation is a power law because both links are.

The tightness of the relation is entirely the narrowness of the instability strip. A strip of finite width introduces scatter, and it is why the period–luminosity relation in the visual band has about 0.25 magnitudes of intrinsic scatter, while in the near-infrared — where the temperature dependence is much weaker — it has 0.08.

The engine: a valve made of ionising helium #

A star does not pulsate merely because it can. Something has to do work on the oscillation each cycle, and in the instability strip that something is a layer of partially ionised helium.

The mechanism is the kappa mechanism, and it is a heat engine with a valve. In most of a star, compressing a layer raises its temperature and lowers its opacity, so the layer lets radiation through more easily and the compression is damped, as the balance requires. In a layer where helium is halfway through its second ionisation, compression instead goes into ionising more helium rather than into raising the temperature — so the opacity rises with compression. The layer becomes more opaque as it is squeezed, dams the radiation flowing through it, absorbs energy, and pushes back out. That is a valve that opens when it should close, and it drives the oscillation.

The condition for it to work is that the partial-ionisation zone sits at the right depth: too shallow and there is not enough mass to matter, too deep and convection carries the energy instead. The depth of the zone is set by the surface temperature, and the range of temperatures for which it is in the right place is the instability strip. So the strip’s existence, its narrowness and its position are all consequences of where helium ionises.

That is a satisfying closure. The relation that measures the universe rests on the ionisation potential of helium.

Reading the light curve #

The pulsation is not sinusoidal, and its shape is as diagnostic as its period.

A classical Cepheid rises fast and falls slowly — the sawtooth is unmistakable — with the maximum brightness occurring not at minimum radius but at maximum expansion velocity, about a quarter of a cycle earlier. The asymmetry comes from the shock that propagates outward through the envelope on each cycle.

The shape matters because the relation is class-specific and the classes look superficially alike. Three kinds of pulsator share the instability strip, and using the wrong relation gives the wrong answer by a factor:

Classical Cepheids — young, massive, metal-rich stars of the galactic disc, periods 1–100 days, absolute magnitudes −2 to −7. Type II Cepheids — old, low-mass, metal-poor halo and bulge stars in the same period range, and about 1.5 magnitudes fainter at the same period. RR Lyrae stars — old, low-mass, periods under a day, and all at about the same absolute magnitude, $M_V \approx +0.6$, which makes them a standard candle of a different kind: not a period–luminosity relation but a single luminosity.

Telling them apart from a light curve is routine — RR Lyrae are far shorter in period, and Type II Cepheids have a characteristic bump — and telling them apart mattered more than anything else in the history of this subject.

The star is not blinking, it is breathing #

The word “variable” invites the wrong mental picture. A Cepheid does not flicker; it changes size, by a substantial fraction of itself, on a timescale of days.

δ Cephei swings between about 41 and 45 solar radii over its 5.37-day cycle — a 10% change in radius, which at the surface means the photosphere moving at up to 20 km/s. The radial velocity curve measures that motion directly, and integrating it gives the change in radius in kilometres:

$$\Delta R = -p\int v_{\text{rad}}\,dt,$$

with $p \approx 1.3$ the projection factor that accounts for the fact that a disc seen face-on shows a range of angles between the line of sight and the surface motion.

That integral is the basis of the Baade–Wesselink method, which extracts a distance from a Cepheid with no calibration against anything. The velocity curve gives the change in radius in kilometres. The colour and brightness together give the change in angular radius. Dividing one by the other gives the distance — geometrically, for a single star, with no ladder beneath it.

The method has been sharpened by interferometry: the angular diameters of nearby Cepheids are now measured directly rather than inferred from colours, with the CHARA and VLTI arrays resolving discs of a millisecond of arc and watching them pulsate. δ Cephei’s distance from this technique is $274 \pm 11$ pc, and it agrees with its Gaia parallax.

There is something worth noticing in that. A Cepheid is used as a standard candle — an object whose brightness is assumed — and it is also, for the nearest few dozen, an object whose distance can be got geometrically. That overlap is what allows the candle to be calibrated, and it is the same overlap-between-rungs structure that the whole ladder is built from.

What was actually measured #

Henrietta Swan Leavitt found the relation in 1908 and stated it properly in 1912, and the method she used removed the hardest problem by choosing where to look.

She was cataloguing variable stars on photographic plates of the Magellanic Clouds taken at Harvard’s Peruvian station. In the Small Magellanic Cloud she found 1,777 variables and measured periods for 25 of them. Plotting apparent magnitude against the logarithm of the period gave a straight line with a scatter of about a tenth of a magnitude.

The insight is in the choice of target. Every star in the Small Magellanic Cloud is at essentially the same distance — the cloud is far away and not very deep — so apparent magnitude differences are absolute magnitude differences, with the distance appearing as an unknown constant common to all of them. Leavitt therefore obtained the slope of the relation exactly, without knowing the distance to anything.

What she could not obtain was the zero point. That needs one Cepheid whose distance is known some other way, and there was not one: no Cepheid is close enough for a nineteenth-century parallax. Hertzsprung supplied a zero point in 1913 from statistical parallax — using the proper motions of thirteen Milky Way Cepheids against the Sun’s motion — and got a distance to the SMC of about 10,000 parsecs, low by a factor of six.

Then Hubble used the relation on M31 in 1923, found Cepheids, and settled the question of whether spiral nebulae were inside the Milky Way. They were not, and the universe acquired a size.

The zero point was wrong for another thirty years. Walter Baade, observing with the 100-inch during the wartime blackout of Los Angeles, resolved individual stars in M31’s core and found that its stars fell into two populations with different colour–magnitude relations — the disc population and the halo population. In 1952 he announced that the calibration had used one class of Cepheid and the extragalactic measurements another. Every extragalactic distance doubled, and the age problem — a universe apparently younger than the Earth in it — went away.

The modern zero point comes from three independent sources: Gaia parallaxes of Milky Way Cepheids, detached eclipsing binaries in the Large Magellanic Cloud, and the water-maser geometric distance to NGC 4258. They agree to about 1%, and the residual disagreement is one of the terms in the Hubble tension.

The generalisation: a period is the easiest thing to measure #

The deeper reason this method works so well is that a period is the most precisely measurable quantity in observational astronomy, by a wide margin.

A photometric magnitude is good to a per cent; a radial velocity to a part in $10^6$; a position to a part in $10^9$. A period is limited only by how long the observing runs, because the fractional error goes as the timing precision divided by the total baseline. A Cepheid watched for a century has a period known to nine or ten figures. A pulsar’s period is known to sixteen.

So any physical relation that connects a period to something else immediately becomes a precision tool. The subject is full of them:

Asteroseismology turns oscillation frequencies into masses and radii for hundreds of thousands of stars. Pulsar timing turns a spin period into tests of general relativity, a probe of the interstellar medium, and a detector for nanohertz gravitational waves. Eclipsing binaries turn an orbital period into masses, given velocities. Transiting exoplanets turn a period into an orbital radius through Kepler’s third law. Mira variables and the tip-of-the-red-giant-branch method extend the candle idea into the infrared.

The common structure is that a period is a time, times are measurable to absurd precision by counting, and counting is the one operation that improves without limit as the observation lengthens.

What a period costs to measure, and what it buys #

The economics of this method are worth stating plainly, because they are unlike anything else on the ladder.

A useful Cepheid period needs the star observed a few dozen times, spread over at least a few cycles — for a thirty-day Cepheid, four or five months of intermittent observation. That is expensive in telescope time and cheap in everything else: no spectrograph, no calibration source, no absolute photometry, and the measurement improves for free with every additional epoch, because the fractional period error falls as the inverse of the total baseline.

Against that, the luminosity it buys is only as good as the relation and its zero point, which are somebody else’s problem and are the dominant error. So the effort profile is inverted compared with most measurements: the part that is hard to observe is nearly free of systematics, and the part that is trivially read off a plot carries all of them.

That is exactly why the surveys were built the way they were. OGLE has monitored the Magellanic Clouds since 1992 and catalogues more than ten thousand Cepheids; Gaia’s photometric time series covers the whole sky. Neither needed a distance to anything. They accumulated periods, and the periods sat waiting for the calibration to improve — which it did, three times, without a single light curve being re-observed.

The period is not quite constant #

A Cepheid’s period is treated above as a fixed property, and over a century of observation it is not — which turns out to be a measurement of where the star is in its life rather than a nuisance.

A star crosses the instability strip while its interior is rearranging itself, and the crossing takes thousands of years. Its mean density therefore changes slowly, and since the period goes as the inverse square root of that density, the period changes with it. The rate is of order seconds per year for a ten-day Cepheid, which is undetectable in any one season and unmistakable in a hundred.

The measurement is made from an O–C diagram: the observed times of maximum minus the times a constant period would give, plotted against date. A constant period gives a straight line; a period changing at a constant rate gives a parabola, and the parabola’s curvature is the rate.

What it says is which crossing the star is on. A star crossing the strip for the first time, on its way from the main sequence, moves quickly and its period changes fast. On the second and third crossings, during the slower blue loop of core helium burning, the star moves in the opposite direction on one and the same direction on the other — so the sign of the period change says which. Roughly two-thirds of Galactic Cepheids have increasing periods and a third decreasing, which is what the relative durations of the crossings predict.

That is a measurement of stellar evolution made by keeping a notebook. Some of the Cepheids with the best-determined period changes have been observed continuously since the 1890s, and the data include visual estimates by amateurs alongside modern photometry, because what is being measured is a time of maximum rather than a brightness and a time of maximum needs no calibration at all.

Where the model stops #

The relation depends on metallicity. A metal-poor Cepheid is not quite as bright at a given period as a metal-rich one, and the size of the effect — about 0.2 magnitudes per dex — is still argued about. It matters because the calibrators are in the Milky Way and the Magellanic Clouds and the targets are in galaxies with different compositions.

Crowding. A Cepheid in a distant galaxy is measured through a photometric aperture containing other stars, and the contamination makes it look brighter and therefore closer. Correcting for it needs the highest available resolution, which is why the calibration moved to JWST.

The relation is not exactly linear. There is evidence for a break near ten days in the visual bands, and the size of any break changes the extrapolation from the calibrators’ period range to the targets’.

The figures show a static relation and a repeating curve, and neither is what is observed. A real light curve is a scatter of a few dozen points taken on whatever nights were clear, folded on a period that had to be found first; the smooth curve here is the fit. And the period–luminosity plot draws seven stars where the modern relation is fitted to thousands, with the scatter that implies.

The ladder from here #

Later rungs on this anchor: the kappa mechanism worked through, and the Eddington valve. The instability strip’s blue and red edges, and what sets each. The period–luminosity–colour relation, which is the two-parameter version. Type II Cepheids and RR Lyrae as separate candles. The Baade–Wesselink method, which measures a Cepheid’s radius directly from its velocity curve and gives a distance with no calibration at all. Mira variables. The pulsation constant, and asteroseismology’s version of it.

Leavitt was employed as a computer at $0.30 an hour, was not permitted to pursue the work further, and died in 1921. Gösta Mittag-Leffler wrote to nominate her for the Nobel Prize in 1925, and was informed that she had been dead for four years. The relation she found is still, a century later, the rung the size of the universe rests on.