The triangle that reaches the stars, and stops

There is exactly one way to measure the distance to a star that assumes nothing about the star. Look at it from two places, measure the angle between the two sight lines, and solve the triangle. No physics, no model of stellar interiors, no calibration against anything — just geometry.

The method has a ceiling, and the ceiling is low. Every distance in astronomy beyond a few thousand parsecs is obtained some other way, and every one of those other ways is calibrated, eventually, on stars near enough for the triangle to close.

The definition that hides the trigonometry #

The baseline available is the Earth’s orbit, so an observation in January and another in July are made from points 2 astronomical units apart. A star’s apparent position shifts against the more distant background, and half of that total shift is called the parallax, $p$.

The trigonometry is $\tan p = 1\,\text{AU}/d$, and since $p$ is always tiny the tangent can be dropped. That leaves a relation so simple it is usually mistaken for a definition:

$$d\,[\text{pc}] = \frac{1}{p\,[\text{arcsec}]}.$$

The parsec is engineered to make it look that way — it is the distance at which the parallax is one arcsecond, so the conversion factor is 1. That is 3.26 light years, or 206,265 astronomical units, and the number 206,265 is the number of arcseconds in a radian, which is where the whole thing comes from.

No star has a parallax of one arcsecond. The nearest, Proxima Centauri, manages 0.77, at a distance of 1.30 parsecs. Everything else is smaller.

Two centuries of failure #

The size of the angles explains why the measurement defeated everyone who tried it for three hundred years.

Copernicus’s model made a clear prediction: if the Earth moves, the nearer stars must shift against the further ones. No shift was found. Tycho Brahe, the best observer of his age, took the absence as evidence that the Earth did not move — a completely sound inference from his data, since the only alternative was that the stars were unimaginably far away.

They are. Proxima’s 0.77 arcseconds is the width of a coin seen at four kilometres. Tycho’s instruments, the finest before telescopes, reached about one arcminute — eighty times too coarse.

The hunt produced two major discoveries by accident. Bradley, searching for parallax in 1725, found a 20-arcsecond annual wobble in every star, in the wrong phase for parallax — stellar aberration, caused by the Earth’s motion tilting the apparent direction of arriving light. It was the first direct proof that the Earth moves, and it gave the speed of light. He also found nutation, a small nodding of the Earth’s axis.

Actual parallax came in 1838, when Bessel measured 0.314 arcseconds for 61 Cygni. He had chosen the star for its large proper motion, on the correct assumption that fast-moving stars are nearby. The value implied 3.2 parsecs, which is within 10% of the modern figure, and it was the first time anyone had known the distance to anything outside the solar system.

The ceiling #

Parallax fails at a distance set entirely by measurement precision, and the failure is gradual: as the angle shrinks toward the error bar, the distance error blows up.

The reason is the reciprocal. If $p$ is measured to $\pm\sigma$, the fractional error in distance is $\sigma/p$, so a star at ten times the useful limit has an error bar that includes infinity. Parallaxes are therefore quoted as angles rather than distances, and inverting a noisy parallax is a well-known way to produce a badly wrong answer.

The history of the ceiling is a history of instruments. Ground-based work reached a few hundredths of an arcsecond — good to perhaps 100 parsecs, covering a few thousand stars. Hipparcos, in orbit, reached a milliarcsecond and mapped 100,000 stars to about 100 parsecs properly. Gaia reaches tens of microarcseconds and has parallaxes for 1.5 billion stars, useful out to several thousand parsecs.

Even that reaches only a fraction of the way across the galaxy, which is 30,000 parsecs to the centre. Everything beyond needs another method.

Why the ladder is a ladder #

Each rung above parallax works by finding an object whose true brightness can be inferred from something observable, then comparing that with how bright it appears. The inverse-square law of light converts the ratio into a distance.

The catch is the inference. A Cepheid variable’s luminosity can be read off its pulsation period — but only because the period–luminosity relation was calibrated on Cepheids whose distances were known independently, which means by parallax. A type Ia supernova is a standard candle only because nearby ones have been seen in galaxies whose Cepheid distances are known.

So the ladder is genuinely a ladder. An error at the bottom does not stay at the bottom; it multiplies all the way up. When Hipparcos revised the Cepheid calibration, every extragalactic distance moved, and with them the Hubble constant and the age of the universe.

The propagation reaches into stellar physics as well: the ages of the oldest clusters moved when the calibration did, because an age is read off a diagram whose vertical axis is a distance.

This is why Gaia matters beyond its own catalogue. It measures parallaxes directly for hundreds of Cepheids, replacing an inference with a geometric measurement at the point where the ladder is weakest.

The other things the wobble contains #

The same measurements that give parallax give two more quantities for free, and both are more useful than they sound.

Proper motion is the star’s steady drift across the sky, distinct from the annual back-and-forth — a real motion through the galaxy rather than an artefact of the Earth’s own. Combined with the parallax it converts to a transverse velocity in kilometres per second, and combined with a Doppler measurement of the radial component it gives the full three-dimensional velocity. That is how the galaxy’s rotation was mapped and how stellar streams — the debris of shredded satellite galaxies — are picked out of a field of a billion stars.

Aberration, the 20-arcsecond ellipse every star traces annually, has to be subtracted before the parallax appears. It is larger than every parallax ever measured, by a factor of at least 26, which is the reason it was found first. Everything astrometry measures is a wobble at the level of one part in $10^5$ of the sphere, or smaller. That is the reason the field spent two hundred years failing and then, once instruments crossed the threshold, produced the geometric backbone of every distance in the universe.

What was actually measured #

The textbook account — observe in January, observe in July, subtract — describes nothing any modern parallax was obtained from. It is worth replacing, because the real procedure explains where the precision comes from and where the residual systematic comes from too.

Gaia does not measure a star’s position. It measures the angle between two stars seen simultaneously in two fields of view 106.5° apart, projected onto a single focal plane. The spacecraft rotates once every six hours, so those two fields sweep a great circle, and the spin axis itself precesses around the Sun’s direction at 63.5° every 63 days. The result is that every part of the sky is crossed repeatedly at many different orientations over the mission, and each crossing contributes one very accurate one-dimensional angular measurement.

No single measurement contains a parallax. What contains the parallax is the global astrometric solution: a simultaneous least-squares fit, over roughly a trillion individual observations, for five parameters per star — two positions, two proper motions and a parallax — plus the spacecraft’s attitude at every instant and the geometric calibration of the focal plane. The parallax of any one star emerges from its consistency with every other star it was ever measured against.

That is where the accuracy comes from. Individual position measurements are good to a few hundred microarcseconds; the parallax of a bright star is good to twenty. The improvement is the square root of a very large number of observations, and the reason the fixed 106.5° basic angle matters so much is that any variation in it feeds straight through the solution into a global parallax error, identical for every star. Gaia carries an interferometer whose only job is to monitor that angle, at the picometre level.

The residual is the zero point, and it is instructive that it survives. A global solution can be shifted bodily: add a constant to every parallax and the fit is nearly as good, because the data constrain differences well and absolutes poorly. Breaking that degeneracy requires objects known to have zero parallax, and the only ones available are quasars. Tying the solution to about half a million of them leaves an offset of roughly $-17$ microarcseconds that varies with magnitude, colour and position on the sky — small, non-random, and the single largest systematic in the catalogue.

The same trick at every scale #

The triangle generalises in a direction that is easy to miss, because the limiting factor is never the geometry. It is always the ratio of the angular precision to the baseline.

Improve either and the reach extends. The Earth’s orbit cannot be enlarged, so the only remaining lever is angular precision — and radio interferometry has far more of it than any optical instrument, because resolution scales with baseline in wavelengths and a very-long-baseline array spans the Earth. Water masers in star-forming regions are compact, bright at 22 GHz, and can be positioned to about 10 microarcseconds. Their parallaxes are therefore measurable out to more than ten kiloparsecs.

That is not a marginal extension. It reaches across the galaxy. The distance to the galactic centre, long an uncomfortable inference from several unrelated methods, is now a geometric measurement: 8.2 kiloparsecs, from the orbit of a star around the central black hole and from maser parallaxes in the same region, agreeing with each other. The spiral arms of the Milky Way have been mapped by maser parallax rather than by kinematic distances, and several of them moved.

The surprising part is what this does to the ladder. A method that reaches ten kiloparsecs geometrically bypasses spectroscopic parallax and cluster fitting entirely, and calibrates Cepheids directly in the Milky Way. Rungs are not merely being made more accurate; some of them are being removed. The ladder’s defining weakness has always been that errors multiply along it, and the fix is not better rungs but fewer.

The rung above, and what it inherits #

Everything past the parallax ceiling works by the same substitution: replace the geometry with an assumption about the object. That is the distance modulus, and it is an inference rather than a measurement. Its accuracy depends entirely on how well the assumed absolute magnitude is known, and the assumed absolute magnitude is known because it was calibrated on objects near enough for a triangle.

The whole structure has one geometric foundation and everything else rests on it. Which is why a revision at the bottom moves everything: when the Hipparcos parallaxes shifted the Cepheid calibration, the distance to every galaxy shifted with it, and so did the ages of the oldest star clusters.

A star’s parallactic displacement is also not a back-and-forth along a line. It is an ellipse, traced once a year, and the ellipse’s shape is a piece of free information.

The reason is projection. The star’s apparent shift mirrors the Earth’s orbit, seen from the star’s direction. A star at the ecliptic pole sees that orbit face-on and traces a circle of radius $p$. A star on the ecliptic sees it edge-on and traces a straight line of half-length $p$. Everything between traces an ellipse whose semi-major axis is $p$ and whose semi-minor axis is $p\sin\beta$, with $\beta$ the star’s ecliptic latitude.

The semi-major axis is therefore always the parallax, whatever the star’s position, which is why the definition is stated that way. And the axis ratio gives the ecliptic latitude — a coordinate the measurement was not designed to produce and which serves as an internal consistency check, since the latitude is already known from the star’s position.

It also dictates when to observe. The displacement along the minor axis is smallest for stars near the ecliptic, so those stars are measured almost entirely along one direction, and the observing schedule for a parallax programme is set by the geometry of each target rather than by convenience.

And the case that does not occur, drawn because it is what the parsec’s definition is about.

Where the model stops #

Only nearby stars. The hard ceiling, and there is no way around it: the baseline is fixed by the size of the Earth’s orbit and cannot be enlarged. A mission to the outer solar system would gain a factor of thirty, and has never been flown for this purpose.

Point sources. A star with a close companion has its photocentre pulled about by the orbit, contaminating the parallax — although that contamination is itself a planet-detection method.

A stable reference frame. The measurement is differential, against background objects assumed fixed on the celestial sphere. Those objects have their own parallaxes and proper motions, and the modern solution is to tie everything to quasars, which are far enough away to have neither.

Systematics that do not average down. Gaia’s parallaxes carry a small zero-point offset — tens of microarcseconds — that varies with position, magnitude and colour. It is not random, so more observations do not remove it, and characterising it has been a major part of the mission’s work.

The figures have their own honest failure, and it is the one the caption already admits: the angles are exaggerated by three or four orders of magnitude. Drawn truthfully, the two sight lines in the first figure would be separated by less than the width of the line used to draw them, and the triangle would be indistinguishable from a single line. Every parallax diagram ever printed has this problem, and it is worth stating the real number: for the nearest star, the baseline is 300 million kilometres and the angle is 0.77 arcseconds.

There is one further reason the method’s reach matters out of proportion to the number of stars it covers. Every other distance indicator is a relation between a distance and something observable — a period, a line width, a light-curve shape — and every such relation has to be calibrated on objects whose distances are already known. Parallax is the only supplier. So the number of Cepheids, RR Lyrae stars and eclipsing binaries within parallax range is not a detail of one measurement; it is the sample size of the calibration that everything beyond it inherits, and until that sample grew from dozens to thousands the ladder’s bottom rung was the noisiest part of the whole structure.

The same point can be made about the ceiling. A method that reaches a hundred parsecs and one that reaches a hundred thousand differ not only in what they can measure directly but in how many calibrators they hand to the next rung, and that second difference is the one that propagates.

The same triangle at three more distances is the whole of what the method is, and the range it covers is exactly the range over which it stops being usable.

The ladder from here #

Later rungs: the parallax ellipse, and how its shape depends on ecliptic latitude. Aberration derived, and the speed of light from it. Proper motion, and space velocities. Secular and statistical parallax, which use the Sun’s own motion as a longer baseline. Moving-cluster parallax, which gave the Hyades. Spectroscopic parallax, which is not a parallax at all. Cepheids and the period–luminosity relation. Type Ia supernovae. The Hubble tension, which is a disagreement between two ways of climbing the ladder. And Gaia’s zero-point problem, which is where the frontier currently is.

Bessel’s 1838 measurement is often called the first stellar distance. It is more accurately the first time anybody could say that the numbers astronomers had been using for two centuries — vast, vague, and unsupported — were roughly right.