Celestial Reference Systems
From Hipparchus to quasars: the ICRS, its realisations and astronomical time scales
Rhodes, around 130 BC
Hipparchus has the best star catalogue in the world, because he made it. He also has something almost nobody else has: older measurements. A century and a half earlier, the astronomer Timocharis of Alexandria recorded where the bright star Spica sat relative to the autumn equinox — the point where the Sun crosses the celestial equator heading south. Eight degrees.
Hipparchus measures the same thing. Six degrees.
It would be easy to blame Timocharis. Instruments were crude; two degrees is four full Moons. But Hipparchus checks other stars, and they have all shifted, by the same amount, in the same direction. The stars have not moved relative to each other. The reference point has moved underneath all of them.
He has just discovered that the sky's coordinate system is drifting — about one degree every 72 years. Two thousand years later, we are still dealing with the consequences. Every orbit you have computed so far is written in a coordinate system. This unit is about what that coordinate system is actually nailed to, and how you move between the dozen that exist.
Watch first · presented by Shoaib Aziz
Unit 3 gave you the working definition: a frame is an origin, three axes and a time. That is enough to write down an orbit. It is not enough to agree on one with somebody else, and agreement is the entire job. Astronomers split the idea in two, and the split is worth learning exactly:
A reference system is the recipe: where the origin is, how the axes are supposed to be oriented, which physical models and constants are assumed, which time scale applies. It is a set of definitions. You cannot point a telescope at it.
A reference frame is the realisation: an actual catalogue of objects with measured coordinates that makes the recipe usable. It is what you load into software. It has error bars.
The International Celestial Reference System (ICRS) is the recipe. ICRF1, ICRF2 and ICRF3 are three successive catalogues that realise it, each better than the last. When a paper says "ICRF", it almost always means the frame — and the difference is the difference between a definition of the metre and a ruler.
Loading interactive…
💡Dynamical versus kinematic frames
There have only ever been two strategies for defining celestial axes.
Dynamical: take the axes from the motion of the solar system itself — Earth's equator (from its spin) and the ecliptic (from its orbit), with at the equinox where the two cross. Everything from Hipparchus to FK5 is dynamical. The problem is in the name: the equator and the ecliptic are moving, so the frame must be tied to a date and corrected for the motion with a model — and the model is only as good as our knowledge of Earth's wobble.
Kinematic: take the axes from distant objects assumed not to rotate relative to the universe. The ICRS is kinematic. It does not care where Earth's equator points; it does not have an equinox at all. It is defined by the directions to roughly three hundred quasars.
The move from dynamical to kinematic, made official in 1997, is the single most important idea in this lesson. It means the reference no longer depends on the thing it is used to measure.
Why not use stars? Because stars move. In 1718 Edmond Halley noticed that Sirius, Arcturus and Aldebaran were measurably out of the positions Ptolemy had recorded. Stars have proper motion — genuine sideways movement across the sky, because they orbit the galaxy at their own speeds and sit at very different distances. Barnard's Star crosses a full Moon's width in 180 years. Arcturus moves about 2.3″ per year.
If your axes are defined by a set of stars, your axes move with them. Catalogues like FK5 handled this by publishing proper motions alongside positions and propagating the frame forward in time — which works exactly as well as the proper motions are known, and degrades every year that passes since they were measured.
Try it yourself below. Nail a frame to any two objects in the sky and run the clock forward.
Loading interactive…
A quasar is the brilliant core of a galaxy whose central black hole is feeding, typically billions of light-years away. Even if one moved sideways at a thousand kilometres per second, at that distance its proper motion would be a fraction of a microarcsecond per year — a thousand times smaller than anything a star does, and below what we can measure.
Quasars are invisible to the eye but bright at radio wavelengths, which is how they are measured: Very Long Baseline Interferometry (VLBI). Radio telescopes on different continents record the same quasar at the same moment, time-stamped by hydrogen-maser clocks. Correlating the recordings gives the difference in arrival time of a wavefront across a baseline thousands of kilometres long — and from that, the quasar's direction to a few tens of microarcseconds.
The same observations measure Earth's orientation as a side effect. That is not a coincidence: the frames in this unit are built and maintained by the same handful of instruments.
| Frame | Adopted | Sources | Defining sources | Axis stability | Notes |
|---|---|---|---|---|---|
| ICRF1 | IAU 1997 (in force 1998) | 608 | 212 | ~20 µas | First kinematic frame; aligned to FK5 J2000 for continuity |
| ICRF2 | 2009 | 3,414 | 295 | ~10 µas | Better southern coverage, five times more precise |
| ICRF3 | IAU 2018 (in force 1 Jan 2019) | 4,536 at S/X band; 824 at K; 678 at X/Ka | 303 | ~10 µas | First multi-frequency frame; models galactic aberration drift (~5.8 µas/yr) |
| Gaia-CRF3 | 2022 | ~1.6 million (optical) | aligned to ICRF3 | µas-level alignment | First optical realisation at ICRF quality — from a spacecraft, not from radio |
💡Where does the ICRS x-axis point?
Not at any quasar. The ICRS axes were deliberately placed as close as possible to the old FK5 J2000 axes, so that a century of catalogues and software would not suddenly be wrong by a large angle. The ICRS pole is within about 18 mas of the J2000 mean pole, and its right-ascension origin within 15 mas of the J2000 dynamical equinox.
So the most modern frame we have is pointed — by convention — at the place where a 2,000-year-old dynamical frame happened to point on 1 January 2000. After that, it is defined purely by quasars, and stays put while the equinox wanders off.
That small leftover misalignment has a name, the frame bias, and it is the first conversion in this unit. Going from ICRS axes to the old J2000 (also written EME2000: Earth Mean Equator and Equinox of J2000) is a fixed rotation of about 23 milliarcseconds:
The frame bias, first-order form (IERS Conventions 2010). Twenty-three milliarcseconds is 4.7 m at GEO, 43 m at the Moon — and 25 km at Mars.
Loading interactive…
⚠The "J2000" that is not J2000
JPL's planetary ephemerides (DE430, DE440) are aligned to the ICRF. NASA's SPICE toolkit labels that frame "J2000", for historical reasons, and treats the two as identical. For almost everything that is harmless — 23 mas is invisible to a planetary flyby.
But it means "J2000" in a SPICE kernel and "J2000" in an equinox-based flight code built from FK5 can differ by the frame bias. On a precise GEO, lunar or deep-space navigation problem, find out which J2000 each side means before you exchange a single vector. Same name, different frame — the Mars Climate Orbiter failure mode, wearing a different coat.
💡Same axes, different origin: BCRS and GCRS
The ICRS axes can be attached to different origins. Two matter:
BCRS — Barycentric Celestial Reference System. Origin at the solar system's centre of mass, which sits just outside the Sun's surface when Jupiter and Saturn line up. Interplanetary trajectories, planetary ephemerides and pulsar timing live here.
GCRS — Geocentric Celestial Reference System. Origin at Earth's centre, axes kinematically aligned with the BCRS. This is the modern 'ECI'. It is where Earth-orbit propagation runs.
Moving between them is mainly a translation by Earth's barycentric position — about 1 AU. There is no rotation in the Newtonian picture; general relativity adds a slow geodesic precession of the GCRS axes of about 1.9″ per century, which the IAU definitions absorb so that you never have to apply it by hand.
The last ingredient of a frame is a clock, and the J2000 epoch shows why it is not optional. J2000.0 is defined as 1 January 2000, 12:00:00 Terrestrial Time — Julian Date 2451545.0. Not UTC.
There are five time scales you will meet in this unit, and each answers a different question:
TAI — International Atomic Time: the weighted average of ~450 atomic clocks. Smooth, uniform, never adjusted.
TT — Terrestrial Time = TAI + 32.184 s exactly. The time argument of precession, nutation and ephemerides seen from Earth. The 32.184 s exists to stay continuous with the older Ephemeris Time.
UTC — what your phone shows: TAI minus an integer number of leap seconds (37 since 2017), kept within 0.9 s of Earth's actual rotation.
UT1 — Earth's actual rotation angle, expressed as a time. It is not a clock: it is measured, by VLBI, after the fact, and published by the IERS.
TDB — Barycentric Dynamical Time, the time of the solar-system ephemerides; it differs from TT by at most ~1.7 ms, periodically, because Earth's clock runs at a varying rate along its elliptical orbit.
In January 2000, TAI − UTC was 32 s, so J2000.0 = 11:58:55.816 UTC. Lab 3 makes you prove it.
Time, he said, is what keeps everything from happening at once.
🕰The planned end of the leap second
Leap seconds exist so that UTC never drifts more than 0.9 s from UT1. They are inserted at irregular intervals announced six months ahead, and they break software with depressing regularity — in 2012 a single leap second took down parts of Reddit, LinkedIn and the Qantas booking system.
In November 2022 the General Conference on Weights and Measures voted to stop inserting them by 2035, letting UT1 and UTC drift apart by more than a second. For astrodynamics nothing changes in principle — you already had to look up UT1 − UTC — but the old shortcut "UTC is close enough to UT1" will quietly stop being true.
🛰Navigating by quasar
The Deep Space Network uses the ICRF directly to fly spacecraft. In ΔDOR (Delta-Differential One-way Ranging), two widely separated antennas observe a spacecraft and then, within minutes, a quasar a few degrees away on the sky. Differencing the two removes most clock, atmosphere and station errors, and measures the spacecraft's direction relative to the quasar to about a nanoradian.
One nanoradian at Mars is about 200 m. ΔDOR was central to the approach navigation of Curiosity and Perseverance into their landing ellipses. The same quasars that define your coordinate axes are, quite literally, the stars these spacecraft steer by.
↓But there is a problem
The ICRF does not move. That is its whole value.
But nobody launches from the ICRF. Satellites orbit Earth, and Earth's equator — the plane every inclination is measured from — is tilted, and the tilt is turning. The axis traces a cone 47 degrees across, once every 25,800 years, and nods on top of that every 18.6 years.
Polaris was not the pole star when the pyramids were built. It will not be the pole star when Vega takes over. Every frame tied to Earth's equator has to be dragged along behind that wobble — and before you can do that, you need to understand what is pushing on the top.