Survey and follow-up telescopes

Why build one telescope that cannot study anything deeply and another that cannot find anything?

Modern transient astronomy runs on a pair of machines that each look badly compromised. One sweeps the whole visible sky every few nights and produces millions of detections a night, almost none of which it can say much about. The other can dissect a single object — composition, velocity, distance — and would take centuries to find that object on its own.

The obvious question is why nobody builds the sensible middle instrument: a telescope that finds things and studies them. It is not for lack of money; the two machines together cost more than one would. So there must be something in the physics or the operations that punishes the compromise, and it is worth finding out what.

Start with what each job actually demands, in the crudest terms.

Finding things means covering sky. The relevant quantity is not aperture alone but the product of collecting area and field of view — etendue, sometimes called grasp. Doubling the field of view is worth exactly as much as doubling the area, for this purpose.

Studying things means collecting photons from one target and, usually, spreading them out. A spectrograph takes the light of a single object and disperses it across hundreds or thousands of wavelength bins, so each bin gets a tiny fraction of an already faint signal. That demands aperture, and long exposures, and nothing at all from field of view — a spectrograph looking at one star gains nothing from a wide field.

Now ask whether these can be optimised together. Here is the pinch: a wide field of view over a large aperture is optically brutal. Keeping stars sharp across many square degrees requires correcting aberrations over the whole field, and the difficulty rises steeply with field size. So a wide-field design pays for its width in complexity and in optical compromises that a narrow-field telescope simply does not make. Rubin Observatory's 8.4-metre survey telescope reaches about 9.6 square degrees per exposure and needed a three-mirror design and a 3.2-gigapixel camera to do it; Keck's 10-metre mirrors see a field a small fraction of that size, and are correspondingly free to be excellent over it.

But the physical constraint is only half of it, and the operational half may matter more. Ask how each machine must be scheduled. A survey's value comes from uniformity: the same fields, on the same cadence, for a decade, so that anything changing shows up as a difference against a reference image. Interrupt it to chase an interesting object and you damage the very comparison that let you notice the object. Follow-up has the opposite requirement — a supernova must be caught within days, a kilonova within hours, so the instrument must be interruptible on demand.

So the two roles conflict on the clock as well as in the optics. A single instrument doing both would have to be simultaneously uninterruptible and interruptible. That is not a budget problem; it is a contradiction in the requirement.

Which gives the actual reason for the split, and it is not merely "each does what it is good at". Specialisation only pays if the outputs can be recombined, and the recombining is done by an interface: the alert stream. Rubin is expected to issue on the order of ten million alerts a night, published within about a minute of the exposure, in a standard format. Broker systems — ALeRCE, ANTARES, Fink, Lasair among them — ingest that firehose, filter and classify it, and hand a manageable shortlist to whoever wants it. The survey does not decide what matters. It publishes evidence in a common currency, and specialists downstream decide.

Notice what that interface buys. Because the survey's output is standardised rather than tailored, one survey can feed dozens of unrelated follow-up programmes — supernova cosmology, near-Earth asteroids, microlensing, variable stars — none of which needed to be anticipated when the telescope was designed. A generalist telescope serving the same communities would have to be time-shared, and every user would get a fraction of it. The split does not just make each machine better at its half; it multiplies how many parties can use the finding half at once.

And the cost of specialisation is real, so it is worth naming. The interface is a bottleneck: follow-up capacity is scarce, so most alerts are never examined by anything. Classification therefore has to happen upstream, on incomplete information, which means the choice of what gets studied is made by filters and brokers rather than by astronomers looking at objects. That is a genuine and much-discussed weakness of the arrangement, not a solved problem.

Three refinements.

First, this is not the old division between "big telescope" and "small telescope". Rubin's mirror is 8.4 metres; it is not a modest instrument. The axis of specialisation is field of view against depth-on-target and, above all, scheduling freedom, not size.

Second, the boundary is neither sharp nor fixed. Wide-field spectroscopic instruments observe thousands of objects at once and sit genuinely in between; space telescopes bend the constraints differently again. The claim is that the extremes are more productive as a pair than a compromise would be, not that nothing lives in the middle.

Third, the split is partly a consequence of the alert infrastructure rather than only a cause of it. Standard formats and public brokers make radical specialisation safe; without them a survey producing ten million nightly detections that nobody could act on would be an expensive way to fill a disc.

• How brokers decide what to promote, and what classes of transient that filtering is likely to miss.
• Why massively multiplexed spectrographs occupy the middle ground successfully where imaging telescopes do not.
• What a gravitational-wave alert demands that an optical transient does not.

Every term the collection defines is gathered in the glossary.