Atmospheric seeing limit
A Socratic walk-through of the atmospheric seeing limit — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why does a ground telescope twice as wide show no finer detail than a small one on most nights?
Everyone knows the rule: a wider telescope sees finer detail. Wavelength divided by aperture, and the bigger the aperture the smaller the angle you can split. Yet an eight-metre mirror on a mountain, on an ordinary night, delivers images no sharper than a garden telescope a hundred times smaller. The rule is not wrong — it is being overruled by something. What?
Reasoning it through
REASONING #Put a number on what is being lost. At a wavelength of 500 nanometres, an aperture of 0.1 metres gives a diffraction angle of 5 x 10^-7 / 0.1 = 5 x 10^-6 radians, and one radian is 206265 arcseconds, so that is about 1.0 arcsecond. Do the same for eight metres: 5 x 10^-7 / 8 = 6.3 x 10^-8 radians, about 0.013 arcseconds. The big mirror should be eighty times sharper. On a typical night it measures around one arcsecond — exactly what the ten-centimetre aperture was already achieving.
That coincidence is the clue. The eight-metre telescope is behaving as though its aperture were about ten centimetres. Which invites the obvious question: what, physically, is ten centimetres wide up there?
Think about what a mirror needs in order to resolve at all. It needs the wavefront arriving from the star to be flat across its whole face, so that the curve of the glass can bring every part of it into step at the focus. Air is not optically empty: its refractive index depends on density, density on temperature, and the atmosphere is stirred by turbulence into parcels of slightly different temperature. Light crossing a warm cell travels a shade faster than light crossing a cool one. By the time the wavefront reaches the ground it is no longer a plane but a gently corrugated sheet.
So the useful question is not "how wide is the mirror" but "how wide a patch of that sheet is still flat enough to work with". That patch has a name — the Fried parameter, r₀ — defined as the diameter over which the accumulated wavefront error is about one radian. At a good site in visible light it runs 10 to 20 centimetres. There is our ten centimetres.
Now the whole thing falls out. If the aperture is smaller than r₀, the wavefront across it is essentially flat and you get the textbook λ/D. If the aperture is much larger, you are not collecting one flat wavefront but a mosaic of many independent patches, each forming its own image displaced by its own local tilt. The long-exposure blur is set by λ/r₀ and stops caring about D altogether. Bigger glass buys light, not sharpness — a division of labour worth naming, since a companion piece on radio interferometry makes the same structural point about a dish: the metal in the middle is sensitivity, and it is the geometry, not the area, that sets resolution.
Two consequences follow immediately, and both are checkable. First, wavelength. Kolmogorov turbulence theory gives r₀ ∝ λ^(6/5) — an exponent I am recalling rather than deriving here — so the blur λ/r₀ ∝ λ^(-1/5) improves only feebly with wavelength, but the ratio D/r₀ collapses. Going from 0.5 to 2.2 micrometres multiplies r₀ by 4.4^1.2 ≈ 5.9, so an eight-metre telescope faces D/r₀ of about 13 instead of 80. Since the number of correcting elements a system needs scales as (D/r₀)², that is 180 actuators instead of 6400. This is why adaptive optics conquered the infrared first and is still fighting for the visible.
Second, time. The corrugation is not static: wind carries it across the aperture at the speed of the turbulent layer, tens of metres per second. The pattern therefore stays coherent for roughly r₀/v — 0.1 metres over 20 metres per second, about five milliseconds. Any correction must be measured and applied faster than that, which is why these systems run at hundreds or thousands of hertz.
The analogy
THE ANALOGY #Reading a printed page through the rippling air above a hot road: your eye is not the problem, and buying a bigger eye would not help. The letters are being reshuffled before the light reaches you, and the only cures are to get closer to the page, to wait for the air to settle, or to measure the ripple and undo it.
road shimmer is a single thin layer near the ground that mostly shifts the image sideways, whereas astronomical seeing is contributed by several layers at different heights and speeds — which is why correcting it well requires knowing where the turbulence sits, not just how strong it is.
Clarifying the model
THE MODEL #Two refinements the reasoning above glosses.
The first is that seeing is not "atmospheric absorption" or "haze". It is purely a phase effect — the photons still arrive, and just as many of them. They arrive at the wrong times, so the interference that would have concentrated them into a diffraction pattern instead scatters them across an arcsecond.
The second is that the diffraction limit never actually goes away. Freeze the exposure below a few milliseconds and the blur resolves into a scatter of speckles, and each speckle is λ/D across. The big mirror is still doing its job; the atmosphere is shuffling its output. That fact is what licenses speckle interferometry and lucky imaging, and it is the sharpest test available: if the one-arcsecond image were the mirror's own fault, freezing time would show one fat blob, not a swarm of fine specks. It shows specks.
The refuting observation: stop the same telescope down with masks of increasing diameter on one night and plot the measured stellar image width against aperture. The account predicts a fall as 1/D up to about r₀ and a flat line thereafter. If the width kept falling as 1/D at all apertures, the atmosphere would not be the binding constraint; if it were flat even below r₀, the fault would lie in the optics or the tracking.
A picture of it
THE PICTURE #How to readThe falling line is the diffraction limit λ/D — what the glass alone would give. The second line is the long-exposure width actually measured through air with r₀ near 10 cm; it tracks the first while the aperture is small, parts company where the two cross, and then flattens near one arcsecond no matter how much more glass you add. The flat stretch is the whole answer: everything to the right of the crossing is bought sensitivity, not bought sharpness. The curve is schematic in its exact shape near the knee, but the two asymptotes are the derived values.
What became clearer
WHAT CLEARED #The aperture in λ/D is not the mirror's diameter but the width of the largest flat patch of wavefront you can lay hands on — and on most nights the sky, not the observatory, decides what that is. Once you see it that way, adaptive optics stops being magic and becomes bookkeeping: measure the corrugation faster than the wind rearranges it, push a deformable mirror into its mirror image, and restore an aperture you already had.
Where to go next
ONWARD #- How a laser guide star manufactures a reference source when no bright natural star sits near the target, and the one aberration it structurally cannot measure.
- Why site testing measures the vertical profile of turbulence rather than a single seeing number, and what that buys.
Key terms
TERMS #| Term | What it means |
|---|---|
Fried parameter (r₀) | the diameter over which the incoming wavefront stays flat to about one radian; the effective aperture the atmosphere allows. |
| Seeing | the long-exposure angular width of a stellar image, roughly λ/r₀. |
| Coherence time | roughly r₀ divided by the wind speed aloft; how long the wavefront distortion holds still. |
| Speckle | one of the many λ/D-sized grains a short exposure resolves the blur into. |
Every term the collection defines is gathered in the glossary.