Large mirror flexure
A Socratic walk-through of large mirror flexure — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why must the biggest telescope mirrors be built deliberately floppy and pushed back into shape all night?
The obvious way to keep a precision surface precise is to make it rigid — thick, stiff, immovable. Yet the eight-metre mirrors built since the 1990s went the other way: barely 17 centimetres of glass across eight metres of span, a proportion closer to a dinner plate than a paving slab. And every one sits on a bed of computer-controlled pistons that push it back into figure all night long. Why deliberately build the floppy version?
Reasoning it through
REASONING #Begin with what the surface has to achieve. Optical figure is judged against the wavelength of light: an error of λ/20 is a reasonable working tolerance, and at 500 nanometres that is 25 nanometres. Not 25 micrometres. Twenty-five billionths of a metre, held across eight metres of glass, while the whole assembly swings from pointing at the zenith to pointing near the horizon.
Now ask what gravity does to a plate. A disc of radius R and thickness t, made of a material of density ρ and stiffness E, sagging under its own weight, deflects by roughly
δ ≈ C · ρ g R⁴ / (E t²)
where C is a dimensionless coefficient of order a tenth that depends on how the plate is held. That grouping is the whole story, and it can be read off without solving anything: sag grows as the fourth power of span, and falls only as the square of thickness.
Put the modern numbers in. Glass-ceramic runs near 2500 kg/m³ and 9 x 10^10 Pa — values I am recalling. With R = 4 m and t = 0.175 m: ρgR⁴ = 2500 x 9.8 x 256 ≈ 6.3 x 10^6, and Et² = 9 x 10^10 x 0.0306 ≈ 2.8 x 10^9. The ratio is 2.3 x 10^-3 metres, so with C near a tenth the sag is a couple of hundred micrometres.
Compare that with the 25-nanometre tolerance. The mirror droops by something like ten thousand times more than it is allowed to. So the question inverts itself: not "why build it floppy" but "how could anything be built stiff enough?"
Try. Hold the shape of the plate fixed — same t/R — and scale it up. Then δ ∝ ρgR⁴/(E R²) ∝ R²: doubling the diameter quadruples the sag while the tolerance stays exactly where it was. To hold δ constant you must instead let t ∝ R², and then the mass, which goes as ρR²t, goes as R⁴. Palomar's five-metre mirror, at roughly 14.5 tonnes (a recalled figure), scaled to eight metres by that rule would be 14.5 x (8/5)⁴ ≈ 95 tonnes of glass — before the cell, the mount, or the building that has to swing it. It also takes months to anneal and hours to come to thermal equilibrium every evening, which ruins the seeing right above it.
So passive stiffness does not merely get expensive. It loses, structurally, and it loses faster the more you spend.
What is left? Look again at C, the coefficient that encodes how the plate is held. A disc supported only at its rim spans R. A disc supported at N points spread over its back spans, on average, something like R/√N per cell — and since sag goes as span to the fourth, it falls as N². That is the lever. We need a factor of about 8000 taken out of 200 micrometres, so N ≈ √8000 ≈ 90 support points. Real eight-metre telescopes use of order 150 axial supports, which is the right order for a crude estimate that ignores edge effects and the fact that the pistons must also correct, not merely support.
And that last point is the design's real move. Once the mirror is a thin shell on many actuators, the supports are not just holding it up — they can be commanded. Send starlight to a wavefront sensor, decompose the measured error into bending modes, and push. Correction becomes a control loop rather than a property of the glass.
The analogy
THE ANALOGY #A tent, not a shed. A shed holds its shape because its walls are stiff enough to resist the wind on their own; a tent holds its shape because a light fabric is held in tension by many guy lines, each adjusted. Scale a shed up and the walls must thicken until the thing cannot be built; scale a tent up and you mostly add guy lines.
tent guys are set once and left, whereas a mirror's supports are re-commanded continuously against a load that changes as the telescope moves — and, unlike a tent, the tolerance is not "looks taut" but a fixed fraction of a wavelength that does not relax as the structure grows.
Clarifying the model
THE MODEL #Two refinements worth making explicit.
First, active optics is not adaptive optics, and the confusion is common. Active optics corrects the mirror's own flexure — gravity as the telescope tips, thermal gradients as the night cools, wind loading — which are slow, so the loop runs every tens of seconds. Adaptive optics corrects the atmosphere's corrugation of the incoming wavefront, which changes in milliseconds and needs a separate, much faster, much smaller deformable mirror. They sit in the same light path and answer different questions.
Second, floppiness is chosen, not merely tolerated. A stiff mirror is not just heavy; it is uncorrectable. If the glass resists bending, the actuators cannot bend it into the shape the wavefront sensor is asking for. Compliance is what makes the control authority possible. The design deliberately trades a property that was never going to be sufficient for one that can be actively supplied.
The refuting observation: track the wavefront error as the telescope is driven from zenith toward the horizon with the correction loop open. The account says the dominant terms are gravity-driven, so they must vary smoothly with the component of gravity across and along the optical axis, repeat exactly when the same elevation is revisited, and vanish on return to zenith. A figure error frozen into the glass by polishing would do none of that — it would sit unchanged in the mirror's own frame. Observatories in fact build elevation look-up tables for precisely this reason, which is the prediction confirmed in operational form.
A picture of it
THE PICTURE #How to readThe two boxed statements at the top are the conditions the design must meet at once; the three below are the hardware that meets them. Read each arrow as "this part discharges that condition". The point is which arrow is missing: the thin blank satisfies the mass constraint and nothing else — on its own it fails the figure requirement outright — and the figure is discharged only by the actuators, verified continuously by the sensor. The family is repurposed here: it is normally used for tracing engineering requirements, and I am using it to show that no single component covers both conditions.
What became clearer
WHAT CLEARED #Stiffness is a losing scaling law. Sag rises as the fourth power of span while the tolerance stays pinned at a fraction of a wavelength, so past a few metres no achievable thickness closes the gap — and chasing it drives mass as the fourth power of diameter. The way out is to stop asking the glass to hold its own shape and start measuring the error and cancelling it, which requires a mirror compliant enough to be pushed.
Where to go next
ONWARD #- Why the largest telescopes now abandon the monolith entirely for segmented mirrors, and what edge-sensing must then replace.
- How thermal gradients within the glass produce figure errors that gravity models cannot predict.
Key terms
TERMS #| Term | What it means |
|---|---|
| Optical figure | how closely a mirror's surface matches its intended shape, measured against the wavelength of light. |
| Meniscus mirror | a thin, uniformly curved shell blank, stiff enough to handle but not to hold figure unaided. |
| Active optics | the slow loop correcting the telescope's own flexure by commanding force actuators behind the mirror. |
Every term the collection defines is gathered in the glossary.