A large diameter helps when the mirror remains correct

Point the telescope higher above the horizon and the image changes. The star has not changed structure; the direction of the gravity load on the mirror and its support has changed. Even an excellent manufactured surface must keep its correct shape during observation. Light also passes through moving air, adding another, faster error.

Active optics maintains optical elements’ figure and relative alignment; phasing brings segments into agreement as parts of one wave surface; adaptive optics rapidly compensates for changing wavefront distortions, including atmospheric ones. One telescope may use all these mechanisms. Neither a motor count nor a single frequency threshold defines them. ESO: active and adaptive optics.

The article on telescope capabilities established that diameter provides collecting area and potential angular resolution. Here we follow how mechanics and control let a telescope use that potential. Our models are hypothetical; the author has not built or physically tested a large telescope or an adaptive-optics installation.

From a correction proposal to different working mechanisms

In October 1953, Horace W. Babcock published a proposal to compensate astronomical seeing through feedback. His scheme already included measurement, electronic processing and a controlled optical response. However, it was not a completed modern mirror with thousands of actuators: he considered an Eidophor with a controlled oil film. The historical proposal must be distinguished from later implementations. Babcock, PASP65, pp.229–232.

On the night of 22–23 March 1989, the NTT at La Silla obtained its first astronomical images. Its active loop maintained optical figure and alignment; building design and thermal control separately reduced local turbulence. That did not mean the primary mirror had removed the entire atmosphere. ESO announcement,23 March 1989.

In October of the same year, a prototype from ESO and French partners was tested on the 1.52 m telescope at the Observatoire de Haute-Provence. Its deformable mirror had 19 actuators and was corrected 100 times/s; the announcement compared infrared images with the loop on and off. This was a specific historical adaptive-correction demonstration, rather than a universal present-day specification. ESO: tests on 12–23 October 1989.

Keck demonstrated another part of the problem: first light on 24 November 1990 used9 segments of the eventual 36-segment primary. The first demonstration was therefore not a completed full aperture. Segmentation required precise surfaces and control of their relative positions. Keck: its own first-light history.

A thick solid blank is not the only route to a large area

Simply enlarging a rigid blank becomes costly in mass, transport, thermal equilibration and manufacture. Alternatives include a thin monolithic mirror with controlled support, a lightweight honeycomb structure, or a segmented aperture. Segmentation also divides production, transport and maintenance into smaller elements. A space telescope adds the need to fold the structure for launch.

A hexagonal outline does not define the optics. A segment must form the right part of the shared aspheric surface, with a specified type and orientation. The 39 m ELT’s M1 documentation describes 798 working segments and 133 optical types repeated across six sectors; an extra sector supports maintenance exchanges. Hexagons cannot simply be placed anywhere. ESO: M1 geometry and manufacture.

The ELT route starts with a low-expansion Zerodur glass-ceramic blank. It receives the required aspheric figure, is polished, cut, matched to its support and optically tested. After transport comes acceptance inspection and reflective coating. ESO’s 14 October 2026 announcement reported nearly 500 of the 931 elements being produced, including exchange segments, already in Chile. That was a delivery status: the ELT was still a project under construction, rather than an observatory with completed science results. ESO: production route and deliveries.

Cutting and mounting cannot be assumed neutral. In an early Keck technical account, Jerry Nelson described polishing a circular blank under controlled elastic stress; removing the forces produced the required asphere. Cutting the hexagon changed its figure and required further testing and correction. This was a specific historical manufacturing route, not a description of every current factory. Nelson: The Keck Telescope Project in the STScI 1989 proceedings.

Material and coating choices follow temperature and waveband, as explained in telescope materials. For example, the ELT receives a silver coating, whereas Webb’s beryllium segments receive thin gold for infrared light. These are coatings on precisely figured substrates, not solid pieces of silver or gold. ESO: ELT coating, text description, NASA: manufacturing Webb’s mirrors.

Supporting the glass, positioning a segment and changing its figure are three jobs

Three controlled heights can set the piston and two tilts of a rigid segment. Calling thin glass a segment does not make it perfectly rigid, however. Its weight must be distributed without the support introducing unwanted sag. Relative element placement and deformation of each element are separate quantities.

In ELT M1, three PACT positioning actuators move the segment assembly. The glass has 27 axial support points arranged in load-distributing whiffletrees; lateral support and clocking restraint serve other functions. Nine separate warping-harness actuators change support forces and moments to make small figure corrections. Drawing only three motors directly holding the glass at three points would omit an important part of the design. Cayrel 2012, Sections 3.2–3.6, Figs.3–7.

Distributed supports carry the glass; actuators move the frame

Distributed supports carry the glass; actuators move the frameP1P2P31234567
  1. Segment and 27 axial contacts
  2. Three nine-point whiffletrees
  3. Moving frame
  4. Three PACT positioning actuators
  5. Fixed frame on the telescope structure
  6. Moment at a support joint: separate figure correction in principle
  7. Example edge-sensor pair
Our simplification of ELT M1 after Cayrel 2012 Figs.3–7 and ESO M1. The 27 contacts and three PACTs are distinct. The spatial mechanism is flattened into a connection tree; coordinates, thicknesses, gaps and glass profile are schematic; lateral support and clocking restraint are omitted. Example edge pairs are not a complete sensor layout; neighbouring segments are partial. One arc illustrates a warping moment, not the locations of all nine shape actuators. Forces, elastic sag and strength are not calculated. The plane in the mathematical branch is a separate numerical example, not dimensions of this assembly. Cayrel 2012; ESO M1.

Sensors near neighbouring edges report relative displacements. The loop can use these to maintain a consistent surface while the overall structure changes slightly. But an electromechanical sensor zero must still be tied to optical performance. Acceptance and calibration must verify figure, relative placement and the combined wavefront. ESO: M1, M4 and M5 control loops.

Branch · Segment positionOptional · may be skipped

Deeper: what three heights really determine

Our hypothetical example places three moving-frame points 120° apart on a circle of radius R=200 mm. Let x and y be in millimetres and height z in nanometres. A small rigid plane is:

z(x,y) = a + bx + cy, z_(i) = a + bRcos⁡θ_(i) + cRsin⁡θ_(i), θ_(i) = 0,(2π)/(3),(4π)/(3).

Here a is piston and b and c are slopes in nm/mm. The symmetric triangle has the separate inverse solution:

a=\frac{z_1+z_2+z_3}{3},\qquad b=\frac{2}{3R}\sum_i z_i\cos\theta_i,\qquad c=\frac{2}{3R}\sum_i z_i\sin\theta_i.

For a=75 nm, b=0.1 nm/mm and c=−0.15 nm/mm, the heights are 95, approximately 39.019, and 90.981 nm. A 0.1 nm/mm slope is 100 nanoradians. The inverse calculation recovers the initial three parameters; adding the same height to all points changes only a.

This model does not calculate glass sag between supports, contact stress, lateral loading or warping-harness behaviour. Its radius and numbers are not ELT support dimensions. It explains controlled degrees of freedom, rather than proving structural strength.

Matching ray directions does not ensure matching phase

A wavefront is a surface of equal wave phase. For a distant axial star, it is approximately planar before an ideal objective; approaching the focus, a converging spherical form is required. Wavefront error is a departure from that required ideal form, rather than a demand for flatness everywhere.

A whole segment may move axially: this is piston. Its figure can remain correct while the optical path changes. Tip/tilt changes the direction and placement of the segment’s contribution. Deformation changes the profile itself. Turbulence adds a changing optical-path pattern across many aperture regions. ESO: wavefront error sources.

Piston, tilt and an uneven wavefront leave different signatures

Piston: different constant optical paths

Piston: different constant optical paths-1000100W / nmu−11

Tip/tilt: different slopes

Tip/tilt: different slopes-1000100W / nmu−11

Variable uneven wavefront: illustrative section

Variable uneven wavefront: illustrative section-1000100W / nmu−11
Our 1D examples: W in nanometres relative to the ideal front, u from −1 to 1 without a telescope diameter. W is not mirror surface height. The segment gap is schematic; the turbulence section is a sum of sinusoids, not an atmospheric measurement or spectral model. At normal reflection an ideal added surface h=−W/2 would cancel this front; real spatial constraints, latency, noise and saturation are absent.

A sharp combined image requires more than bringing separate spots to approximately one location. At each wavelength, contributions from different pupil regions must add with the required relative phases. Phasing therefore demands a more precise criterion than a mosaic that looks level. A common phase added to the entire pupil does not change image brightness; differences between regions matter.

Branch · Phase and fluxOptional · may be skipped

Deeper: a quarter wavelength of height can extinguish the centre

Divide an ideally illuminated pupil into two equal areas. Move one surface by δ at nearly normal reflection. Choose the sign of δ so that positive displacement increases optical path. Path difference W is approximately twice the surface movement; incidence angle i introduces a cos i factor:

W = 2δcos⁡i, Δϕ = (2πW)/(λ).

Our normal-incidence example uses i=0. Equal complex amplitudes in the central direction give normalised intensity:

I_(0) = (∣(1 + (e)^(iΔϕ))/(2)∣)^(2) = (cos⁡)^(2) ⁣((2πδ)/(λ)).

At λ=550 nm, displacements δ=0,68.75,137.5 and 275 nm give I₀=1,0.5,0 and 1. The restored central brightness at 275 nm reveals phase ambiguity: one full cycle has been added at this wavelength, but not necessarily at another. One bright centre in one spectral band therefore does not prove correct broadband phasing.

One wavelength allows an ambiguous result

Surface step δ = 0 nm
Phase difference = 0π

δ=0 nm; I₀=1.000ReImI₀ = 1.000

Surface step δ = 68.75 nm
Phase difference = 0.5π

δ=68.75 nm; I₀=0.500ReImI₀ = 0.500

Surface step δ = 137.5 nm
Phase difference = 1π

δ=137.5 nm; I₀=0.000ReImI₀ = 0.000

Surface step δ = 275 nm
Phase difference = 2π

δ=275 nm; I₀=1.000ReImI₀ = 1.000
Our two-equal-area pupil model: λ=550 nm, normal reflection, W=2δ; positive δ increases path. The circle has unit radius. The solid vector is the first amplitude, the dashed vector the second; the bold vector with a circle is their mean. At zero phase difference they coincide; at π their mean vanishes. The bar is normalised central-direction intensity, not the percentage of all photons in a real core. A 275 nm step returns I₀=1 at 550 nm only; at 650 nm I₀≈0.784. An independent discrete Fourier check conserves total flux; absorption is not undone. This is not a PSF, observation or ELT/Webb segment geometry.

A zero at the centre does not mean photons have disappeared across the entire image plane. Our independent discrete Fourier check of 64 pupil samples shows redistribution: changing only phase leaves the sum of intensities across all discrete directions constant. This Parseval check is independent of the single-centre formula. If the atmosphere transmits a power fraction T, amplitude has a √T factor; after perfect phase correction the peak relative to the lossless case is T, not 1. This is a 1D teaching model, not an ELT or Webb two-dimensional photographic PSF.

The adaptive loop measures the residual and corrects it again

The atmosphere changes during observation. One optical adjustment at the beginning of the night is insufficient. Some reference-star light is directed to a wavefront sensor. A Shack–Hartmann sensor, for example, divides the pupil with a small-lens array; movements of its star spots provide information about local wavefront slopes. A camera does not see a ready-made table of “error in nanometres”: calibration and reconstruction are required.

Before operation, known control changes must be linked to sensor response: surface motion, camera spot displacement and electrical signals use different units. Calibration establishes scales, signs and mutual effects. A wrong sign can increase an error instead of reducing it. The modes accessible to the chosen measurement must also be checked; a fast camera does not make its sensor all-knowing.

The computer estimates observable components, converts them to commands and drives a deformable mirror. The mirror introduces an opposite optical error. The next measurement shows the residual after correction, and the cycle repeats. Science instrument and sensor paths should be matched; otherwise some science-image errors remain unseen by the sensor. ESO: sensing and reconstruction.

Corrected light brings information back into the loop

Corrected light brings information back into the loop12345LightDataCommands
  1. Deformable mirror changes optical path
  2. Science detector receives corrected light
  3. Sensor provides images or local slope estimates
  4. Reconstruction and controller estimate observable modes and residuals
  5. Calibrated conversion of commands into actuator motion or force
Generic closed loop for a reference star in the working direction, not a particular ELT or Keck drawing. Reflections are unfolded; lines give order and direction, not exact rays or focusing surfaces. The sensor views the corrected channel. Images/slopes become estimated wavefront errors and then commands; a camera does not measure nanometres without calibration. Laser and off-axis modes add paths. Solid: light; dashed: data/control. Principle after ESO Active and Adaptive Optics. ESO.

The ELT’s adaptive M4 is a separate2.4 m mirror with six thin petals, rather than the 39 m collecting M1. Nominal M4 petal thickness is 1.95 mm, compared with about 50 mm for an M1 segment. The petals also use Zerodur glass-ceramic. A glass-ceramic name alone does not define mechanical function: geometry, support and control matter. More than 5000 magnetic actuators deform the shell; a rigid reference structure and local capacitive sensors help control it. Documentation describes up to1000 shape updates/s and 70000 local position measurements/s. These belong to different stages: local measurement rate is neither atmospheric camera rate nor whole-system correction bandwidth. ESO: M4 construction.

Even M1 has fast internal position control: the ELT control page gives 500 Hz. Thus “active means slow; adaptive means everything above some frequency” is too crude. Distinguish error cause and spatial structure, optical sensing, internal actuator loops and slower command offloading between elements. ESO: nested loops and offloading.

Speed without a good measurement does not give the right correction

A late command describes an earlier atmospheric state. Excessive correction gain can make a loop oscillate. A shorter exposure also supplies fewer photons per measurement, raising the demands on reference brightness and sensor noise. The design must balance spatial and temporal sampling, measurement accuracy, latency and available mirror stroke.

Branch · Discrete controlOptional · may be skipped

Consider one already observable wavefront mode. Disturbance d and correction command u are optical-path coefficients in nanometres; residual r=d+u. This does not claim a normal slope sensor measures global piston. After a step, let the disturbance be constant, with no noise or saturation and integrator gain g.

If measured rₖ is used for the next sample, uₖ₊₁=uₖ−grₖ, giving:

r_(k + 1) = (1 − g)r_(k).

Decay requires 0 < g < 2. Now add another full sample of delay, making the command depend on rₖ₋₁. Then:

r_(k + 1) = r_(k) − gr_(k − 1), (q)^(2) − q + g = 0.

Both roots q must have magnitude below 1; this hypothetical loop requires 0 < g < 1. Gain 1.2, which works without extra delay, is now unstable. Numerical series were checked against analytical solutions; sinusoidal disturbance response was separately checked through the discrete transfer function.

One extra delay changes stability

g=0.5 · no extra delay

g=0.5 · no extra delay-125012506121824r / nmSample number

g=0.5 · one extra sample

g=0.5 · one extra sample-125012506121824r / nmSample number

g=1.2 · no extra delay

g=1.2 · no extra delay-125012506121824r / nmSample number

g=1.2 · one extra sample

g=1.2 · one extra sample-10000100006121824r / nmSample number
Our single-mode integrator: r₀=100 nm; the preceding unmeasured residual for extra delay is set to zero. Command units are equivalent optical path, not actuator stroke. The first three panels span ±125 nm, the fourth ±1000 nm. No noise, saturation, elastic dynamics or mode coupling; sample duration is unspecified. The unstable case is not a real observatory specification. Series agree with analytical solutions and a separate frequency-response check.

Time is in sample numbers. Without a specified sample duration, it cannot become a particular instrument’s hertz. Real loops are multichannel and have noise, calibration, actuator coupling and saturation; these inequalities are not ready-to-use ELT settings.

The reference star must also reveal sufficiently similar atmospheric distortion. Far from the science target’s direction, rays pass through different air regions. This limits the well-corrected field. More stars and deformable mirrors enable more elaborate modes but do not remove the need to assess the measured atmospheric volume.

A laser reference helps where a suitably bright natural star is absent. In Keck’s system, a laser excites sodium about 90 km above Earth. It is a nearby artificial source, not illumination of a distant galaxy. Returning light samples a cone, while distant starlight covers a different volume. This Keck mode also needs a natural tip/tilt reference: the laser measurement includes an unknown contribution from the upward path. Keck: laser-guide-star limitations.

Space removes terrestrial seeing, but not the need for alignment

Webb does not correct Earth’s atmospheric turbulence. Its 18 segments nevertheless had to unfold, have their individual images identified, match figure and phases, and be checked across the science instruments’ fields. After launch, a mechanically deployed mosaic was not yet an optically ready primary.

Each primary segment has six positioning and orientation actuators plus one for curvature. This differs from an ELT segment assembly’s three PACTs. During manufacture, Webb’s mirrors were measured cold, repolished using those results, coated and checked cold again. Shape under operating conditions matters, rather than only in a warm room. NASA: actuators and cryogenic testing.

In flight, information comes from images, particularly NIRCam, including exposures with known defocus. Processing retrieves wavefront error; analysis and command preparation take place on Earth. STScI describes a system without continuously operating edge sensors: slow thermal changes permit periodic optical measurements. Measurement and actual mirror motion are different events; corrections happen as needed. STScI: wavefront sensing and control concept, STScI: alignment maintenance.

The absence of ground-like self-weight sag does not mean “there is no gravity in space”. Free fall, a different loading regime, a stable thermal structure and continued monitoring matter. Mirror alignment must also be distinguished from observatory pointing or image stabilisation.

Correcting phase does not recover all lost light

Adaptive optics can return more already collected stellar energy to the useful image core, recovering some lost sharpness. It does not return photons absorbed by atmosphere, cloud or an optical component; it cannot open an opaque waveband or replace aperture. A laser guide star adds no cosmic photons to the galaxy measurement either.

Incomplete correction may result from unmeasured spatial scales, latency, reference noise, different directions, insufficient stroke, science-path errors or the optics themselves. One attractive central star does not test every field point and condition. Even a correct basic reflecting layout needs separate surface, alignment and mechanical checks.

Large-system acceptance requires segment measurements in defined supports, actuator and sensor calibration, optical phasing, loop-response checks and science-image assessment. Recording wavelength, field, reference brightness, atmosphere, temperature and timing makes a result interpretable. A large mirror becomes useful through a consistent chain of manufacture, support, measurement and correction.