Deformable mirror
Applies continuous reflective tip, tilt, and paraxial defocus.
Open in the canvas →In the real world
A deformable mirror corrects a wavefront by changing its own shape. Light does not always arrive with the flat, well-behaved wavefront that optical design assumes: the atmosphere scrambles starlight, the eye's own cornea and lens distort a view of the retina, and tissue aberrates a focus long before a microscope reaches the depth it was built for. In every case the instrument is fine and the wavefront is not, so the fix is to add the conjugate of the distortion and cancel it.
The device is a thin reflective faceplate — a metallised membrane or a polished silicon layer — sitting on an array of actuators that push and pull it. Piezoelectric stacks, electrostatic pads, voice coils, and MEMS all appear, but the principle is the same: drive each actuator and the surface bends locally.
Reflection is what makes the mechanics easy. Displacing the surface by h shortens or lengthens the path twice, once on the way in and once on the way out, so a very small movement buys a large optical correction:
Working in a loop
A deformable mirror is almost never set by hand. It runs closed-loop with a wavefront sensor — usually a Shack–Hartmann, a lenslet array whose spot displacements measure local wavefront slope. The measured wavefront is decomposed into Zernike modes — tip, tilt, defocus, astigmatism, coma, spherical aberration, and higher — the actuator commands that best cancel them are computed, and the cycle repeats at hundreds or thousands of hertz, fast enough to keep up with atmospheric turbulence.
The low-order modes carry most of the power. Tip and tilt alone account for the largest share of atmospheric distortion, so big telescopes often split the job: a small, fast tip–tilt mirror handles the bulk motion while the deformable mirror, with hundreds or thousands of actuators, takes the higher orders. How well the higher orders can be corrected is set by actuator count and spacing — a mirror cannot reproduce structure finer than its actuator pitch, and the residue left over is called fitting error.
Designs trade smoothness against independence. A continuous facesheet gives a smooth surface but neighbouring actuators pull on each other, so each has an influence function rather than acting alone. A segmented mirror gives independent control at the price of gaps between segments, which diffract. MEMS devices are compact and cheap with limited stroke; bimorph and voice-coil mirrors offer large stroke with fewer actuators.
The applications are wherever a wavefront arrives spoiled: ground-based astronomy, adaptive-optics retinal imaging, deep-tissue and two-photon microscopy, laser beam shaping, and free-space optical communication.
In OpticalSetup
This element models the two lowest-order corrections a deformable mirror makes — the ones that dominate real aberration budgets — as a mirror with an adjustable curvature and an adjustable deflection. Three controls:
- Aperture, the size of the reflective face; the blue handle resizes it.
- Defocus focal length, which curves the surface. Positive values make it concave: light reflects converging, and the focus lands that many millimetres in front of the mirror — set 100 mm and the beam crosses the axis 100 mm away, set 200 mm and it crosses at 200. Negative values make it convex, so the return beam diverges from a virtual focus behind the surface. Leave it at zero for a flat mirror.
- Tip / tilt, on the purple knob, which steers the reflected beam.
Pair one with a wavefront detector and the correction becomes measurable rather than merely drawn: a flat mirror returns a collimated beam, a positive focal length returns a converging one, and a negative focal length a diverging one, with the detector naming the state and reporting the cone angle.
One convention to know
The Tip / tilt control applies its angle directly to the outgoing beam: set 5° and the reflected beam leaves 5° away from where it would have gone. That is the beam deviation, not the mechanical tilt of the surface — a real mirror tilted by 5° would deflect the beam by 10°. Rotating the whole element on the canvas does behave physically, giving the usual factor of two, so the two routes to a tilt are not equivalent. If you are reasoning about actuator stroke, halve the number.
Only tip, tilt, and defocus are modelled — the lowest Zernike orders. There is no astigmatism, coma, spherical aberration, or arbitrary surface shape, which is awkward given that correcting exactly those higher orders is the reason deformable mirrors exist; this element captures what they do first, not what makes them special. Nothing represents the mechanism either: no actuators, no actuator count or pitch, no influence functions or inter-actuator coupling, no stroke limit, and therefore no fitting error. The surface is perfectly smooth, so segment gaps and print-through never diffract, and there is no temporal response — the shape changes instantly rather than over the milliseconds a real mirror needs. Nothing closes the loop: there is no sensor driving the correction, so the shape is one you set by hand rather than one the system finds.