Adaptive Optics (AO) techniques allow ground based observatories to dramatically improve resolution by actively compensating for the effects of Astronomical Seeing.

The atmospheric effects are quite variable in both time and location. A parameter called Isoplanatic Angle (IPA) is used to express the angular extent over-which a given wavefront correction optimized for one point (usually a guide star, artificial or natural) will be effective. As an example, Table 9.1 in this Giant Magellan Telescope resource shows values for IPA scaling almost linearly (actually: $\sim\lambda^{6/5}$) from 176 arcseconds at a wavelength of 20 microns to only 4.2 arcseconds at 0.9 microns.

This suggests an IPA of 2 to 3 arcseconds for visible wavelengths, which taken by itself is not a killer limitation.

However, it seems almost all currently active AO work is done exclusively in various infrared wavelengths, apparently down to 0.9 microns but no further. (AO is also implemented computationally to array data in radioastronomy.)

Is this because the observed wavelength needs to be longer than the guide star monitoring wavelength? Because it is simply much harder and there is always Hubble above the atmosphere for visible work so it's not worth the extra effort, or is there another more fundamental reason?

I'm not looking for speculation or opinion, I'd like a quantitative explanation (if that applies) - hopefully with a link for further reading - thanks!

  • 2
    $\begingroup$ I'm not proposing this as an answer because it's an opinion - and I cannot speak to the justification being made by the pros. But I think that the reason it's being done in NIR is largely because the 'exciting' astronomy is now being done in non-visible wavelengths, and NIR has a lower extinction rate in our atmosphere when compared to other non-visible wavelengths. $\endgroup$ May 6, 2016 at 13:18
  • $\begingroup$ @EastOfJupiter thanks! The reason I asked this is I'd recently heard about the Hubble being chronically heavily oversubscribed. I'm not asking why most of the work is in IR, I'm asking why none of the work is ever in visible. If Hubble is the (seemingly) only dee-sub-arcsec visible wavelength telescope for all of humanity, it seems there is significant pressure to open up at least one alternate source. It's the zero I'm wondering about. $\endgroup$
    – uhoh
    May 6, 2016 at 13:25
  • 1
    $\begingroup$ There are instruments working down to about 600nm now, but the question still stands. $\endgroup$
    – ProfRob
    Oct 25, 2018 at 23:31
  • $\begingroup$ @RobJeffries I'd love to hear about that! You may have already eluded to the possibility in your 2016 comment. There is also the somewhat-related question Will the E-ELT use Adaptive Optics at visible wavelengths? $\endgroup$
    – uhoh
    Oct 25, 2018 at 23:44

2 Answers 2


There's a pretty good discussion at this page.

There are several factors at work:

  1. The smaller isoplanatic angle, as you note. This limits how much of the sky you can observe with AO, since your target needs to be within the isoplanatic angle of a bright enough references star. (Even with laser guide stars, there is still a need for a reference star for "tip/tilt" correction.) The difference in angular area on the sky means that the area of the sky that can theoretically be observed with AO will be about 20 times larger in the near-IR than in the optical, just from the difference in isoplanatic angle.

  2. The effects of turbulence are stronger and have shorter timescales in the optical. This has three effects:

    A. The corrective optics (e.g., deformable mirror) need to have more movable parts ("a near-perfect correction for an observation done in visible light (0.6 microns) with an 8-m telescope would require ~ 6400 actuators, whereas a similar performance at 2 microns needs only 250 actuators.") and need to operate on a faster timescale.

    B. In addition to the electromechanical complexity, you'll have to do far more in the way of calculations to drive all those actuators, and on a shorter timescale. So the computing power required goes up.

    C. In order to provide the inputs for the corrective computations, you need to observe the reference star on a much finer angular scale ("A large number of actuators requires a similarly large number of subapertures in the wavefront sensor, which means that for correction in the visible, the reference star should be ~ 25 times brighter than to correct in the infrared."). This limits how much of the sky you can do AO for even more: a star that might be bright enough in the near-IR to correct a region 20-30 arcsec wide isoplanatic patch won't be bright enough to correct the corresponding 5-arcsec-wide isoplanatic patch in the visible.

  3. In order to make corrections, you need to observe the reference object in the optical. This is easy to do with a near-IR setup using an optical/IR beamsplitter: send the optical light to the AO equipment and send the near-IR light to the near-IR instrument. In the optical, you use an optical beamsplitter to send half the light to the instrument and the the other half to the AO equipment. This means that the AO equipment only gets half the light it would if it were used with a near-IR instrument, which makes it (even) harder to do the corrections.

Finally, there is an issue unrelated to the AO itself, which is that you need different science instruments depending on whether you're working in the optical or the near-IR. Optical instruments use silicon CCDs for detection; these are only sensitive out to about 0.9-1 microns. Near-IR instruments use different detectors (usually HgCdTe-based), which are good from about 1-3 microns. (Near-IR instrument also need a different design to reduce contamination from thermal emission from the telescope and optics for observations at wavelengths longer than 2 microns.) So in practice the choice has been: combine AO with a near-IR instrument and get good performance with affordable/feasible technology, or combine AO with an optical instrument and get very limited performance with more expensive (or even, until recently, unattainable) technology.

Nonetheless, there are some optical AO systems starting to appear, such as MagAO on the Magellan telescope (which has both an optical instrument and a near-IR instrument, and can correct for both simultaneously).

  • $\begingroup$ Interesting! I'm asking why AO is used down to 0.9um but no further - could you do your numerical comparisons for say 0.9 vs 0.5? Do all these difficulties simply scale roughly linearly with $1/\lambda$ or is there something that's getting much more difficult at a rate much faster than that? Have minimum wavelength for astronomical telescope AO been steadily decreasing as technology and understanding has improved, or has there always been a wall between 0.9um and visible? $\endgroup$
    – uhoh
    May 9, 2016 at 2:24
  • 1
    $\begingroup$ @uhoh I actually obtained observations about 7 years ago, in the R and I bands (600-800 nm), with an AO system called NAOMI on the William Herschel telescope. It didn't get to the diffraction limit, more like 0.2-0.3 arcseconds, but was more-or-less unique at the time. Lucky Imaging is usually viewed as cheaper and more successful at optical wavelengths. $\endgroup$
    – ProfRob
    May 9, 2016 at 6:36
  • 1
    $\begingroup$ @uhoh I think the missing piece in your understanding is 0.9-1 micron is magical, but not because of the AO -- it's because you need different science instruments for the optical vs the near-IR. I've edited my answer to include this point (and another point about additional loss of light in the optical AO case). $\endgroup$ May 9, 2016 at 9:25
  • 1
    $\begingroup$ There are also working optical AO systems used by the US Air Force (and probably some other countries) to spy on satellites. These are on small (1-3 m) telescopes, which means there's less correction required to reach the diffraction limit, and they're looking at extremely (by astronomical standards) bright objects, which probably makes things more feasible. $\endgroup$ May 9, 2016 at 9:27
  • 1
    $\begingroup$ Speaking as a former employee at a company called, you guessed it, "Adaptive Optics Associates," I can confirm pretty much everything in the answer and the comments here. $\endgroup$ Sep 26, 2016 at 13:59

The simple answer for the wavelength part is that performance of AO systems degrades the shorter in wavelength you look. The basics of what happens is as you go to shorter the wavelengths of light, you need a finer plate scale to detect variations in seeing which requires very expensive (and in some cases non-existant) hardware. You also need a higher AO frequency (ability to measure the light and deform/refocus the telescope) to account for the higher frequency of light, this again takes very expensive hardware if it exists at all at the frequency required.

This is because some of the basic calculations (not taking Zernike polynomials into account) are based on the Strehl ratio and Here (ratio of peak intensity of an aberrated image compared to perfect image) to figure out what the intensity of the source should be and the FWHM (Full-Width Half Max - width of the profile of light at half intensity) to essentially measure where the light should be. Both of these measurements are wavelength dependent.

Basic further reading can be found at The Isac Newton Group of Telescopes. Much more in-depth reading can be found at the university of Arizona Optics department.

  • $\begingroup$ Thanks. You've listed a number of things that scale with wavelength, and said that they are harder or move expensive - I can do that too. But which one is the one that is so hard or so expensive that it is a show-stopper? Am I correct that AO is simply not ever done in the visible? How-much harder is it? How-much more expensive? As I mentioned I'm hoping for something quantitative. Considering the amount of science that can't be done because the Hubble is so oversubscribed. Do any of those links have the answer to this question? $\endgroup$
    – uhoh
    May 6, 2016 at 18:44
  • $\begingroup$ There is no good metric for calculating hardness of a calculation so I can't really speak to that. The problem really arises when you are diffraction limited because you aren't able to get the information you need, which happens at shorter wavelengths. Diffraction limit: (1.22 * λ(in cm))/diameter(in cm) $\endgroup$
    – veda905
    May 6, 2016 at 19:24

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .