TTHEHE HHISISTTORORYY ANDAND DDEVELOPMENTEVELOPMENT OFOF THETHE ESOESO AACTIVECTIVE OOPTICSPTICS SSYYSSTEMTEM

THE ACTIVE OPTICS SYSTEM IS THE FUNDAMENTAL OPTICAL CHARACTERISTIC OF THE ESO NEW TECHNOLOGY (NTT) AND (VLT). THE NTT PIONEERED THIS SYSTEM. WITHOUT IT, THE VLT, WITH ITS THIN, VERY FLEXIBLE , COULD NOT GIVE A USABLE OPTICAL IMAGE AT ALL. R. N. WILSON

HE TERM “ACTIVE OPTICS” , normally made as thick and stiff the errors in a star image, although Fou- is normally, and I believe as possible to avoid flexure. The idea of cault's invention of the knife-edge test correctly, associated with the segmenting goes back to Lord Rosse in enabled a very sensitive qualitative as- ESO system developed for 1828, but was first realised in practice by sessment. The detector available, the eye, with monolithic Horn d’Arturo in the 1950s using a fixed was sensitive but highly non-linear; pho- TprimariesT and applied in the ESO NTT primary. Keck I, finished in 1992, was the tography was terribly slow and insensi- and VLT telescopes. Technical systems first such telescope with a normal 2-axis tive, and also non-linear. Third order based on the same principles, also with mounting. So it was logical that earlier aberration theory, due to Seidel, was only

elescopes and Instrumentation thin meniscus primaries, are used in the ideas of active optics should have been published in 1856 and existed only for

T other very large telescopes (8 m) of limited to monoliths. The history of the spherical surfaces, not for a Newton par- GEMINI (2x) and SUBARU (1x). Other development of mirror support systems abolic primary. The complete theory for important telescopes with monolithic pri- and of active optics for monoliths is given telescopes was only published in 1905 by maries, but using stiffer lightweighted in relatively complete form in my second Karl Schwarzschild. Lack of theory was blanks, are also actively controlled, the book, Optics II also the reason Foucault could not have most notable being the WIYN (3.5 m) tel- (Wilson 1999; W99 in the following). The invented the aplanatic (RC) telescope. escope, the three 6.5 m telescopes of the present article is a simplified and much Both Schwarzschild (1905) and Chrétien MMT upgrade and the two Magellan tel- abbreviated version, with more emphasis (1910) used the Abbe “sine condition” escopes, and the two 8.4 m telescopes of on the personalities involved. (Wilson, 1996) as the basis for setting up the Large Binocular Telescope (LBT). A friend of mine at ESO, who comes aplanatic (i.e free not only of spherical The other major branch of modern opti- from the rich French tradition in tele- aberration, but also of field coma) tele- cal telescope development, that using seg- scope optics, recently suggested to me scope forms, unknown to Foucault. No, mented mirrors, was pioneered and ex- that active optics (and the Ritchey-Chré- realistically Foucault, a scientific and emplified by the two Keck 10 m tele- tien (RC) aplanatic telescope) might have technical genius, conceived and adjusted scopes. This has its own system of active been invented by one of my great French his mirror supports to get the best image control of the segmented primary. heroes in optics: Jean Bernard Léon Fou- he could. But this was not active optics: it Although the aim is the same, the tech- cault. Reference is made in this connec- was a procedure which had been used nologies involved in the control systems tion to his largest − 80 cm− Newton tele- throughout the history of the reflecting of monoliths and segmented mirrors are scope completed in 1862 (Wilson 1996). telescope. It had been used empirically by essentially different. This is the case be- With all respect to the great genius Fou- James Short, William Herschel, Lord cause the flexure function of a monolith cault, I believe that neither of these in- Rosse and others before the invention of has no discontinuities, which are funda- ventions would have been possible at that modern support forms by Lassell in 1842 mental to the nature of segmented mir- time. For active optics, no technology ex- (astatic lever) and T. Grubb, also about rors. The ESO active optics system is isted for measuring in a systematic way 1842 (whiffle tree) (see W99). “closed loop” in the sense that correction is made by measurements in real time of the quality of a star image. This is not the case with the Keck telescopes which cor- rect the primary with an internal active system.“Closed loop” in the ESO sense is also not necessarily used in the other ac- tive systems for monolithic primaries: I should like to dedicate this article to GERHARD some of them rely on “precalibration” of SCHWESINGER (b. 08.01.1913 in Krappitz, Upper Silesia, d. the flexure effects (see page 9). 03.11.2001 in Heidenheim, Württemberg: see photo), who In this article, I shall confine myself to developed the first complete Fourier theory for the support active optics with monoliths. When was of primary mirrors of telescopes and thereby also stimulat- the idea of active optics for monoliths ed my thinking on active optics; and to LO WOLTJER whose first conceived? We must remember that vision and support led to the NTT and VLT based on my ac- all “classical” telescopes had monolithic tive optics concept.

2 The Messenger 113 We must conclude that active optics plete system of correction, which later be- wesinger, Zeiss showed no interest in pur- was anyway a concept of the 20th century. came the ESO active optics system. In the suing it. I think an industrial concern, I believe the first ideas of a systematic course of further discussions with Dr however brilliant and engaged in the mat- process came from the great French opti- Gerhard Schwesinger, a brilliant engineer ter, would anyway have been too far re- cian Couder in 1931 and (probably inde- and mathematician who also had excel- moved from practical telescope use. In pendently) from the great Russian opti- lent knowledge of optical aberration the- other words, an observatory concerned cian Maksutov in 1948* (W99). Couder ory, I became aware of his general Fouri- with practical telescope development for recognised the high sensitivity to the er theory of the flexure aberrations intro- a functioning observing site was essential. aberration astigmatism of mirrors which duced by support errors in primary mir- Of course, I had this in mind when I were inadequately supported and sug- rors. I realised that the circular nature of joined ESO. However, I soon learned that gested that such “regular” errors might be the mirror led to polynomial functions ESO had other problems far more urgent: corrected by “a system of forces suitably which were completely equivalent to the successful realisation of the 3.6 m tel- applied”. He concluded that astigmatism those defining optical aberrations, al- escope, on which its reputation and, in- left over from manufacture could be thus though the mathematical boundary con- deed, its future existence depended. This corrected. But as with Foucault, Couder ditions are not the same. A light went on was a conventional telescope following had no means of measuring astigmatism in my head! It would be perfectly possible the line of the “Bowen-class telescopes” in a star image. It could only be done to interpret all the flexure effects in terms (Wilson, 1996) and had no significant in- qualitatively,off-line, in a slow process of the classical optical polynomials of novative features. Nevertheless, at its (effectively dc in modern terms of active Hamilton or Zernike. Again, if one could completion, the 3.6 m telescope gave me optics). A rapid, repetitive, quantitative measure their amount and direction, one the opportunity, during its optical set-up, correction process was technically out of could correct all such flexure terms by ap- alignment and test in 1976, to simulate the the question. Similarly, Maksutov pro- propriate force changes of the supports, whole theoretical basis of an active optics posed the adjustment of Lassell type asta- calibrated from the Schwesinger theory. system. The test system used was “classi- tic support levers to correct such errors Above all, this would apply to the lowest cal Hartmann”.This was a painfully slow observed with an ocular or a Foucault order term, by far the most important as and exhausting process, measuring photo- knife-edge, recognising too that the result Couder had recognised, astigmatism. This graphic Hartmann plates in a semi-auto- was only valid for one zenith angle of the was particularly interesting, as Sch- matic mode. But it enabled our team (es- telescope. Since he had no mathematical wesinger’s calculations showed that sentially Francis Franza, Maurice Le Luy- algorithm for applying the correction, he maintenance of the absolute tolerances er and myself) to do a rigorous aberration suggested trial-and-error. Again, as with for the astigmatism specification was just analysis of the finished telescope.This led Couder, such a procedure was so cumber- as impossible in practice as the mainte- to my definition of the “Intrinsic Quality” some that its application could normally nance of an absolute decentering toler- (IQ) (see Fig. 1) of a telescope as that op- only be a once-off process at initial set-up ance. tical quality which would be achieved in (i.e trial-and-error, qualitative, dc active It followed that, when I formally joined principle if all the correctable terms optics). Understandably, although these ESO in September 1972, I had had the measured could also be corrected in prac- suggestions were highly perceptive, they whole theoretical basis of active optics in tice. In the 3.6 m telescope, there was no never led to any practical results or to se- my head for several years. However, not means of doing this: the primary mirror rious further thinking. The time was still only the central problem of image meas- was too thick and rigid to allow it, even if technically not ripe! urement and analysis remained to be a suitable support system had been avail- In 1968 I was already working, at Carl solved, it was also necessary to convince able. Furthermore, a far simpler,rapid and Zeiss in Oberkochen, indirectly for ESO, other colleagues and astronomers of the on-line image analysis system than that on a study for the optics of the ESO 3.6 m immense possibilities. In spite of the in- used would be essential. telescope.This made me aware of the cen- terest and vital information given by Sch- The essential elements of my active op- tering problem of such Cassegrain tele- scopes. A lateral decentering tolerance of the axis of the secondary mirror to that of the primary, in order to maintain the op- tical specification of this telescope, would have to be set at well under 1 mm. In dis- cussions with my colleagues in mechanics, it was clear that this was virtually impos- sible in practice, bearing in mind ex- change operations of the top-end units. It became clear to me why the aberration produced (decentering coma, a particu- larly unpleasant, asymmetrical degrada- tion of the star images) was the main curse of the Cassegrain telescope in prac- tice: I called it “Cassegrainitis”. It struck me then that, if one could measure its amount and direction, its correction on- line would be a relatively simple mechan- ical operation requiring only a small lat- Figure 1: Results of classical Hartmann tests of the conventional ESO 3.6 m telescope in 1976, eral shift of the secondary to correct the illustrating the theoretical improvement after successive removal of polynomial terms. The mean error. This was the mental start of a com- right-hand point of the functions gives the Intrinsic Quality (IQ) of the telescope. (W99)

*In W99, I expressed my gratitude to D. Enard and K. Bahner for drawing my attention to these pro- posals by Couder and Maksutov, respectively. © ESO - September 2003 3 tics system were first published formally telescope community at the time and not successful completion of the 3.6 m – in- in Wilson (1977). At that time I called the suitable for practical application, these dubitably essential for the future exis- concept a “feedback” telescope. The term studies had some valuable theoretical fea- tence of the organisation. Immediately “active optics” was used in a more explic- tures, notably modal control with so- following the conference, Woltjer asked it publication in Wilson (1982), which es- called “natural modes”.The measurement Richter, Chief Engineer in charge of the sentially gave the whole system of the system proposed was not “closed-loop” in 3.6 m, to coordinate a study of three op- NTT except for the practical details of the the ESO system sense, using a natural star tions: 1 x 16 m, 4 x 8 m, 16 x 4 m.Although image detection and analysis. At the pre- image in real time, but an experimental Richter favoured the 1 x 16 m option, vious verbal presentation in 1981 at an precalibration of deformations of the pri- there was general agreement at ESO in optical conference in Graz, an American mary for given forces and determined by favour of 4 x 8 m. I myself was strongly in in the audience asked me at the end: interferograms. In any event, the astro- favour of this option. A 16 m telescope “Have I understood you correctly, that nomical community in the United States, was technologically too big a step in size, you propose a telescope in which the op- with a few exceptions such as Aden while 16 x 4 m represented no step in size tical system continuously checks itself Meinel, showed little awareness or inter- and was politically too banal not to suffer and optimizes itself fully automatically?” est in the potential of active optics until reduction in the number of telescopes to I replied: “Yes, I congratulate you, you the late 1980s when the ESO NTT pro- save money. However, at that time, there have understood it perfectly”.He replied: duced its first results. Before the NTT de- was no clear idea what optical solution “Well,my feeling is that such a system will velopment at ESO, the same was true in would be available for 8 m unit tele- never be realisable in practice”.I hope he Europe. In the early 1970s, a very cheap scopes, let alone a 16 m. The concept of has since followed developments and reg- low quality 4 m IR spectroscopic tele- the 10 m Keck was just published, but de- istered what has emerged with the NTT scope was built in France by Connes, sign studies were only just starting. The and VLT. Chevillard et al. (1989). This used a pri- Angel technology of lightweighted blanks As is always the case with radically new mary with 36 square segments, similar in (a further development of Palomar) had developments, parallel thinking had been principle to the later Keck 10 m project. not then taken off, and the Multi-Mirror- going on independently by other groups. However, unlike the Keck, it used a Telescope (MMT) was only completed in In 1970, a paper was published by Cree- closed-loop feedback control system 1979 and comprised only small tele- don and Lindgren in the American jour- based on measurements of a natural star scopes.Already in these discussions on an nal “Automatica”,a journal hardly known image, in principle like the ESO system ESO VLT, I had my active optics solution in the astronomical or optical communi- for monoliths. The detector was a circular in mind. But a trial on a smaller-size tele- ties.This work had been commissioned by aperture passing flux to a photo-multipli- scope seemed to me essential. Shortly af- NASA in connection with the 2.4 m Hub- er for each segment. Image analysis in the ter these discussions, the extension of ble Space Telescope (HST) and was re- ESO system sense would hardly have ESO by the membership applications of ported in detail in secret NASA reports been possible with this detection system. Italy and Switzerland provided a marvel- (W99). I only became aware of this work The image quality aimed for was very low, lous new perspective at just the right about 1985 through Oberto Citterio. The 10 arcsec, but the segments were so poor time: the entrance fees could perhaps authors, who were brilliant control engi- that the image was more like 10 – 20 arc- fund a new test telescope to try out new neers but not versed in optical aberration secs.The project was apparently known to technolgy.Woltjer asked Richter to make theory, proposed a very complex mathe- some ESO astronomers, but was not com- a cost estimate for such a 3.5 m New Tech- matical scheme for the active control of municated to the engineers: apparently it nology Telescope (NTT). At that stage, the HST. In a (secret) NASA study of generated no interest. Finally,it was aban- the only “new technology feature” envis- 1973, Howell and Creedon developed this doned in 1975 because of total lack of in- aged was an alt-az mounting, which had approach further, improving it fundamen- terest and support in the French astro- been pioneered by the Russian 6 m tele- tally by proposing a modal approach, also nomical community. scope completed in 1976, but no western a fundamental feature of my own concept project had had the courage to take this as thought out at Zeiss in 1968 with Sch- THE ORIGINS AND DEVELOPMENT fundamental step. This cost estimate en- wesinger, from his own modal theory. OF THE ESO NTT AND VLT abled political support for the NTT to be However, Howell and Creedon's algo- The ESO Technical report No. 8 in 1977, marshalled. rithm was extremely complex and re- concerned with the image analysis of the At this time, I was spending a most in- quired the optimization of the support newly set up 3.6 m telescope, gave the structive year on La Silla, learning about geometry according to the errors meas- first account of the theoretical basis of an the practical (maintenance) problems of ured. In contrast, the ESO system algo- actively controlled telescope with a the many (and varied) telescopes at our rithm can be applied directly to any nor- monolithic mirror with the ESO system. ESO observatory. Everything I saw con- mal passive support, whatever its geome- It was followed in the same year by the vinced me further that active optics was try. Since the Howell and Creedon pro- formal paper (Wilson, 1977) at the ESO the only answer to the problem of main- posal was completely impractical, it was Conference on “Optical Telescopes of the taining optimum optical quality with tele- rejected by NASA. Because the HST pri- Future”.At this conference, Prof. Woltjer scopes in practice. Back in Europe in the mary has a very stiff, lightweighted pri- also gave a paper assessing the great mer- summer of 1980, the big event for ESO mary,the forces required for active optics its, above all for spectroscopy, of larger was the move from Geneva to Garching correction of the initial spherical aberra- ground-based telescopes and thereby jus- in Germany. Since Richter left ESO at tion error,discovered after launch in 1990, tifying astronomically the construction of this point, the position in charge of the Te- would have been unrealisable in practice. a 16 m telescope. He left it open whether lescope Group was free and Woltjer With a thin, relatively flexible primary this should be a single 16 m telescope, or asked me to fill it. I was happy to do this, similar to the NTT, the ESO active optics an array of smaller telescopes. Such for- for my central interest had always been system could have corrected it immedi- ward thinking was legitimate and neces- the telescope optics side, although my fi- ately. Although unknown to the general sary for ESO at that time following the nal position in Geneva had been in charge

4 The Messenger 113 of the Instrumentation Group. For the Te- ed much work and was finally successful led by Daniel Enard. He decided to fol- lescope Group in Garching, although oth- in demonstrating the practicability of the low completely the active optics, thin er tasks were not negligible (e.g. the ESO active optics system. This was also monolithic primary concept of the NTT. building for the 2.2 m telescope of the our first trial of the image analyser based However, the VLT would require a much MPIA or the achromatic plate for the on the Shack-Hartmann principle. I had bolder approach than the (correctly) cau- Schmidt telescope), the NTT was the cen- learned of this invention of Roland Shack tious approach of the NTT regarding the tral project and was exactly the project I (whom I had met at the Imperial College flexibility of the primary. The basic tech- had longed for. I think I can truly claim of London University in the early 1960s) nical decisions on the VLT had therefore that I determined every major character- about 1979, when he was a professor at to be taken before the completion of the istic of this telescope (the alt-az mount the Optical Sciences Center in Tucson. I NTT First Light in March 1989. Essential- was taken over from Richter and was self- visited him with Francis and he was de- ly, the 1 m-experiment enabled those de- evident for a New Technology Telescope) lighted at last to find someone who was cisions to be taken with confidence. – not always to the pleasure of all the as- deeply interested in his system of optical Before the “Astronomical First Light” tronomers. For example, I rejected not quality measurement and who wished to of the NTT we had what I called the only a prime focus but also a Cassegrain apply it immediately. The astronomical “Technical First Light”.This was the first focus, equipping the telescope only with community in Tucson had shown no in- time I saw a star image in the newly erect- two symmetrical Nasmyth foci. This terest at all. Shack gave me an “S-H ed telescope. No adjustments had then meant that complex changes of foci been made: this phase of the op- as in the 3.6 m were totally avoided. eration was now starting and in- With an act of great courage and ex- cluded the so-called “dc phase” of pression of confidence in my active optics, i.e. the fixed , once- knowledge of optics, Woltjer ac- off corrections – see pages 6-8. cepted my active optics concept, but The star image I saw in a hand- imposed one entirely reasonable held eyepiece did not please me. I and prudent condition: that the looked at the defocused image in- NTT should work in the passive side and outside focus, a classical mode (i.e. without active optics) to test procedure I had used for the same specification as the classi- decades on many telescopes. The cal (passive) 3.6 m telescope. This appearance indicated strong condition forced me to increase the spherical aberration, a defect thickness of the NTT primary from which became world famous two the 1:18 ratio I had envisaged to years later, when this aberration, 1:15. This reduction in flexibility re- indicating a ”matching error” in duced the dynamic range of the ac- the forms of the primary and sec- tive correction – see below concern- ondary mirrors, was revealed in ing “First Light”. With the proven the Hubble Space Telescope. I success of the active optics, 1:18 hoped for the best, thinking per- would have been better, but I still haps there was a strong thermal accept Woltjer’s imposed condition effect in the local air of the build- as correct at the time. The optics ing at that time. However, with team was minimal in 1980 and con- The NTT Active Optics Supports the tests of the image that fol- sisted essentially of Francis Franza with Francis Franza at work lowed, my fears were confirmed: and myself with valuable assistance there was indeed a strong spheri- from Bernard Delabre on the optical de- screen”,a raster of lenslets, which he had cal aberration present. We were able to sign side. Quite early on I decided that it made mechanically on a lathe. This had a prove it was an error in the form of the was too risky to go ahead with the final serious phase defect, but was usable and primary. Exhaustive investigation, also active optics system without a smaller enabled us to operate our image analyser. with the manufacturer Carl Zeiss, showed scale experiment. This led to the 1 m-mir- Initially, the detection was with photo- that a spacer error had been made, as with ror experiment in the optics lab, in which graphic plates and, to get rapid results, I the HST, in one of the so-called ”null- Paul Giordano played a fundamental favoured doing the whole 1 m-experi- test” systems. In fact, this error had been part. However, there was a serious lack in ment with plates. But Lothar Noethe detected by a careful check of the system our optics group: a physicist to deal with wished to initiate CCD technology and at Zeiss. However, owing to a misunder- the image analysis side and the necessary worked also with a CCD which had just standing of the sign of the spacer error, software development.This gap was filled become available. The final results were the error was corrected in the wrong di- by the engagement of Lothar Noethe, only published in 1988 (Noethe et al. rection, thereby doubling the resulting who came from Siemens. His application 1988, W99), but already earlier they had spherical aberration instead of eliminat- for the ESO job was one of the greatest given us full confidence to complete the ing it! The amount of spherical aberration pieces of good fortune in our whole active NTT active optics system. The CCD as was about the same as that found later in optics development. The presence of detector was fundamental: there was no the HST. Our active optics system was Francis Franza was the other essential pil- serious alternative for our closed-loop able to correct it completely, although it lar in the NTT development. Francis was system working rapidly in real time. The used up about 80% of the dynamic range engaged in 1973 in Geneva and from that 1 m-experiment was also fundamental for of correction available – see pages 8-9. time on we developed a perfect working the VLT study, initiated by Woltjer in the This correction, saving a very costly re- symbiosis. 1980s after the financial approval of the working of the primary, was a marvellous The 1 m-mirror experiment represent- VLT as an ESO project and technically demonstration of the power of active op-

Wilson R., History of Active Optics © ESO - September 2003 5 Figure 2: CCD a world record at that time for a ground- pictures obtained at based telescope. Richard West identified “First Light” with the the field (only 12 x 12 arcsec because of ESO 3.5 m NTT in March 1989, the small size of the CCD used directly at compared with previ- the Cassegrain focus) and set up a beau- ous records of the tiful comparison (West 1989, W99) which same field (globular is reproduced here (Fig. 2). Upper left cluster ω Centauri). Upper left, ESO 1 m shows the field, suitably magnified, taken Schmidt; upper right, from a plate by the ESO 1 m Schmidt tel- ESO 3.6 m telescope; escope in 1984 under modest seeing con- lower left, ESO NTT ditions (ca. 2 arcsec). Upper right is from raw image; lower a plate, considered excellent by normal right, NTT processed image. See text for standards, by the passive 3.6 m telescope details. with seeing about 1 arcsec. Already with From West, R. (1989). this improvement, the “clumps of cotton wool”representing star images in the Sch- midt plate have vastly improved. At the bottom left is the “First Light” frame with the NTT. The five separate images of the Schmidt became about 15 with the 3.6 m and number almost 100 with the NTT. (The bottom right frame shows further improvement by off-line processing; but this cannot be compared with the other three which are “raw images”). The enor- mous gain in resolution is striking and tics to extend vastly the manufacturing telescope optics.The finite expansion co- well-illustrated by the triple star right of tolerances of correctable aberrations. efficient of aluminium is largely compen- centre. The Schmidt shows no resolution, The “Astronomical First Light” results sated by its excellent thermal conductivi- with the 3.6 m the triple nature can be in- of the NTT were so fantastic that they es- ty and active optics can easily handle ferred without resolution, while the NTT tablished it as optically the best telescope residual expansion effects. Although in- resolves the three components complete- in the world at that time (1989).As is well terest in aluminium has since been ly.But at least as significant as the gain in documented (W99), the conditions were shown, above all in France, no telescope resolution is the gain in light concentra- extraordinarily good both for external of significant size has been equipped with tion per star image, giving a huge increase seeing and dome seeing: we were remark- an aluminium primary since the brilliant in depth penetration for the same expo- ably fortunate. This was also due to an- pioneer work of Mottoni for the Merate sure time. other new technology feature of the NTT, 1.37 m telescope (1969) in Italy (W99). Figure 3 reproduces the frontispiece of taken over and improved, from the MMT. This is unfortunate and demonstrates my second book (W99) and shows the UT This was the building concept, whereby once again the inherent technical conser- No. 2 (Kueyen) of the VLT together with the building rotates with the azimuth vatism of the astronomical community: two beautiful photographs. The photo at movement of the telescope. We improved the refusal to abandon glass corresponds the top right, a three-colour composite of the MMT building concept by removing exactly to the inverse refusal to abandon the Spiral Galaxy NGC 2997, was taken the back wall, thereby allowing ventila- metal (speculum) in the 1860s and to in- with UT No. 1 (Antu) and the FORS 1 in- tion to pass laminarly through the whole troduce chemically silvered glass! This led strument on 5 March, 1999. In the near IR building “slit” for the telescope. This fea- to the disaster of the Melbourne reflector, band, the FWHM of the best star images ture has been very important for the ex- set up in 1869 (Wilson, 1996). was 0.25 arcsec, a record at that time. cellent optical quality and, in somewhat It follows from the above account that modified form, has been taken over for three very important aspects of the NTT SOME BASIC PROPERTIES OF THE the VLT. technology came from the USA: the CCD ESO ACTIVE OPTICS SYSTEM Only one new technology feature detector, the Shack-Hartmann image (a) Automatic optical Maintenance which I envisaged was not realised. This analyser and the building concept. But As indicated in the first section, the ESO was a second primary with an aluminium the active optics concept for thin menis- active optics system was conceived to blank.This was finally abandoned for cost cus monoliths was a purely European de- remedy what seemed to me the most in- and time-scale reasons, but I believe that velopment, which, apart from Roland tractable problem of the “passive” tele- this decision was an error. The NTT Shack and Aden Meinel, was ignored or scopes built up till about 1980: the prob- would have been a perfect telescope to actively rejected in the USA until the lem of optical maintenance of the fin- test the viability of aluminium as a blank “First Light” success in 1989. ished telescope. The optical specifications option. Excellent and reasonable offers The success of the ESO active optics of such telescopes (e.g. the ESO 3.6 m) existed both for the manufacture of the concept has been wonderfully demon- were much inferior to those which have blank and its “Canegen” coating and for strated by the best Full-Width-Half-Max- become normal for “active” telescopes, the optical figuring. The existing blank in imum (FWHM) star images recorded for but they were still good. Furthermore, Zerodur is, of course, excellent: but the the NTT and VLT. With “First Light” in they were largely met by the manufactur- extreme zero expansion property of March 1989, the NTT revealed a best star ers. The problem was that they could Zerodur (or ULE fused quartz) is no image for a CCD frame in the globular rarely be maintained in practical observa- longer necessary for actively controlled cluster Centauri of 0.33 arcsec FWHM – tories. Often, they could be re-established

6 The Messenger 113 Figure 3: ESO VLT Unit Telescope No. 2 photographed in March 2000 by Hans- Hermann Heyer. The photo upper left shows the Crab Nebula taken with UT No. 2 and the FORS2 instrument on 10 November 1999. The photo upper right was taken with UT No. 1 and the FORS1 instrument on 5 March 1999 and shows the Spiral Galaxy NGC 2997. The best star image quality (in the near IR band) had a FWHM of 0.25 arcsec.

by careful adjustment in a complex off- line operation, but most frequently they declined again long before the next such operation. The aim of active optics was therefore to automate the whole optical maintenance procedure. In the NTT, the design aim was to re-optimize the quality automatically with a cycle time of about 10 min.This automation was never re- alised in my time (it was still initiated by hand). The NTT is very robust in its de- sign (see previous section), and this man- ually initiated procedure could also en- sure good quality. However, it depended on procedures being followed and, in the real world, this does not always happen. The VLT telescopes are not at all robust in this sense, because the primaries are about 50 times more flexible than that of the NTT. Without its active optics, the VLT cannot produce a usable image. Ful- ly automatic operation is thus essential, and optimization is performed every 40 s. Therefore, the period over which the op- tical performance can decline is reduced from what used normally to be weeks for a passive telescope to 40 s. Furthermore, the optimization is always complete, re- covering fully the maximum potential of the telescope, whereas the old-fashioned, off-line procedures were rarely fully ef- fective simply because the telescope was inevitably out of commission and there was always great time pressure. Also, the telescope designs were rarely “mainte- nance friendly” for the optics. The opti- mization cycle of the VLT means essen- tially that the optical quality must main- tain itself for the change of zenith dis- tances involved in tracking for 40 s. Originally, the NTT software had no provision for automation of the active op- tics correction because we knew we had to learn from experience how this could best be realised. The initial huge success after First Light was therefore achieved by purely manual operation. By the end of 1990, we had sufficient practical expe- This included, of course, the fully auto- ness and colour are less critical. This was rience to make an attempt at automation mated active optics correction cycle. Re- from the start a problem with the NTT, possible, but organisational changes pre- cently,the VLT software for the active op- for which the availability of sufficiently vented any further advance in practice. tics has been further improved to elimi- bright stars was uncertain. This problem Finally,the wise decision was taken to use nate aberration effects of the air and to was exacerbated by the raster of the the NTT as a test bench for the new VLT make the choice of reference star more Shack-Hartmann detector, which was laid software, which was installed about 1996. flexible (i.e. easier) because the bright- out cautiously with 40 40 sub-aper-

Wilson R., History of Active Optics © ESO - September 2003 7 tures. With more experience, this was re- duced to 20 20 sub-apertures for the VLT, which anyway gives more light be- cause of the larger aperture.Although the S.-H. raster is unchanged, it has been pos- sible (according to current information from Olivier Hainaut) to operate a partial automation of the NTT active optics for astronomical exposure times longer than Figure 4: The bandpasses for active and correction. From original publication by Wilson and Noethe 1989 and W99. ca. 2 − 3 minutes and under stable condi- tions for the residual aberrations. Typical- ly, about 5 – 7 image analyses are aver- quality,the classical definition of “seeing”. quency component can only be actively aged out, giving then a correction every (If the integration time is inadequate, the corrected within the limits of the isopla- 5 – 10 minutes. This procedure is there- image analyser will give completely errat- natic angle. For the external seeing in fore close in cycle time to my original pro- ic and wrong results since random aberra- general, unlimited correction will only be posal. tions of the external seeing are included). possible using artificial, laser-generated For a frequency higher than 1/30 Hz (for reference stars. (b) Frequency bandpasses of inferior seeing at somewhat lower fre- active and adaptive optics quencies), we enter into the adaptive op- (c) Modal control Table 1, taken from W99, shows the basis tics bandpass C, going to beyond 103 Hz, A modal concept (i.e. successive terms of of all this thinking in terms of the “Band- for the external seeing. In this bandpass, some polynomial with increasing powers pass” or frequency of all the sources of we are confronted with the phenomenon of the parameters involved) has always image degradation. The most important of the isoplanatic angle Θ (W99), that an- been normal practice in optical design conclusion from this Table is that all the gle over which the phase of the error in- based on the theory of optical aberra- error sources are dc or of bandpass <10−2 troduced by atmospheric seeing is essen- tions. As explained in the first section, I Hz except (8), (9) and (10) and partly (7). tially constant. Θ is a function of seeing realised from discussions with Dr Sch- This is of central importance, because it is quality,wavelength and the frequency.For wesinger at Carl Zeiss in 1968 that the roughly the frequency limit of normal ac- visible light and an extended frequency flexure terms in his theory of elastic flex- tive optics correction in closed-loop and band, the value of Θ is only a few arcsec ure of circular mirrors could be interpret- implies that two thirds of all the errors and even at the lowest frequencies of ed in a similar way.The whole theoretical listed are amenable to it.The definition of bandpass C amounts to only one or two basis of active optics was already clear to the correction bandpass is an essential arcmin at most. For closed-loop opera- me.A modal basis was thus clear from the feature of any control system. The situa- tion, for which a reference star within the start. It is a fundamental property of phy- tion with my definition is shown in Fig. 4. isoplanatic angle is required, this is a very ics, linked to thermodynamics, that so- The normal active optics bandpass A, serious limitation compared with band- called “higher order terms”, involving as defined for the NTT, goes from dc to pass A for active optics, for which there is higher powers of the polynomial parame- 1/30 Hz. The limit of 1/30 Hz simply cor- no isoplanatic angle limitation at all and a ters (essentially the radius, the thickness responds to the well-known fact that, in reference star at any convenient point in and the azimuthal orientation in cylindri- the presence of good astronomical exter- the field can be used. The bandpass B, cal mirrors) require more energy for their nal seeing, an integration time of 30 s is which I call the extended active optics generation and are more stable than sufficient to “integrate out” the external bandpass, goes from 1/30 Hz to about 10 “lower order terms”.This is embodied in seeing completely, giving a round image Hz. This is particularly important for Er- the principle of St Venant,fundamental to corresponding to the external seeing ror 7 of Table 1, of which the higher fre- this application of elasticity theory. The conclusion is of great impor- Tabllle 1: The ten sources of error giving degradation of image quality in ground-based telescopes, and tance for active optics: low or- their corresponding bandpasses. Diffraction, which is inevitable and continuous, is excluded since (for der terms such as defocus and a given signal wavelength) it cannot be influenced. In space, the three errors dependent on air vanish astigmatism can occur in tele- (W99). scopes readily and vary rapidly and require relatively low SOURCE OF ERROR BANDPASS (HZ) forces to generate them, such as (1) Optical design dc (fixed) the gravity effects due to tele- (2) Optical manufacture dc (fixed) scope movement. Beyond a cer- (3) Theoretical errors of: tain (high) order, conversely, - Mirror supports dc → 10−3 (fixed → minutes) - Structure (focus, centering) 10−3 (minutes) gravity effects produce effects (4) Maintenance errors of the structure which are optically negligible. and mirror supports 10−6 → 10−5 (weeks → days) The corollary is a very simple (5) Thermal distortions basic axiom of active optics: if - Mirrors 10−5 → 10−4 (days → hours) - Structure 10−3 (minutes) forces of the order of the gravi- (6) Mechanical distortion of mirrors (warping) 10−7 (years) ty forces on the supports can (7) Thermal effects of ambient air produce an optical error of sig- (telescope, dome and site “seeing”) 10−4 → 102 (hours → 0.01 s) nificance, then correcting forces −2 1 (8) Mirror deformation from wind gusts 10 → 10 (minutes → 0.1 s) of the same order of magnitude (9) Atmospheric turbulence (external “seeing”) 210−2 → 103+ (50 s → < 10−3 s) (10) Tracking errors 5 → 102 (0.2 s → 10−2 s) can correct it, if we can deter- mine how and where to apply them! Conversely, a higher or-

8 The Messenger 113 der error which cannot be corrected by Figure 5: The first 25 forces of this magnitude will also not be natural vibration modes generated, i.e. it will not be present. In of the VLT primary. The first 16 of these are cor- other words, any error present which is rected by the active op- due to elasticity and gravity can be cor- tics system. The modes rected. Later, I realised that this modal shown were calculated approach is also mathematically essential for a thickness of 200 mm, whereas the final for finding a practical solution. Mathe- thickness of the mirror is matically, it would seem simple and ele- 175 mm (W99). gant to measure the total aberration error at many points of a rectangular raster over the pupil. If calibrations exist for the aberrations produced by known force dis- tributions, then a so-called matrix inver- sion would give a solution reducing the optical error at all the raster points to zero. However, this procedure would in- clude all the aberration orders, including higher order effects which are negligible in practice, but not zero.These terms would only be correctable by impossibly high forces.The result would be a solution matrix with an inaccessible solution: mathematically a solution matrix with enormous eigenvalue ratios. The modal approach avoids all such problems, pro- vided the modes are reasonably deter- mined. In the NTT, because the primary was relatively stiff, seven modes were suf- ficient. In the VLT, with a primary about but can never rival the repeated direct active optics technologies, both for their 50 times more flexible, 16 modes are cor- measurement of the actual aberration in optical function and for their cost. Finally, rected. Fig. 5 shows the nature of the de- the telescope image. Since there is no iso- for the reader interested in a full account formations produced by these modes.The planatic angle problem and there is no of the current status of active optics, the use of these so-called “natural vibration unsolved practical problem of applying best reference is a recent review article by modes” is one of the great contributions image analysis using CCDs in big tele- Lothar Noethe (2002). of Lothar Noethe (Noethe 1991, W99). scopes, my view is that the closed-loop The obvious optical alternative of system with image analysis is the opti- ACKNOWLEDGEMENTS Zernike polynomial modes, orthogonal mum way of performing active optics. I am most grateful to Prof.Woltjer for dis- modes commonly used in optical design, This was my intention from the start of cussions and clarification regarding the is quite feasible. But the dynamic range of my complete theoretical concept of active early years of the NTT and VLT develop- correction for a given range of forces is an optics in 1968, although I knew of no tech- ments, and to Lothar Noethe for techni- optimum for the natural modes, a very nical solution for real-time image analysis cal discussions. Thanks are also due to important advantage. at that time. Olivier Hainaut for information on the current status of the NTT. The editor of (d) Closed-loop operation CONCLUSION The Messenger,Peter Shaver, has also Reference has been made several times As with all technical developments de- been most helpful regarding the content. above to the fact that the ESO active op- parting radically from accepted technolo- My gratitude is due to him for suggesting tics system is a closed-loop system per- gy, it took a long time, 21 years, between the article in the first place. forming corrections at frequent time in- my first theoretical basis of active optics tervals by measuring the errors in a star in 1968 to its final practical confirmation REFERENCES image in real time. The development of with the NTT in 1989. Without the confi- Connes, P., 1989, SPIE Conf. Active Telescope this image measurement system, based on dence and support of Prof.Woltjer for the Systems, F.Roddier ed SPIE vol. 1114, 238 the Shack-Hartmann detection principle, NTT and VLT, who knows whether it Dierickx, P.et al., 2003, SPIE vol. 4840, 151 Noethe, L. et al., 1988, J.Mod.Optics, 35, 1427 was not trivial, but has long been standard would have been tried in practice to this Noethe, L., 1991, J.Mod.Optics, 38, 1043 technology at ESO through the NTT and day? The significance of active optics Noethe, L., 2002,Active optics in modern large VLT. In a VLT unit telescope, up to 1000 seems to me, in hindsight, greater today optical telescopes, Progr.Optics 43, E.Wolf image analyses might be made in a single than in those early years. Together with ed, 1 West, R., 1989, The ESO Messenger, 56, 3 winter night! In the first section, it was in- the segmented technology of the Keck Wilson, R.N., 1977,Procs ESO Conf. Optical Te- dicated that an alternative approach is by telescopes, it enabled the breakthrough of lescopes of the Future, ESO, Geneva, 99 precalibration of aberrations as a func- both the technological quality barrier and Wilson, R.N., 1982, Optica Acta, 29, 985 tion of zenith distance in an alt-az mount- the cost barrier presented roughly by the Wilson, R.N., 1996, Reflecting Telescope Optics ed telescope, and that this approach is 5 m Palomar telescope. Future huge tele- I, Springer-Verl. Berlin & Heidelberg, 2nd corr. repr. 2002 used in some other projects. The ESO scope concepts, such as the 100 m OWL Wilson, R.N., 1999, Reflecting Telescope Optics viewpoint is that precalibration can be a of ESO (Dierickx et al. 2003), would be II, Springer-Verl. Berlin & Heidelberg, corr. reasonable approximation in many cases, inconceivable without these modern repr. 2001(W99)

Wilson R., History of Active Optics © ESO - September 2003 9