1

No. 97 – September 1999

T E L E S C O P E S A N D I N S T R U M E N T A T I O N

News from the VLTM. TARENGHI, ESO

During the last months, the work at ESO’s Very Large Telescope (VLT) project, both at the ParanalObservatory and in Europe, progressed very well. On the following pages is a brief description of therecent achievements and the activities that will take place in the next months.

Figure 1: From the window of a commercial airplane on a flight from Santiago to Antofagasta, G. Wayne Van Citters from NSF took this dra-matic photo of the Atacama desert. In the isolation of the mountain desert the constructions of the Paranal Observatory resemble a “mirage”.

2

Figure 2: UVES. General view from above the KUEYEN Nasmyth platform where the high-resolution UVES spectrograph is being installed. Therotator-adapter is to the left, within the blue-painted support. (This digital photo was obtained on September 2, 1999.)

ANTU (UT1)

It is now half a year since 1st Aprilwhen the first 8.2-m VLT Unit Tele-scope (ANTU) was handed over tothe astronomers and it started to de-liver excellent observational data. Morethan 35,000 scientific frames havebeen obtained so far with FORS andISAAC. Many of the images have aseeing better than 0.4 arcsec. Severalresearch projects for which time wasallocated during the past months have re-ported very good and even spectacu-lar results. In some cases, the datawere fully reduced and the as-tronomers involved are now in the fi-nal stages of preparing the associatedscientific reports and papers. Our scien-tific operation team guided by R. Gil-mozzi gained enormous experience bothin the service and visitor modes. Nowit is time to transfer the telescopesand instruments to the others. A fun-

damental contribution was given bythe Garching front end and back endoperations teams led by D. Silva andB. Leibundgut that interact with thecommunity in the preparation of theobservation and the support of the VLTarchive.

KUEYEN (UT2)

Following "First Light" in March thisyear, the commissioning work for thesecond Unit Telescope successfullyreached a milestone when instrumentscould be integrated at the different

Figure 3: UVES. When the UVES cover is putin place, it protects the sensitive inner partsof the instrument from unwanted light and dust.(This digital photo was obtained on September16, 1999.)

E

3

telescope foci. The KUEYEN Commis-sion Team, led by Jason Spyromilio,tuned the telescope control system insuch a way that the pointing tests areproceeding extremely well. During a se-ries of 145 test stars just completed andcovering the entire operating range of thetelescope, on no occasion was the starout of the patch. From the real perfor-mance measurements, the overall RMSvalue was found to be only 0.85 arcsecand no stars were rejected during thetest. This is an excellent result for anytelescope.

The next major activity, currently un-derway, is the mounting of the large, high-dispersion spectrograph UVES onto oneof the Nasmyth platforms (see Figs. 2 and3). UVES is the second major VLT in-strument to be built at ESO. It is also oneof the heaviest and most complex onesat this astronomical facility. This delicatework includes the installation and ex-ceedingly accurate alignment of the manyoptical components on the heavy baseplate, necessary to ensure the highest op-tical quality during the forthcoming ob-servations. The moment of “UVES FirstLight” with the first test observations ofspectra of celestial objects is expected thelast week of September.

Figure 5: Mirror cell for UT4. The fourth M1 Cell (for YEPUN) in its protective cover arrives atParanal, after a safe passage from Europe. (This digital photo was obtained on August 23, 1999).

Work on the assembly of FORS2 con-tinued at the Integration Laboratory inthe Mirror Maintenance Building (MMB)and the first tests were made. This in-strument will be installed at the Casse-grain focus of KUEYEN in late October1999.

MELIPAL (UT3)

The dummy M1 mirror cell was re-moved in early August from the third UnitTelescope (MELIPAL) and the real MirrorCell (the third “M1 Cell”), although still witha concrete (dummy) mirror, was installedat the telescope. With a dummy M2 Unitin place at the top of the telescopeframe, pointing and tracking tests wereinitiated. The first image with the 20-cm

guide telescope mounted on the centre-piece and used for these tests was ob-tained in late August (see Fig. 4). As wasthe case for the first two Unit Telescopes,the mechanical quality of MELIPAL wasfound to be excellent and the pointing hasnow been tuned to a few arcsec over theentire sky. The tracking is also verygood. Following these tests, the M1 Cellwas again taken down to the MMBat the Base Camp and the dummy con-crete mirror was removed. The M1 Cellwas moved back up the mountain andstored under the MELIPAL structure un-til the (third) 8.2-m Zerodur mirror is in-stalled in mid-November. “First light”with the VLT Test Camera at theCassegrain focus of this telescope isplanned for February 2000.

Figure 4: UT3 first pointing. First image of astar obtained with the CCD camera at the 20-cm guide telescope on MELIPAL, at the be-ginning of pointing and tracking tests.

Figure 6: “Residencia” model. Photo of the architect's model of the ParanalResidencia and Offices. The building is fully integrated into the landscape.Its southern and western facades emerge from the ground, providing aview from the bedrooms, offices and restaurant towards the Pacific Ocean.

Figure 7: Start of construction. View of the building site where thefirst part of the new Paranal Residencia and Offices are now underconstruction.

4

Performance of the VLT Mirror Coating UnitE. ETTLINGER (LINDE), P. GIORDANO and M. SCHNEERMANN (ESO)

In August 1995 ESO signed a contractwith LINDE AG to supply the coating unitfor the mirrors of the Very LargeTelescope. The coating unit was firsterected and tested in Germany and thendisassembled, packed and shipped toChile. After integration into the MainMaintenance Building on Paranal, the firstmirror was coated on May 20, 1998("First Light" on May 25–26, 1998). Thefinal runs for provisional acceptance wereperformed in January 1999.

The Coating Unit

The Coating Unit encloses the follow-ing main components (see Fig. 1):

• the vacuum chamber with a diame-ter of more than 9 m and an inner volumeof 122 m3

• the roughing pump system and thehigh vacuum pumping system, consistingof 8 cryo pumps and a Meissner trap

• the mirror support system, a whiffletree structure with lateral and axial padsto support the glass mirror during coat-ing

• the mirror rotary device including a fer-rofluidic vacuum feedthrough, to rotate themirror underneath the magnetron duringcoating

• the coating cart, enclosing an aircushion system to drive the lower cham-ber section and a lifting device, to close

the vacuum chamber after mirror load-ing

• the magnetron sputter source with awater-cooled shutter system and cryo-genic shields

• the glow discharge cleaning device,to heat up and clean the mirror surfaceprior to coating.

Thin Film Deposition Equipment

The main component for the coatingprocess is the thin film deposition equip-ment of the Coating Unit comprising thefollowing components:

• sputter source for aluminium includ-ing power supply

• shields to trim the aluminium coatingdeposited

• shutter panels • cryo panels attached to the shutters• Glow Discharge Cleaning Device

(GDCD) including power suppliesThe DC Planar Magnetron Source

consists of the target 99.995% pure alu-minium cathode bonded to a water-cooledbacking plate to reduce the heat radiat-ed to the mirror. The discharge is pro-duced by the use of an inert gas (Argon)to support the flow of current betweencathode and anode. The planar mag-netron source uses magnetic fields to fo-cus electrons in the region of the sput-tering target.

Stainless steel trim shields, which areplaced below the cathode to trim the de-position of the sputtered aluminium, en-sure that the mirror is evenly coated witha uniform thickness of aluminium.

The shutter panels are stainless steelsheet box constructions, which are cooledby water, fed through the panels underpressure. There are two shutters, one toform the leading edge of the coating, andone to form the trailing edge. Both shut-ters are pivoted about the centre of themirror, and at the beginning of a coatingrun both shutters are closed togetheralong the line of the joint band.

Below the shutter panels, copper cryopanels are suspended, filled with liquid ni-trogen.

Their purpose is to provide an area ofhigh purity and homogeneity between theshutter opening below the magnetron andhence aid and improve the quality of thereflectance of the aluminium depositedonto the mirror.

The GDCD consists of two aluminiumelectrodes, shaped to give the requiredprofile. The glow discharge electrodesare water-cooled and are suspendedfrom a dark space shield, which is alsomade of aluminium. The purpose ofthe GDCD is to reduce adsorbed watermolecules from the mirror, the inner sur-faces of the chamber and all compo-nents mounted inside the chamber. The

YEPUN (UT4)

The fourth Unit Telescope (YEPUN) isnow in the final phase of mechanical as-sembly. Acceptance tests with the sup-plier will start soon, after which the con-struction of the four telescope structureswill have been finished. The fourth M1Cell arrived at Paranal in late August(Fig. 5). "First Light" for YEPUN is sched-uled for July 2000.

VLTI

In Europe, major achievements oc-curred in the area of the VLT Inter-ferometer (VLTI) construction. The twosiderostats are being tested on the cloudyEuropean sky (see article by F. Derie etal. in this issue). ESO concluded a con-tract with REOSC (France) for delivery ofall mirrors to equip the four VLT UnitTelescopes with coudé foci optics. An im-portant milestone was reached with thecompletion of the polishing of the opticsfor the first delay line for the VLTI atREOSC. The integration of the cat's eyewill be done at TNO (Netherlands) andthis part will then be integrated into thedelay line at Fokker Aerospace (The

Netherlands). Shipment to Chile will takeplace immediately afterwards. In addition,the 1.8-m primary mirror for the first VLTIAuxiliary Telescope has reached the fi-nal stage of the light-weighting processand the optical polishing will soon beginat AMOS (Belgium). The already manu-factured optics (null correctors, M2 ma-trix) have shown excellent optical quali-ty results. A contract for the purchasingof the Auxiliary Telescope number 3 wasconcluded with AMOS.

The “Residencia”

It has always been ESO’s intention toprovide better quarters and living facilitiesat Paranal. Plans for the design and con-struction of the “Facilities for Offices,Board and Lodging”, also known as theParanal Residencia and Offices, weretherefore begun some years ago. Theproject was developed by Auer andWeber Freie Architekten from Munich(Germany). Their design was selectedamong eight proposals from Europeanand Chilean architects.

The unusual conceptual design isstructured as an L-shaped building, lo-cated downhill from the Paranal access

road on the southwest side of the presentBase Camp. The building is fully inte-grated into the landscape, essentially un-derground with its southern and westernfacades emerging from the ground, pro-viding a view from the bedrooms, officesand restaurant towards the Pacific Ocean.The common facilities, restaurant, offices,library, reception and meeting rooms arearticulated around the corner of the build-ing, while the hotel rooms are distributedalong the wings of the L-shape. A circu-lar hall, 35 m wide and four floors deep– covered with a dome emerging atground level – is the building's focalpoint with natural light. At the floor of thehall there is an oasis of vegetation witha swimming pool. Special protectionagainst light pollution is foreseen. To alarge extent, the ventilation is with naturalairflow through remotely controlled air inoutlets.

The Chilean firm Vial y Vives won theconstruction tender. Work on the first partof the building began on July 1, 1999, andis scheduled to be completed within twoyears. The excavation work for theParanal Residencia is proceeding ac-cording to schedule, with the first concreteto be cast shortly.

5

glow discharge also heats the mirrorsurface, which improves film adhesionand film purity.

Process Description

The method selected to deposit a thinaluminium film on the mirror is the sput-tering process.

The sputter coating process is done byinserting a mirror into a vacuum chamber,which contains a specially designed cath-

ode and an inert process gas. A negativevoltage is applied to the cathode (target)and a glow discharge (plasma) igniteswithin the vacuum chamber when the ap-propriate environment (or conditions) areachieved.

At this point, positively charged atomsof the gas (ions) are attracted to thesurface of the target which is nega-tively charged. The positive atoms strikethe negatively charged target with suchforce that atoms of the target are eject-ed and deposited on the mirror surface,

building up a thin layer atom by atom.What differentiates a magnetron cath-

ode (used here) from a conventionaldiode cathode is the presence of a mag-netic field which confines electrons in theregion of the sputtering target. A magneticfield is supplied from below the cathodetarget, which is arranged so that free elec-trons are captured in the crossed mag-netic and electric fields in the region. Thecaptured electrons enhance the ionisationin the inert gas atmosphere. Overall, theeffect is to lower the operating voltage re-quired, from several kilovolts to less than1000 volts. The most striking gain of themagnetron design, however, is an order-of-magnitude increase in coating depo-sition rates per unit area of target.

Coating Procedure

The mirror is loaded into the lower halfof the coating unit. The two halves of thevacuum chamber are brought together,sealing the coating unit. Then the cham-ber is pumped down to achieve the re-quired vacuum level.

The mirror is rotated beneath the sput-ter source and GDCD so that it can firstreceive a surface treatment and then itscoating of Aluminium.

After the surface treatment via theGDCD has been completed the mag-netron is powered up to pre-sputter untilthe cathode’s surface is clean.

The mirror is rotated at a pre-deter-mined speed calculated to give the in-tended coating thickness at the intended

Figure 1: Main components of the VLT Coating Unit. Cart chassis and drive (1), Lateral pads (2), Mirror rotary device (3), Mirror support system(4), Axial pads (5), Supporting frame (6), Magnetron, shield and shutter (7), M1 mirror (8), Vacuum vessel (9).

Figure 2: The lower part of the coating unit vacuum chamber. For loading and unloading of themirrors, the lower half of the chamber is moved on 8 air-cushions from the coating unit locationto the mirror handling tool and vice versa.

6

sputtering rate. The leading edge shutteris opened as the intended position of thejoint band passes underneath. Thereby,the rotational speed of the shutter exactlymatches that of the mirror.

After the mirror has completed one rev-olution, the trailing edge shutter is closedby rotating it at the same rotational speedas the mirror, as the film joint passes un-derneath. This produces sharply steppededges to the coating, the trailing edge ly-ing on top of the leading edge. Shadowingof the coating by the shutter edges pre-vents them from being perfect steps.

Because both shutters are pivotedabout the centre of the mirror, all pointsalong the joint band take the same timeto cross the mirror opening.

Figure 4 shows an overview of thepressure conditions inside the coatingchamber during the glow discharge clean-ing treatment of the mirror and the sput-ter process. Prior to glow dischargecleaning, the chamber is pumped downto less than 0.001 Pa. Then air is fed inand at a pressure of 2 Pa (cryo pumpsthrottled) the GDC process is started.After the mirror has completed one rev-olution, the glow discharge cleaning is fin-ished and the chamber is pumped downto less than 0.0002 Pa within less than10 minutes, which demonstrates the highpumping speed of the high-vacuumpumping system. Finally, argon is fed inand sputtering is started at an argon pres-sure of 0.08 Pa.

Test Procedure

During the tests of the coating unit theM1 mirror was replaced by the dummysubstrate, a test platform which repre-sents the surface of the M1 mirror. It al-lowed the mounting of glass slides (50 ×

50 mm, 1 mm thick) onto which the alu-minium coating was deposited.

Up to 60 glass slides were mountedonto the dummy substrate at intervalsalong the M1 radius. The position of theslides was recorded on the back of theslides, together with the coating run num-ber.

During a long experimental phase,which covered more than 30 test runs, thefollowing parameters were optimised:

• pressure of the argon process gas• clearance between aluminium target

and mirror surface• adjustment of shutter for an ex-

tremely small joint line• rotation speed of mirror rotary device.The influence of the glow discharge

cleaning device was investigated with re-

spect to the film adhesion and reflectivi-ty of the samples. In addition, the tem-perature raise on the sample surface wasmeasured during glow discharge clean-ing treatment and coating.

The optimisation goal was:• a film thickness of 80 nm with a thick-

ness uniformity of 5%• a sputter rate of 5 nm/s at 80% of the

mirror surface• the best possible reflectance of the

aluminium film in the wavelength rangebetween 300 and 3000 nm.

Aluminium Film Reflectance

After coating in the VLT CoatingChamber, the reflectance of the sampleswas investigated in a reflectometer withan accuracy of about 0.1%. The re-flectance measurement was absolute, notrelative to a standard material whose re-flectance must be known.

The reflectance data of the sampleswere compared (see Fig. 5) with two re-flectance standards:

• The reflectance standard used in theUSA: “Standard Reference Material 2003,Series E”, issued by the National Instituteof Standards and Technology (NIST).

• “Infrared Reflectance of AluminiumEvaporated in Ultra-High Vacuum”, pub-lished by Bennet et al. in J. Opt. Soc. Am.53, 1089 (1963).

Figure 5 shows the absolute re-flectance versus the wavelength of thecoated samples (solid line) in the ultra-violet, visible and infrared region. The cir-cles (filled and not filled) represent the ex-perimental points measured by Bennet(fresh and aged samples), while the as-terisks represent the reflectance accord-ing to the US NIST standard.

Over the whole range from 300 to2500 nm the results obtained in the VLTCoating Chamber are between the agedand not-aged Bennet data and are clear-ly above the NIST data.

Figure 3: A look into the coating unit vacuum chamber. The upper vacuum chamber part car-ries the DC planar magnetron sputter source. In the lower part the rotatable whiffle tree for sup-port of the primary mirror is visible.

Figure 4: Chamber pressure during GDC- and coating process. After the GDC treatment, thehigh vacuum pump system pumps down the pressure inside the coating chamber (volume 122m3) from 2 Pa to 0.00015 Pa in less than 10 minutes.

7

Considering the size of the coating unit(chamber diameter more than 9 m, innervolume about 122 m3) and the size ofthe mirror surface (more than 50 m2) tobe coated, the reflectance data measuredare excellent.

Bennet’s “ideal” data were obtainedin a bakeable, grease-free, all-glass,small laboratory system. To reduce thepossibility of oxidation of the aluminiumduring deposition, the evaporation wascarried out in an ultra-high vacuum ofabout 10–8 Pa (vacuum inside VLT coat-ing chamber 10–4 Pa prior to coating).During the measurement of the re-flectance inside a reflectometer, theBennet samples were flushed with drynitrogen to minimise the rate ofchemisorption of oxygen by the alu-minium.

Due to these optimum test condi-tions Bennet’s results are in very goodagreement with the theoretically predict-ed reflectance data for highly pure(99.999%) aluminium and can be con-sidered as the best possible experimen-tal results.

Aluminium Film Thickness

Beside the reflectance, the film thick-ness of the sputtered aluminium was in-vestigated.

With a special tool a line of coating wasremoved from the sample. The thicknessof the step at the edge of this line wasmeasured in at least three places with aprofilometer.

Figure 6 shows the film thickness uni-formity along the radius of the mirror(measured at single glass samples). Ata mean film thickness of 82.48 nm thestandard deviation was 4.16% (3.43 nm).This high film thickness uniformity over aradius of more than 4 metre shows theperfect adjustment of the trim shields andthe shutter.

Depending on the magnetron dis-charge power, a dynamic deposition rateof more than 5 nm per second could bereached.

This rate is the film thickness dividedby the time taken for a point on the mir-ror to cross the shutter opening. If themeasured film thickness is δ nm and themirror rotational speed is ω degreesper second, and the shutter opening is θ,then

Rate = (δ*ω)/θ

Film Joint Zone

Special attention was given to thearea where the film coverage was fin-ished. The azimuth position of the joint lineand the inclination angle with respect tothe radius vector were exactly adjusted.The defined shutter control and the shad-owing of the coating by the shutter edgesresulted in a joint line zone, which couldno longer be detected by variations in thefilm thickness uniformity. The only indi-cation was a slight decrease in the re-flectance (see Fig. 7) in the ultraviolet andvisible region which could be measuredin a band of less than 20 mm width. Thereflectance in the infrared spectrum wasnot noticeably influenced.

Temperature Increase During Coating

In order to measure the temperatureraise of the mirror during coating, ther-mocouples were glued to the bottom

Figure 5: Measured reflectance values of coated samples (5 × 5 cm) as function of the wave-length in the range from 300 to 2500 nm (ultraviolet, visible, infrared). The results of the VLTCoating Unit are compared with the US standard NIST data and best possible measurementsin a small laboratory system (Bennet et al. in J. Opt. Soc. Am. 53, 1089 (1963)).

Figure 6: Aluminium film thickness, measured on glass samples along the radius of the mirror.Film thickness mean value 82.48 nm, standard deviation 4.16 % (3.43 nm).

Figure 7: Decrease of reflectance in the ultraviolet and visible region of wavelength inside thejoint line of the mirror at a radius of 2.1 m. The dotted lines show the beginning and the centreof the joint line zone, which had a width of less than 20 mm.

8

side of glass samples. Then they were po-sitioned at 7 different radii and fixedthermally insulated onto the platform ofthe dummy substrate.

During the passages underneath themagnetron, the temperature increase onthe samples could be determined at thesame time when the aluminium film grewup.

The result was a temperature raiseof less than 10°C on the surface. Thesamples at the outer radii warmed upmore due to their larger clearance tothe liquid nitrogen cooled shields under-neath the shutter and the larger shutteropening.

Summary

The sputtered aluminium film on the 8-metre-class mirrors produced in the VLTCoating Unit is very uniform along thewhole radius (standard deviation 4.16%).The reflectance in the ultraviolet, visibleand infrared regions is near to that whichideally can be expected. Furthermore, thedynamic deposition rate, defined as thefilm thickness cumulated in a single passof the mirror past the source opening, is

higher than 5 nm per second. The shut-ter system produces sharply steppededges at the start- and end line of the

coating and minimises the joint line to awidth of less than 20 mm. All results arewithin or above ESO specification.

Figure 8: Temperature increase of the mirror surface during coating, measured by thermocou-ples positioned at 7 different radii and rotated underneath the magnetron. The magnetron dis-charge power was 102 kW, the rotation speed 20 degree per minute.

Climate Variability and Ground-Based Astronomy:The VLT Site Fights Against La NiñaM. SARAZIN and J. NAVARRETE, ESO

1. Introduction

Hopes were raised two years ago(The Messenger 89, September 1997) tobe soon able to forecast observing con-ditions and thus efficiently adapt thescheduling of the observing blocks ac-cordingly. Important steps toward this goalwere achieved recently with the start ofan operational service forecasting precip-itable water vapour and cirrus (high alti-tude) cloud cover for both La Silla andParanal observatories1. This was madepossible thanks to the collaboration of theExecutive Board of the European Centrefor Medium Range Weather Forecasts(ECMWF) who accepted to distributetwice a day to ESO the Northern Chileoutput of the global model. The ground-level conditions2 are also extracted fromthe ECMWF datasets with immediatebenefits for observatory operation: badweather warnings could be issued 72hours in advance during the July snow-falls at La Silla. Once embedded into theObservatory control system, the tempera-ture forecast is due to feed the control loopof the VLT enclosure air conditioning.

These forecast systems still have to beoptimised and equipped with a properuser interface tailored to astronomer’s ex-pectancies. They might then join estab-lished celebrities like the DMD DatabaseAmbient Server 3, paving the way tomodernity for ground-based astronomi-cal observing.

2. Unexpected, Improbableand Yet, Real

In this brave new world, however, noteverything is perfect: as if the task of build-ing a detailed knowledge of our siteswas not hard enough, climatic eventssuch as the El Niño–La Niña Oscillation(The Messenger No. 90, December 1997)can turn decade long databases into poor-ly representative samples. As was re-ported with UT1 Science Verification(The Messenger No. 93, September1998), the VLT site started to behaveanomalously in August 1998. A re-analysis of the long-term meteorolog-ical database pointed out an ever-in-creasing frequency of occurrence ofbad seeing associated to a formerlyquasi-inexistent NE wind direction. This

contrasting behaviour is illustrated inthe examples (Figs. 1 to 4) as displayedby the Database Ambient Server. Whilegood seeing occurs under undisturbedNNW flow from above the Pacific, thisNE wind is coming from land, it israpid (Fig. 5), turbulent, and up to 2degrees warmer than ambient. Findingits exact origin became a challengingtask.

3. Searching for Clues

It is commonly stated that obser-vatories are worse when operationstarts than they appeared during the pro-ceeding site survey. This is the well-known 3σ effect, the most outstand-ing site being probably at the top ofits climatic cycle when it is chosen. Thistheory did not apply to the VLT site forwhich we had records since 1983 show-ing an impressive climatic stability.Therefore, before accusing The Weather,a complete investigation was undertak-en, starting with man-made causes. Inparticular the chronology of the stress-es imposed to the site since the con-struction of the VLT had started (e.g.,water sewage, power generation, heat ex-changing) were reviewed in collabo-

1http://www.eso.org/gen-fac/pubs/astclim/fore-cast/meteo/ERASMUS/

2http://www.eso.org/gen-fac/pubs/astclim/fore-cast/meteo/verification/ 3http://archive.eso.org/asm/ambient-server/

9

ration with the VLT staff at Paranal. This,completed by an infrared survey of thesurroundings (Fig. 6), confirmed thatman-made heat pollution was orders ofmagnitude smaller than the phenomenonwe were witnessing.

Measurements of the turbulence in theground layer of the telescope area werethen conducted by the University of Nice4.The study results delivered in spring of1999 revealed that the first 30 m aboveground accounted for only 10% of the to-tal. The turbulence borne by the NEwind was probably extending over onethousand metres or more up in the at-mosphere and thus could not have beengenerated locally.

4. An Explanation in View

The inquiry turned then towards syn-optic air motion but no obvious correla-tion could be observed between condi-tions at Paranal and the output of globalmodels. The answer had to lie justin-between and the picture was slowly tak-ing shape: if apparently stable conditionlike the good NNW Paranal wind werebased on a fragile equilibrium of energybalance, a slight change in the global cir-culation could indeed induce notice-able modifications of the wind pattern incontinental areas, locally tuned by themesoscale orography. But, did such achange really occur?

It was then natural to ask the Chi-lean meteorologists who kindly andrapidly delivered the report reproducedherewith. This report stating that a weak-ening of the westerlies observed dur-ing the recent La Niña period consti-tutes the most plausible explanation to

date. Indeed when compared to the laststrong Niño event (1982–1984), the cur-rent slowly terminating oscillation hasbeen fairly stronger in its Niña phase(Fig. 7), which coincides with the periodof bad seeing at Paranal (July 1998 toApril 1999).

Figure 1: Seeing under standardParanal NNW wind direction.

Figure 4: Infrequent Paranal NEwind direction.

Figure 2: Standard Paranal NNWwind direction.

Figure 3: Seeing under infrequentParanal NE wind direction(data aretruncated for storage purposes.

Figure 5: High wind velocities aretypical with infrequent ParanalNE wind direction.

Figure 7: Comparison of the two strongest Niño events. A negative index corresponds to warmerwaters (El Niño), a positive index to cooler (La Niña). The 1998–99 Niña period is stronger thanthe Niña of the previous cycle (1982–84) (http://www.vision.net.au/~daly/elnino.htm).

4Report on the GSM Measurement Campaign atthe Paranal Observatory, Nov27–Dec20, 1998; F.Martin et al., Ref: VLT-TRE-UNI-17440-0006

Figure 6: Thermal infrared night image of the VLT basecamp with the hot (white) powerplantplume in the centre. These heat sources are located several kilometres to the ESE on the leeside of the mountain. The thermally neutral access road is visible on the foreground.

T h e r m a l E m i s s i o n A n a l y s i sB a s e C a m p P o w e r P l a n t

General View• 19 Feb. 1999• 3 h 31 Local Time• Wind summit:

WNW, 3 m/s• Wind Base Camp:

W. 1 m/s

10

Analysis of the Anomalous Atmospheric Circulation in Northern Chile During 1998C. CASTILLO F. and L. SERRANO G., Department of Climatology of the Direction of Meteorology of Chile

In the far north of our country, wind components in the low and middle at-mosphere until 700 hPa are normally from the North and from the East (Fig. 1),while above them and up to 100 hPa, West winds are normally prevailing dur-ing the whole year (Fig. 2).

It has been noticed that, during the years with La Niña or Antiniño occurrence,changes in the at-mospheric circula-tion are of a globalscale, similar towhat is recordedduring Niño years.

In Niña periods(as is the casesince early 1998),the subtropical an-ticyclone of theSouth East Pacificwhich dominatesthe circulation inour country, be-comes more in-tense and moves afew degrees south-wards, resulting ina weakening of thehigh altitude (west-erlies) winds overthe far north of thecountry. To thisadds the fact thatthe circulation onthe eastern side ofthe Andes, which brings air from the Atlantic, becomes more intense from thelow altitudes to the high atmosphere, up to the point at which the East compo-nent of the wind starts prevailing at high altitudes. This circulation pattern maybe persistent enough to be seen on the Antofagasta radio soundings and is, aspreviously stated, directly related to La Niña events.

Figure 1: Circulation in 850 hPa.

Figure 2: Circulation in 200 hPa.

Figure 3: Circulation in 500 hPa (January 1999).

5. And What Comes Next?

Can we readily assume, looking atFigure 8 that Paranal is on its way to re-cover its original excellence? Hopefullyyes, because although the average see-ing still reflects some fairly bad events,the best 5 percentile were back into nor-mal over the last two months, just as LaNiña was vanishing . . .

6. Acknowledgements

We would like to thank Myrna Ara-neda Fuentes, Associate Director ofthe Department of Climatology of theDirection of Meteorology of Chile whoprovided the report reproduced here-

with and José Vergara of the Departmentof Geophysics of the University of Chile

for his helpful advice about mesoscale cir-culation.

Figure 8: Seeing statistics at Paranal since UT1first light: monthly average (red), median(black) and 5th percentile (green). The dashedlines give the respective long-term site char-acteristics. Seeing is reconstructed from DIMMmeasurements for an equivalent 20-min. ex-posure at 0.5 µm and at zenith.

E

11

within a range from 84 mm–1 to 2800mm–1 for the various configurations of8-m Unit Telescopes and AuxiliaryTelescopes.

Realisation of the Mirror

The Optical Laboratory of MarseilleObservatory (Laboratoire d’Optique del’Observatoire de Marseille, LOOM) was

selected for the design and realisation ofthis special device, because of its ex-pertise in the active optics domain.

The location of the VCM in the delay-line system, on the piezo-translator usedfor small OPD compensation, led to min-imise its weight and to realise a very smallactive mirror with a 16-mm diameter. Therange of radii of curvature, with such asmall optical aperture, corresponds to ƒratio from plane to ƒ/2.5, and the maxi-mal central flexion achieved is 380 mi-crons.

In the VCM system, presented with itsmounting in Figure 1, the curvature isobtained with a uniform loading appliedon the rear side of the mirror and pro-duced by a pressure chamber. In orderto use easily achievable pressure (<10bar), the active part of the system is a verythin meniscus with a 300-micron centralthickness.

Due to the large bending of the mir-ror, the full domain of curvature is notachievable with a classical optical ma-terial as vitroceramic glass (Zerodur).Metal alloys having 100 times higher flexi-bility than glasses, a stainless steel sub-strate (AISI 420) was chosen for the re-alisation.

Optical performances of the VCM

The VCM optical surface quality re-quirements were really tight for such anactive mirror.

A careful analysis of the meniscus de-formation, using theory of large amplitudeelastic deformations, allowed to fulfil withsuccess these requirements as present-ed in Table 1 and Figure 2.

Variable Curvature MirrorsM. FERRARI, F. DERIE, ESO

Since the very beginning of the VLTIproject, it has been foreseen that obser-vations with this unique interferometershould not be limited to on-axis objectsand the possibility to have a “wide” fieldof view (2 arcsec) was planned. It turnedout that the “wide” field of view goal couldnot be achieved without a precise pupilmanagement.

A variable-curvature mirror system(VCM), tertiary mirror of the delay linecat’s eye, has been developed for thispupil management purpose. While theVLTI is tracking an astronomical object,the delay line provides for the equalisa-tion of the optical path of the individualtelescope by varying positions in the in-terferometric tunnel, and the VCM vari-able focal length permits positioning of thepupil image at a precise location after thedelay line.

The range of radii of curvature requiredfor the VCM is determined by the posi-tion of the delay lines. The two major pa-rameters for the determination of thisrange are (1) the distance between theposition of the pupil image following thecoudé optics and the entrance of the cat’seye and (2) the distance between the exitof the cat’s eye and the required positionof the pupil image in the recombinationlaboratory. Considering the field of viewplanned for the VLTI (2 arcsec) and theOPD (optical path difference) to com-pensate for, the cat’s eye secondary cur-vature must be continuously variable

• For radius 2800 mm:

Required performance: λ/4 PTV over ∅ 5 mmDesign Goal: λ/10 RMS over ∅ 14 mm

Achieved performances: λ/6.5 PTV over ∅ 5 mmλ/10.6 RMS over ∅ 14 mm

• For radii in the 230–2800 mm range:

Required performance: λ/2 PTV over ∅ 5 mmDesign Goal: λ/2 PTV over ∅ 14 mm

Achieved performances: λ/2.8 PTV over ∅ 5 mm

• For radii in the 84–230 mm range:

Required performance: 3λ/4 PTV over ∅ 5 mmDesign Goal: λ/2 PTV over ∅ 14 mm

Achieved performances: λ/2.5 PTV over ∅ 5 mm

Table 1: Required and achieved optical surface quality for the VCM.

Figure 1: Variable Curvature Mirror mounted and unmounted with its support. The active partis a very thin (300mm) meniscus and the deformation is achieved with a uniform air pressureapplied to the back side of the mirror. The optical aperture is 16 mm.

12

Figure 2 presents the optical quality(PTV) achieved with one of the VCM mir-rors on the whole curvature range, com-pared to the required performances.

On the central part of the VCM, 5 mmdiameter, analysis shows that the devi-ation from a sphere remains below λ/2

ptv during the whole curvature range(2800–84 mm).

Curvature Control Accuracy

The other important point, related tothe pupil positioning in the interferomet-

ric laboratory, is the adjustment of a pre-cise curvature during the operation of theVCMs.

In order to achieve a precise curvature,the air pressure on the back side of themirror is controlled on open loop with a5.10–4 accuracy in the 0–10 bar range.This high accuracy on the pressure con-trol has been combined with an “hys-teresis” model for the meniscus defor-mation. The effect is present only duringthe phase of decreasing pressure and de-pends on the maximum pressure reachedduring the cycle.

The model allows to take into accountthe history of the mirror and computesthe right pressure to achieve the re-quired curvature. An output of the mod-el is presented in Figure 3 where ∆P isthe correction to be applied (to correct forthe hysteresis effect) as a function ofthe maximum pressure reached duringthe cycle.

Using the high-accuracy pressure con-trol and the hysteresis model, the result-ing error on the VCM curvature is less than5.10–3 m–1 and leads to a 15-cm pupil po-sition accuracy in the interferometric lab-oratory.

The error on the pupil position will bereduced by the beam compressor system,located at the entrance of the laboratory,to less than 1 cm at the instruments en-trance and this value has to be comparedwith the 70 metres stroke of the delay linecarriage.

Conclusion

This is not the first time that such anoptical device, a varifocal mirror, has beenachieved. But this one, thanks to theexpertise of the team in Marseille Obser-vatory, has an exceptionally wide rangeof curvature combined with a very goodoptical quality and a high curvature ac-curacy. This will allow the VLTI to deliv-er a good entrance pupil to the interfero-metric instruments.

Today this VCM has been deliveredto ESO and is being included in thedelay lines in order to set up and cal-ibrate the interfaces with the other devicesof the interferometric mode. The integra-tion will be done during this year, thefirst complete systems (delay lines withVCMs) will be installed on Paranal dur-ing the year 2000.

Figure 2: VCM optical surface quality over a 5-mm diameter. The wavefront error (WFE) is giv-en as a peak-to-valley (PTV) in fraction of λ.

Figure 3: Correction to be applied to correct for the hysteresis effect. The lines represent thehysteresis model and the dots the measured values for various maximal pressures reached dur-ing a cycle.

The VLTI Test Siderostats Are Ready for First LightF. DERIE, E. BRUNETTO and M. FERRARI, ESO

To allow the technical commissioningof the VLTI in its early phases, without theVLT Unit Telescopes or the AuxiliaryTelescopes, two test siderostats havebeen developed to simulate the mainfunctions and interfaces of these tele-scopes.

The principal objective of the opticalconcept (Fig. 1) of the siderostat is to op-timise the collecting area over a sky cov-erage defined by the scientific targets cho-sen for the commissioning of the VLTI withthe test instrument VINCI and the IR in-strument MIDI. The diameter of the 400-

mm free aperture of the Alt-Azimuthsiderostat has been chosen to allow theobservation in the N Band of all MIDI tar-gets (Table 1). The angles between thedifferent mirrors are optimised for the lat-itude of Paranal. To simulate the 80-mmdiameter pupil size of a Unit Telescope,

13

a beam compressor(5:1) is introduced be-tween the Siderostatmirrors. The develop-ment of the beam com-pressor is made in col-laboration with MPIAHeidelberg. The outputpupil shape is, as ex-pected by principle of asiderostat, an ellipsewhose size and orien-tation vary in function ofthe direction of obser-vation (Fig. 2). Thesmall circle in the cen-tre (at 0.0, 0.0 spaceZenith position) repre-sents the full apertureof the siderostat mirror.Field rotation due toAlt-Azimuth siderostatmovement (bottom partof the figure) and vi-gnetting due to the re-lay optics (central upperpart) can be clearlyidentified.

Based not only ondedicated requirementsfor the optics, mechan-ics and control, but alsoon the above opticaldesign and on the siteinterface, the development of twosiderostats has been contracted toHalfmann Teleskoptechnik GmbH locat-ed at Neusäß near Augsburg. The detaildesign and the manufacture wereachieved in less than one year. Figure 3,taken during the last assembly, showsthe system ready for alignment tests pri-or to packing and delivery to Paranal. The

five mirrors have an optical quality bet-ter than 25 nm RMS and are gold coat-ed. The full structure is made of steel thatpreserves high rigidity. The dimensionsare about 2.5 m in height and 2.2 m indiameter at the level of the yellow frame.Only the upper part will be visible aboveParanal ground. The frame and the low-er part will be located inside the AT sta-tion pier to have the output beam exact-ly at the level of the delay line entrance.With a weight of about 1.75 tons, the com-plete structure will be easily transportedwith a crane to allow, as requested dur-ing commissioning, relocation of thesiderostat between AT stations. For thispurpose a minimum of 8 stations will beequipped with a dedicated anchoringsystem.

The functions of pointing and trackingare realised by a control system coupledto a CCD camera. The CCD is located atthe focus of a guiding scope.

In the background of the picture onecan see the two enclosures ready forpacking. For these enclosures it was de-cided to re-use the same concept as forthe Astronomical Seeing Monitor (ASM)that has been used for two years atParanal. Taking advantage of the ASMexperience, few modifications have beenretrofitted to the design of the siderostatenclosure. What is not shown on the pic-ture is the electrical cabinet that holds thecontrol system. This cabinet will be lo-cated a few metres away from the en-closure.

Some performance tests have beencarried out during the last weeks withgood results. But, the fine-pointing per-

Figure 1: Three-dimensional representation ofthe VLTI Test-Siderostat Optical Path. The di-ameter of the Siderostat mirror is 400 mm andthe second plane mirror (upper part of theperiscope) is at 25° angle of elevation. The useof a Beam Compressor allows to have an 80-mm output pupil (similar to the ones of the UnitTelescopes).

Figure 3.

Figure 2.: Output pupil shape as a function of the direction of observation.

14

formance is still to be measured whengood weather conditions will allow us toobserve a star in the sky of Neusäß.

With an optical design optimised for thetargets of the technical commissioning ofthe VLTI and its instruments, with high-quality optics, high rigidity of the me-chanics and good control electronics, thetwo siderostats to be installed early 2000will offer the capability to characterise theVLTI in its first phase. As done for theAstronomical Seeing Monitor, it is en-visaged to upgrade the CCD camera bya VLT technical CCD and to perform thenecessary upgrade of the control elec-tronics and software to fully meet the ESOVLT standard.

tion process and to produce specificdata-reduction tools. The observing datapresented here were taken with Adonisat the La Silla 3.6-m telescope duringtechnical time, specifically dedicated tothis data-reduction programme.

Adaptive Optics (AO) is now a proventechnology for real-time compensation ofspace objects, to remove the degradingeffects of the Earth’s atmosphere. How-ever, the compensation is never “perfect”and residual wavefront errors remain,which in some cases can lead to signif-icant uncompensated power. This de-creases the image contrast making, insome cases, necessary to use some formof image post-processing to remove theeffects of the system’s point spread func-tion (PSF). The knowledge of good de-convolution techniques suitable on AOimages is also important to boost imageswith low Strehl, e.g. from low-order AOsystems.

AO compensation is achieved via aservo-control loop which uses a referencesignal from a guide star (GS) to zero thewavefront error at each iteration, typicallyevery 2–40 msec chosen depending onthe observing wavelength and referencesignal strength. It is well known that theperformance of an AO system is limitedby the reference signal strength as wellas by the atmospheric coherence lengthr0, the atmospheric coherence time to,and the isoplanatic angle θ0.

In astronomical imaging, exposuretimes from a few msec to tens of minutesare used. The variability of r0 and of theAO PSF can be quite large in short inte-gration times (seconds), while it smoothesout in timescales of minutes. There

# Object 1950 Coordinates Magnitudes

1 Alfa Ori 52 27.8 +07 23 58: V=0.5 C=–4.52 Alfa Sco 26 17.9 –26 19 19: V=1.0 C=–4.03 Alfa Tau 33 03.0 +16 24 30: V=1.5 C=–3.04 L2 Pup 12 01.8 –44 33 17: V=5.1 C=–3.05 R Leo 44 52.1 +11 39 41: V=6.0 C=–3.06 IRC + 10 216 45 14.2 +13 30 40: C=–3.07 V766 Cen 43 40.3 –62 20 25: V=6.5 C=–3.08 Alfa Her 12 21.6 +14 26 46: V=3.5 C=–3.09 VX Sgr 05 02.5 –22 13 56: C=–3.2

10 R Aqr 41 14.1 –15 33 46: V=6.4 C=–3.1

Table 1: List of sources bright enough to be used with 400-mm diameter siderostats (C meansN magnitude corresponding to correlated flux).

Abstract

Adaptive Optics produces diffraction-limited images, always leaving a residualuncorrected image and sometime PSFartifacts due, e.g., to the deformablemirror. Post-processing is in some cas-es necessary to complete the correctionand fully restore the image. The resultsof applying a multi-frame iterative decon-volution algorithm to simulated and ac-tual Adaptive Optics data are presented,showing examples of the application anddemonstrating the usefulness of the tech-nique in Adaptive Optics image post-pro-cessing. The advantage of the algorithmis that the frame-to-frame variability of thePSF is beneficial to its convergence, andthe partial knowledge of the calibratedPSF during the observations is fully ex-ploited for the convergence. The analy-sis considers the aspects of morphology,astrometry, photometry and the effect ofnoise in the images. Point sources andextended objects are considered.

1. Introduction

This paper reports on work done to es-tablish and evaluate data reduction pro-cedures specifically applicable to Adap-tive Optics (AO) data. The programmeevaluated for multi-frame iterative blinddeconvolution, IDAC, is provided by J.Christou and available in the ESO Webpage at http://www.ls.eso.org/lasilla/Telescope/360cat/adonis/html/datared.html#distrib for theUnix platforms.

This work is part of a dedicated effortto guide the AO users in the data reduc-

Myopic Deconvolution of Adaptive Optics ImagesJ.C. CHRISTOU1, D. BONACCINI2, N. AGEORGES2, and F. MARCHIS3

1US Air Force Research Laboratory, Kirtland AFB, New Mexico, USA (christou@as.arizona.edu)2ESO, Garching bei München, Germany (dbonacci@eso.org, nageorge@eso.org)3ESO, Santiago, Chile (fmarchis@eso.org)

is also a slow decline in Strehl Ratio (SR),the system’s performance parameter1,across an imaging field of radius θ0 dueto field anisoplanatism effect. The latterproduces mainly an elongation of the PSFin the direction of the GS, which is func-tion of the object radial distance. The PSFat the reference star has a diffraction-lim-ited core, surrounded by a broad gauss-ian halo.

Static and dynamic artifacts in the im-age due to either print-through of the de-formable mirror actuators, and differen-tial aberrations between the science andWFS optical paths, have also been ob-served. For ADONIS, the former have aGaussian shape with peak intensities inthe range of 0.5% to 1% of the centralPSF core. Although their energy contentis undoubtedly small, if not removedthey make it difficult to unambiguouslyidentify faint objects or faint structuresaround bright sources.

Thus the observer has to worry abouttime-varying and space-varying AOPSFs.In general, the observation of a pointsource as PSF Calibrator, in addition tothe target, is used to further improve thecompensated image via post-processing.However, the AO compensation ob-tained on the target and the PSF cali-brator are not necessarily the same: the

1The Strehl Ratio is a standard measure for theperformance of an AO system. It is the ratio of themaximum intensity of the delivered PSF to the max-imum of the theoretical diffraction-limited PSF whenboth PSFs are normalised to unity. A Strehl Ratio of1 means achievement of the theoretically best per-formance.

15

AO compensation depends upon the at-mospheric statistics, i.e. the coherencelength r0 and the correlation time t0, andhow they relate to the sub-aperture sizeand sampling time of the AO system.These parameters vary in time betweensets of observations of the target and thePSF calibrator, so that the AO system per-forms differently on each. In addition, theAO system is sensitive to the brightnessand extent of the source used to close theloop. Both affect the signal-to-noise(SNR) on the wavefront sensor (WFS).In other words, the PSF calibrator ob-tained ‘off-line’ from the science obser-vation does not give exactly the samePSF of the science frames, and the PSFis only approximately known.

So far, WFS using Avalanche Photo-diodes (APD) in photon-counting modehave been used to rebuild the PSF fromWFS data (J.P. Véran et al., 1997), ob-tained during the science acquisition. Themethod proves rather precise as long asthe guide star is a point object and nottoo faint. Still, only an approximate PSFis recovered, especially with faint GS orwith short exposure times.

Thus, in order to deconvolve the AOobservations, it is useful to use an algo-rithm that is flexible with regard to the PSFas the latter is always known only with lim-ited precision. A “blind” deconvolution al-gorithm is very suitable for this applica-tion. We call it ‘myopic’ as the AO PSFis rather well known, although not per-fectly, either via the PSF calibratormethod or via the PSF reconstructionfrom WFS data.

We report here on the iterative ‘myopic’deconvolution technique, which takesadvantage of multiple frames taken un-der variable PSF conditions. Startingfrom an approximately known averagePSF, and assuming that the science ob-ject is constant in time, the algorithm runsrelaxing the PSF in each frame and find-ing the best combination of PSFs whichdeliver the common science object in allthe frames. The input is an approximat-ed average PSF and a set of N scienceframes taken with the same instrumen-tal set-up. The output is the science ob-ject recovered, and a set of N PSFs. Ingeneral, the more the PSF varies be-tween frames, the easier the science ob-ject detection is. As it is known that thePSF in closed loop varies in time scalesof seconds, this myopic iterative decon-volution algorithm is very well suited toAO data. Long-exposure images taken indifferent seeing conditions, e.g. with dif-ferent PSFs, also benefit from this de-convolution method.

Deconvolution algorithms use a“known” PSF to deconvolve the mea-surement. This is classically an ill-posedinverse problem which has been wellstudied in recent years, and algorithmssuch as Lucy-Richardson are readilyavailable (Lucy, 1974). Multiframe itera-tive blind deconvolution (IBD), whichsolves for both the object and PSF si-multaneously, is ill-posed as well and also

poorly determined. However, the appli-cation of physical constraints upon thisalgorithm permits both the object and PSFto be recovered to an accuracy that de-pends mainly upon the SNR of the ob-servations (Sheppard et al., 1998).

We discuss the physically constrainediterative deconvolution algorithm, de-scribed in section 2, and its applicationto both simulated and real data sets de-scribed in sections 3 and 4. The simula-tions permit us to investigate the algo-rithm’s performance given a known tar-get. We briefly show the effect of noiseon photometry, astrometry and mor-phology. Different data types are inves-tigated including multiple point source tar-gets, binary stars, a galaxy image, Io andRAqr for different SNR conditions. Thispermits performance evaluations on realdata sets by comparison to the predic-tions of the simulations. As previouslymentioned, the algorithm also recoversthe PSFs for the observations, and in-vestigation of these permit evaluation ofthe AO system’s performance e.g. onnon-point-like targets.

We are assuming that the PSF is con-stant across the processed field of view,i.e. we neglect anisoplanatism effects.Work on the latter is in progress at ESOin collaboration with the OsservatorioAstronomico di Bologna, and it will be re-ported later.

2. The Algorithm

The IDAC code is based on the con-jugate gradient error-metric minimisation,blind deconvolution algorithm of Jefferiesand Christou (1993) and is currently un-der further development and testing byan extended group. The advantage ofmultiple frames is to have more infor-mation to break the symmetry of the prob-lem. For a single frame, the target andPSF are exchangeable without usingprior information. For multiple frames, acommon object solution is computedalong with the corresponding PSF foreach observation in the data set. Notethat PSF static artifacts in the image willbe considered part of the object, and haveto be calibrated out. This is done in sev-eral ways, e.g. processing the PSF cal-ibrator data with this same code, ex-tracting the static components and re-moving them from the science object atthe end of the processing, or using im-ages of an artificial point source on theAO setup.

The standard isoplanatic imagingequation can be written as

(1)

or in the Fourier domain as

(2)

where g′(r→) is the measurement, ƒ(r→ ) is

the target, h(r→) is the blur or PSF of the

system and * denotes convolution. n(r→)

represents noise contamination whichcan be some combination of photonnoise and detector noise. The Fouriertransforms are indicated by the corre-sponding uppercase notation, where r

→ isthe spatial index and ƒ

→is the spatial fre-

quency index.For the blind deconvolution case, both

the target and PSF are unknown quanti-ties. In order to solve for these we applyan error-metric minimisation scheme us-ing a conjugate gradient algorithm simul-taneously minimising on several error-metric terms. The first is known as the fi-delity term which measures the consis-tency between the measurements and theestimates.

(3)

where k is the frame index and i is the pix-el index and sik is a “bad” pixel mask whicheliminates cosmic-ray events and “hot”and “dead” pixels from the summation.The ^ indicates the current estimates ofthe variables.

One of the problems with many ofthese iterative algorithms is knowingwhen to terminate the iterations. From (1)and (3) it can be seen that when an ex-act solution is reached, the error-metricdoes not go to zero but to a noise bias,i.e.

(4)

where zero-mean Gaussian noise of rmsσΝ has been assumed for K frames andN pixels per frame. Thus, truncating theiterations at this limit is consistent with thenoise statistics and is an effective regu-larisation procedure to minimise “noiseamplification error”.

It is also necessary to apply physicalconstraints to both the object and PSF.Both are positive and following theapproach of Thiébaut & Conan (1994),we reparameterise both as square quan-tities, i.e.

(5)

It is also noted that the PSF is a band-limited function due to the physical na-ture of the imaging system, i.e. the tele-scope used has a finite aperture andtherefore the PSF has a finite upper-bound to its spatial frequency range ƒc =∆/λ. Thus the PSF estimate is penalisedfor containing information beyond thisspatial frequency limit by the following er-ror metric,

(6)

Figure 1 Left: simu-lated multiple-pointsource truth (4sources). Note thateach star is centredon a single pixel.Right: sample PSFfrom the data cube.Note the diffraction-limited core and Airyrings broken up byresidual aberrations.(Both images dis-played on a square root scale.)

16

2.1 Myopic deconvolution

Prior PSF information is utilised re-ducing the PSF parameter space, betteravoiding local minima solutions, and alsohelping to break the symmetry. This is ap-plied in the form of a mean PSF for themultiple-observation data set. In order toremove the effects of mis-registration fromone frame to another, the shift-and-add(SAA) or peak-stacked mean of the PSFestimates is computed,

(7)

where ipk is the intensity peak location ofthe kth frame. The SAA image is then com-pared to the SAA image of a referencestar by using the following error-metric,

(8)

Equation 3 computes the fidelity termin the image domain. The advantage ofthis is to permit the application of supportconstraints to the measurement as wellas bad-pixel masks. However, Jefferies& Christou (1993) originally computed EFin the Fourier domain where there is theadvantage of applying a bandpass-limitto the observations, i.e. so that those pix-els in the Fourier domain which lie out-side the cut-off frequency ƒc are exclud-ed from the summation. A modification tothis is to apply a spatial frequency SNRweighting for those spatial frequencieswhich lie within the cut-off frequency, i.e.

(9)

where Θuk in the simplistic case is simplya band-pass filter, i.e. is unity for spatialfrequencies lower than ƒc and zero out-side. However, it can also be defined asa low-pass filter customised to the SNRof the data, e.g. an Optimal Filter (in thelinear minimum mean squared errorsense.)

(10)

where |Nuk|2 is assumed to be white noiseand is obtained from the mean value of|G’uk|2 at spatial frequencies > ƒc.

The algorithm minimises on the com-bined error metric of the individual onesdescribed above, i.e.

(11)

where α j are weights which regulariseeach of the additional error-metric terms.

3. Application to Simulated Data

In order to demonstrate the algo-rithm’s performance on AO data, duringour study it was first applied to simula-

tions. In this section we present resultsfrom application to a simple object – a setof multiple point sources for which the as-trometry and photometry can be recov-ered. The algorithm is next applied to afar more complicated target, a galaxy.

3.1. Multiple-Star Case

A data set comprised of four pointsources was created and convolved witha data cube (64 frames) of simulated at-mospheric PSFs for D/r0 = 2 such thateach of the PSFs was dominated by a dif-fraction-limited core but there was alsospeckle noise. Figure 1 shows the “truth”object and one of the “truth” PSFs. Six dif-ferent levels of zero-mean Gaussiannoise then contaminated these simulat-ed data. Table 1 gives the SNR levels (ordynamic range) for these cases as de-fined by the ratio of the peak signalin the data cube to the rms of the noise(PSNR). Figure 2 shows the first frameof the data cube for the six different SNRcases.

For the initial tests,all 64 frames of thedata cubes were re-duced simultaneous-ly. The algorithm wasminimised on the fi-delity term (comput-ed in the image do-main) and the band-pass constraint only,E = EF + EBL, usingthe SAA image of themeasurements asthe initial object esti-

mate (see Fig. 3) and the SAA image ofthe brightest source was used as the ini-tial PSF estimate. Note that this sourceis in fact a close binary. Thus, the algo-rithm was not run “blind” but reasonablestarting “guesses” or estimates wereused based on the data. The iterationswere terminated when either the noiselimit (equation 4) was reached or whenthe error-metric changed by less than onepart in 106.

The resulting reconstructed objects areshown in Figure 4. In all six cases. All foursources have been recovered, and innearly all cases, the individual sources arerestored to a single pixel demonstratingthe algorithm’s ability to “super-resolve”.Comparison of the lowest SNR case re-construction clearly shows detection ofthe faintest source which is very am-biguous and masked by noise in the rawdata (Fig. 2). Comparison to the SAA im-ages (Fig. 3) shows the faintest sourceto be a “brightening” of the Airy ringaround the brightest object. The super-

Figure 2: First frameof the data set for thesix different SNRcases with the noiseincreasing from leftto right and top tobottom. (All imagesdisplayed on asquare root scale.)

PSNR (P) DB [20 log10(P)] ∆ Magnitude

819 58 7.3410 52 6.5205 46 5.8102 40 5.051 34 4.226 28 3.5

Table 1: Signal-to-noise ratios for the simulated multiple-star data setsgiven as the ratio of the peak signal to the rms of the additive noise(PSNR) as well as in decibels and stellar magnitudes.

17

resolution of the algorithm permits it to bedetected as a separate source.

Using multiple point-source objectsmakes it relatively easy to determine thereconstructed object fidelity permitting di-rect measurements of not only the as-trometry, i.e. the relative positions of thesources, but also the relative photome-try, i.e. the brightness of the sources. Inall cases, the relative locations of thesources matches the truth. The relativephotometry was computed, in stellarmagnitudes, by measuring the peak pix-el values of the four sources since thesources were all initially located at inte-ger pixel locations. Figure 5 shows theresiduals of the photometry as a functionof the sources’ true brightness values.Note that as the SNR decreases, the fi-delity on the fainter source becomesworse in a systematic way with the mag-nitude difference increasing as the noise.

The reason for the underestimation ofthe fainter sources is because of the pres-ence of the noise. This is further illustratedin Figure 6. This plots the residuals fromFigure 5 against the SNR of each of thefour sources for the six different SNR con-

Figure 4: Reconstructed objects for the six different SNR cases with thenoise increasing from left to right and top to bottom. (All images displayedon a square root scale.)

Figure 3: Initial object estimates for the six different SNR cases withthe noise increasing from left to right and top to bottom. Note thatthe brightest source (bottom left) was used as the initial PSF esti-mate. (All images displayed on a square root scale.)

Figure 5:Residual pho-tometry errorsfrom measuringthe brightestpixel for the foursources in theobject recon-structions for the6 different SNRcases comparedto the sources’true brightnessnormalised to thebrightest source.

Figure 6:Residual photom-etry errors ofFigure 5 plottedagainst the SNRof the 4 sourcesfor the differentSNR conditions.The horizontaldashed linerepresents per-fect reconstruc-tion and thevertical dashedline representsthe 3σ noiselevel.

Figure 7: Reconstructed PSFs for (left toright) decreasing SNR for the first frame of thedata cube. Top: the brightest source used asthe initial PSF estimate. Note its contamina-tion by the faintest source. Bottom: recon-structed PSFs. H

18

ditions. Note that the poorest recon-struction is below the 3σ level and thatgood reconstructions are obtained forSNRs greater than 6σ. This demon-strates that the sources that become “lost”in the noise are difficult to extract accu-rately.

As mentioned above, the object andthe PSF are both reconstructed using de-convolution. Figure 7 shows the recon-structed PSF for the first data frame of thedata cube for each of the SNR casescompared to the initial estimates for thefirst frame of the data cube. Comparisonof the highest SNR case shows goodagreement between the reconstructionand the “truth”. As the SNR decreases,there is little difference in the recon-structions with more noise for the lowestSNR case. This analysis is useful whenplanning the observation’s integrationtimes, according to the desired photo-metric accuracy.

3.1.1 Effect of noise propagation

We ask “How well does the algorithmdeal with the noise?” As seen above,some noise does appear in the object re-constructions and the PSFs, especially forthe higher noise cases, but most of itshows up in the residuals. This is illus-trated in Figure 8 which shows his-tograms of the residuals for the 6 differ-ent SNR reconstructions. As can beseen, these histograms appear to haveGaussian distributions. The rms value ofthe residuals compare very favourably to

little difference between them. The rela-tive photometry of the four sources wascalculated from the peak pixel values asabove for all reconstructions. The meanand standard deviations were computedfor the multiple reduction cases, i.e.when the number of frames was less than64, and these are shown plotted in Figure10. This shows the repeatability of thephotometry and indicates that when thenumber of frames used is relatively small,i.e. ~ 8 or less, then the photometry onthe fainter sources can show significantvariation. There is also a trend that as theframe number decreases, the photome-try of the fainter sources is underesti-mated. This is probably indicative of thecombination of high noise and smallnumber of frames making it difficult for thealgorithm to distinguish between signaland noise.

3.2 Galaxy case

The multiple star represents imagingof a relatively simple object. A more typ-ical extended target is that of a galaxy.Two sets of AO observations of a bare-ly resolved galaxy were generated. Thefirst using simulated AO PSFs for goodcompensation with SRs of ~ 50% whichis typical of that expected by using e.g.a 60-element curvature system in K-band. The second had a much poorercompensation with SRs ~ 10% typical ofa faint GS. The former data set were com-puted from simulated PSFs, whereasthe latter was obtained from ADONIS ob-servations of a faint isolated star. EachPSF data set comprised ten separateframes. These PSFs were over Nyquistsampled by a factor of two correspond-ing to an image scale of 32 mas/pixel as-suming K-band imaging on a 3.6-m tele-scope.

The galaxy reference image for thesimulations was obtained from the 512 ×512 galaxy image contained in the IRAFsystem. This was reduced using 8 × 8 pix-el block sums to a 64 × 64 pixel imagewhich was then embedded in a 128 × 128pixel field. This image was then convolvedwith the PSFs and then contaminated byzero-mean additive noise with PSNRs of1000, 200 and 50 corresponding to dy-namic ranges of 7.5, 5.8 and 4.2 magni-

the input noise statistics and the residu-als do have a zero-mean. The reducedphotometric accuracy for the lower SNRcases shows the effect of the presenceof noise on the algorithm’s ability tobreak the symmetry between the PSFand the object, especially when using ini-tial noisy PSF and object estimates.

These results demonstrate the algo-rithm’s ability to recover not only the ob-ject distribution but also the correspond-ing PSFs for different SNR conditions.The residuals compare well to the inputnoise conditions, and relative photome-try of the reconstructed object shows re-sults consistent with the different SNRs.In all cases, 64 frames were reduced. Forthese data, the PSFs have a com-mon diffraction-limited core but have dif-ferent speckle noise. Thus, the multipleframes are significantly different fromeach other.

3.1.2 Effect of frame multiplicity

The next question we ask is “How welldoes the algorithm perform as the num-ber of frames is reduced?” This was in-vestigated by taking the third highest SNRdata and reducing it using 1 set of 64frames, 2 sets of 32 frames, 4 sets of 16frames, 8 sets of 8 frames and 16 setsof 4 frames. The reconstructed object forthe first of each of these sets is shownin Figure 9. Note that there appears to be

Figure 8: Histo-grams of theresiduals for thesix different SNRcases. The mea-sured residualsrms value isshown for eachcase compared tothe rms value ofthe input additivenoise (in paren-theses).

Figure 9: Reconstructed object for the thirdhighest SNR case for (left to right and top tobottom) 4, 8, 16, 32, and 64 frame reductions.

Figure 10:Residual pho-tometry errorsfrom measuringthe brightestpixel for each ofthe four sourcesin the objectreconstructionsfor the differentmultiple-framereductions. Themean and stan-dard deviationsare shown forthe reductionsusing less than64 frames.

19

tudes respectively. Figure 11 shows theuncontaminated mean observation for thetwo SNR cases obtained from shift-and-add analysis.

Figure 12 illustrates the effect of theadditive noise on the resolution of the im-ages. The complete average power spec-trum of the ten frames for the 3 SNR cas-es of the lower SR data radial profiles areshown. Also shown are the noise bias lev-els. Where they intersect with the dataspectrum is more clearly seen in the toppanel which shows the SNR of the mea-sured power spectrum. Note that the high-est spatial-frequency passed by the pupilis set to unity and for a PSNR of 1000,this is reached. However, as the noise in-creases, the effective frequency cut-offdecreases to 0.85 for a PSNR of 200 andto 0.52 for a PSNR of 50. Thus, high spa-tial frequency information is further cor-rupted by the additive noise.

The data sets were reduced using thenoisy SAA image as the initial object es-timate and a SAA of a reference star asthe initial PSF. This PSF estimate was ob-tained from the same set of data, whichcreated the galaxy images, but excludingthe used PSFs. The iterations were ter-minated when the noise bias level wasreached. Full-field support was used forthe object and the PSFs as well as theband-limit. All six cases have beenanalysed (3 SNRs each for the two SRcases). For both SR cases, it can be seenin the curves of Figure 13 that as the SNRdecreases, the effective resolution alsodecreases. Also, the higher the initial SR,the better the reconstruction. Super-res-olution is achieved over the diffraction-lim-ited image for all but the lowest SNR case

4. Application to ADONIS Data

In this section, we present results ofthe IDAC algorithm as applied to differ-ent types of measured data. This includesbinary stars, then the Galilean satellite Ioas an extended object, and R Aqr, whichhas faint structure around a bright cen-tral source.

4.1. Closeby point sources: τ Canis Majoris

Data were taken of the bright binarystar τ CMa (= HR2782 = ADS 5977A) atthe ESO 3.6-m during a technical run inFebruary 1996. The Yale Bright StarCatalogue lists it as being an O91b starof mv = 4.40, having an equal magnitudecompanion at a separation of 0.2″. Morerecently, the CHARA group have mea-sured it obtaining a separation of 0.160″in 1989.9 (Hartkopf et al., 1993). The dif-fraction-limits of the telescope are 0.072″at J (1.25µm), 0.095″ at H (1.65µm) and0.126″ at K (2.2µm). Thus the data wereNyquist-sampled at J, i.e. at 0.035″/pix-

Figure 11: Simulated AO galaxy observationsfor high SR (left) and low SR (right). (Displayedon a square-root scale intensity.)

Figure 12: Hownoise affects resolu-tion. The bottompanel shows azi-muthally averagedpower spectra forthe three SNRobservations of thelower SR data. Thetop panel shows thecorresponding SNR.Note that as thenoise increases thecut-off frequencydecreases.

for the lower Strehldata, which is re-stored to diffraction-limited. The recon-structions show thedetectability of thespiral arm structurelost in the raw com-pensated images.

Figure 13 showsthe radial profilesof the Fourier mod-uli of the low SR re-constructed imagescompared to thetruth (the top line ofthe four). The verti-cal lines indicatethe noise-effectivecut-off frequency fN(see Fig. 12) for thethree SNRs. Theseplots show that within fN, there is good re-construction but for the super-resolutionregime, in these cases for f > fN, the high-er spatial frequencies are attenuatedwith respect to the truth. In order to quan-titatively measure how good the recon-struction is, the normalised cross-spec-trum was computed between the recon-struction and the truth, i.e.

(12)

which will be unity for a perfect correla-tion. Note that the numerator is the mod-ulus of the cross-spectrum only so thatmis-registration of the two images doesnot affect the value. The summation of XSi,yields the correlation coefficient for the re-construction, i.e.

(13)

where Λ represents the spatial frequen-cy region over which the summation iscomputed. Setting Λ to the SNR support(see Figure 12) yields correlation coeffi-cients of 1.1, 1.0 and 0.9 for decreasingsignal-to-noise showing excellent recon-struction within those regimes.

Figure 13: Azimuthally averaged radial profiles of the Fourier moduliof the reconstructed images for the lower SR data. Decreasing SNRgives lower spatial frequency cutoff and worse MTF.

Figure 14: J (top left), H (top right), and K-band(bottom) peak tracked images (smoothed to re-move the pixel-to-pixel variations) with super-imposed logarithmic contours emphasisingthe extended halo structure in the images. Thecontour levels are at 1, 1.6, 2.5, 4, 6.3, 10, 16,25, 40, 63% of the peak value in each of theimages.

20

el and over-sampled at the longerwavelengths easily resolving the com-panion. The data were taken in “speck-le” mode, i.e. using a series of 200 shortexposures (texp = 50 ms) with a sampletime of ∆t = 0.74s. Seeing was estimat-ed to be 0.8″ at visible wavelengths(0.55µm). The AO loop was operated atmaximum gain with the Shack-HartmanReticon sensor in line mode, at a rate of200Hz.

The initial data post-processing con-sisted of background correction and flatfielding with bad-pixel correction. Figure14 shows SAA images for the three ob-serving bands. These clearly show thecompanion but they are also contami-nated by extended halo due to the resid-ual wavefront errors of the AO compen-sation as well as a “fixed-pattern” error inthe form of triangular coma giving rise toa “lumpy” Airy ring about the image cores.These clearly illustrate the need for de-convolution.

In the initial application of the algorithm,the original 200-frame data sets were re-duced to sub-sets of 16 frames of 64 ×64 pixels i.e. 2.24″ × 2.24″. Thus, therewere (M+1)N2 = 69632 variables to beminimized for M convolution images of

size N × N. For all three wavebands, thefirst 16 frames were used and the initialobject estimates were the SAA imagesshown in Figure 14. As no prior PSFknowledge existed, the initial PSF esti-mates were chosen to be Gaussians. Thesuper-resolved images were filtered backwith a Gaussian of FWHM = 0.083″which slightly over resolves the diffraction-limit at this wavelength, i.e. λ/D P 0.12″.Figures 15, 16 and 17 show the recon-structed objects for the three wavebandsobtained from the first and second setsof 16 frames of the 200-frame data sets.Note the strong similarity between the im-age pairs with differences occurring onlyat the few-percent level and also note howclean the reconstructions are.

Keith Hege at Steward Observatoryhas suggested that Speckle Holography(SH) (Hege, 1989) imposes a further con-straint on the data. This is applied outsideof the blind deconvolution loop and mayprevent stagnation into a local minimum.It makes use of a multiple-frame data setalong with PSF estimates for each frame.The object estimate was obtained as fol-lows, i.e.

(14)

so that when the PSF estimates equal thetrue PSFs then the band-limited object es-timate will equal the true band-limited ob-ject. This multiple-frame deconvolutionhas been shown to be less noise sensi-

tive than deconvolving on a frame-by-frame basis. The object estimate from thisprocedure coupled with the PSFs, whichproduced it, can then be fed back into themyopic deconvolution code as new esti-mates.

For the three data sets reported here,a variant of this was performed. The PSFsfor all 200 frames at each waveband werecomputed by simply taking the super-re-solved object estimate (from the first 16-frame start-up sub-set, using GaussianPSFs) and deconvolving into each mea-sured data frame. These were then usedas the PSF estimates in equation (2) togenerate a new object estimate. Blind de-convolution was then applied to all 200frames for each waveband. The resultingobject estimates are shown in Figure 18.

These images show a very cleandeconvolution with lack of backgroundfeatures below the 1% levels. The as-trometry and photometry of the final im-ages obtained by Gaussian fits are giv-en in Table 2 and show very self-consis-tent astrometry for the three wavebands.

4.2. Io observations

The Jovian Galilean satellite, Io, is typ-ical of the type of Solar-System objectswhich are benefitted by AO observations.At thermal infrared wavelengths, the vol-canic hot spots can be detected againstthe cooler background surface withoutwaiting for the moon to go into eclipse.Its volcanic activity is monitored from theground using the thermal camera facility(COMIC) which is available on the ADO-NIS system at the ESO 3.6-m (Le Mignantet al. 1998). It was observed in September1998 in the L’ band (λ = 3.809, ∆λ = 0.623µm) with an image scale of 0.100″/pixel,corresponding to ≈ 400 km on the surfaceof Io, and an integration time of 500 msec.The source had a visual magnitude of mv= 5.1, and measurements of the seeingusing the Differential Image Motion

Band ∆r (mas) PA (°) ∆m

J 149 34 0.84H 149 33 0.80K 149 33 0.94

Table 2: Astrometry and photometry for the200-frame τ CMa blind deconvolutions recon-structions for the three wavebands.

Figure 15: J-band τ CMa object reconstructionsfor the first (left) and second (right) 16 frames.Same contour levels as Figure 14.

Figure 16: H-band τ CMa object reconstruc-tions for the first (left) and second (right) 16frames. Same contour levels as Figure 14.

Figure 17: K-band τ CMa object reconstruc-tions for the first (left) and second (right) 16frames. Same contour levels as Figure 14.

Figure 18: Final Gaussian smoothed J (top left),H (top right) and K-band (bottom) τ CMa ob-ject reconstructions using all 200 data framesand an SH start. Same contour levels as Figure17.

Figure 19: ADONIS imaging of lo. The en-semble average of a set of ten exposures isshown on the left and the reconstructed ob-ject is shown on the right.

Figure 20:Ensemble averagereference star(PSF calibrator)image for the loobservations.

21

Monitor (DIMM) was 0.75″. At this wave-length the 3.6-m has a diffraction-limit of0.226″ so that the object was well over-sampled. A PSF calibrator with similar vis-ible brightness HD174974 (mv = 5.0, G5)was observed to yield initial PSF esti-mates. The field of view of the observa-tions was 12.8″ × 12.8″.

The data set comprised ten separateobservations of Io, the complete averageis shown in Figure 19. This was used asthe initial object estimate. The mean ofthe point source calibrator was used asthe initial PSF estimate, shown in Figure20. Note that the PSF is diffraction-limit-ed with a Strehl ratio of ~ 50%.

The reconstructed object is also shownin Figure 19. This has been regularisedby suppressing the high-spatial frequen-cy information with apodisation by a per-fect telescope transfer function. Thus, theeffective resolution is the same as for theobservation, but now the “hot spot” on thesatellite’s surface, Loki, is more clearlydiscerned against the smooth distributionof the surface.

The algorithm’s ability to super-re-solve has been discussed earlier. For ob-jects with sharp edges, such as a plan-etary satellite or asteroid, truncation of thehigh spatial frequencies results in “ring-ing” in the image domain. The super-res-olution achieved on these types of objectshas been discussed previously (Christouet al., 1994) and is further illustrated be-low with simulations of Io imaging. Figure21 compares a simulated Io object, i.e. auniform disk with two hot-spots, to the re-construction, both of them for super-res-olution which is ~ 1.5 × diffraction-limit-ed case. As can be seen, these two im-ages (top-right and bottom-right) agree

very well. Just how well, is better seen inthe Fourier moduli of the truth image com-pared to the reconstruction before filter-ing. This is shown with the azimuthally av-eraged radial profiles in Figure 22.

The radial profiles match well out to~ 80% of the diffraction-limit where the re-construction has greater power than thetruth. Beyond the diffraction-limit, the re-construction has non-zero power but ata lower level than the truth. The spatialfrequency range shown indicates the re-gion over which the filtered images inFigure 21 were restored.

A thorough analysis of Io adaptive op-tics imaging and the resulting science canbe found in a recent article by Marchis etal. (1999).

4.3 R Aquarii

R Aquarii is an eclipsing symbiotic bi-nary surrounded by morphologically com-plex nebulosity which extends to ~ 1 arc-minute from the central source. At small-er scales, there is an elongated jet-likeemission feature (Hollis et al., 1985). Itwas observed in December 1996 withADONIS equipped with a Fabry-Perot atthe ESO 3.6-m in both the 2µm continu-um and the Hydrogen emission line, Brγ(2.1655µm) in order to detect and resolvestructures of the nebulosity. The imagescale was 0.05″/pixel with exposure timesof 0.6s for a total integration time of 48s,for the Brγ observations reduced here.The individual frame PSNR was com-puted to be ~ 2000 for the central sourcebut for the nebulosity, ~ 3″ away, it was~ 5. In order to improve the SNR, the datawere binned down from 80 to 10 frames,increasing the nebulosity SNR to ~ 15.

The 10 eight-frame-binned data werereduced using the PSF constraint (seeequation 8) in addition to the band-limitand fidelity terms, where the “known” PSFwas obtained from the shift-and-add im-age of the central source for all 80 ob-servations. After convergence, the pro-cessing was continued with the PSFconstraint removed. Figure 23 comparesthe nebulosity and central source for asingle observation, an eight-frame aver-age, as used for the deconvolution show-

ing the improved SNR, and for the 80-frame sum. The reconstructed objects areshown in Figure 24 for both the Brγ andcontinuum observations, both with andwithout the PSF constraint. The initial es-timate in each case was the Richardson-Lucy reconstruction. Note how after thePSF constraint is “relaxed”, the mor-phology of the nebulosity changes to bet-ter fit the observations. This nebulosityhas structure on the scale of ~ 0.1″, i.e.diffraction-limited and there are obviousstructural differences between the twowavebands.

5. Discussion

In this article we have shown the ef-fectiveness of using a multi-frame itera-tive myopic deconvolution technique onAO data. Using the IDAC SW on both sim-ulations and real data, it is shown that itis a powerful tool for deconvolving AO im-ages, to remove the effects of the un-compensated components of the PSFs

Figure 21: Simulated observations and re-ductions of an Io-type object: truth (top left),filtered truth (top right), observation (bottom left)and filtered reduction (bottom-right).

Figure 22:Azimuthally aver-aged radial profilesthrough the Fouriermodulus of the pixel-limited truth image(solid-line, seeFigure 24) and thereconstructed image(dashed line).

Figure 23: R Aquarii observations in Brγ.Single frame (left), 8-frame average (centre)and 80-rame average (right).

Figure 24: R Aquarii nebulosity reductions: Brγ(top) and 2 µm continuum (bottom). Left to rightare the Richardson-Lucy reductions used forthe initial object estimate, the PSF constrainedIDAC reduction and the PSF relaxed recon-struction.

22

and reach full diffraction performances.The ESO ADONIS PSFs show a vari-ability, which is common to AO systems.It is this variability and the lack of a pre-cise frame-by-frame PSF determinationthat makes post-processing of the datadifficult. The iterative multiframe decon-volution method described takes suc-cessfully advantage of this intrinsic vari-ability.

The linearity of the algorithm and itspreservation of photometry is demon-strated on the four star simulations.

Multi-frame iterative myopic decon-volution enforces the very important con-straint that observations of the same ob-ject will yield a common object result.Some prior PSF information is extreme-ly useful in reducing the search space forthe solution as demonstrated by theR Aquarii results.

Observations of a PSF calibrator, asdemonstrated here, make an excellentinitial PSF estimate, and this can be fur-ther strengthened by application of thePSF constraint. A variation on this ap-proach has been utilised by Conan et al.(1997) and Véran et al. (1997). They havedemonstrated that a very good initial PSFestimate can be obtained from a statis-

tical analysis of the residual wavefront er-rors. Currently, this approach is applica-ble only to AO systems using photoncounters, but it should be extendable towavefront sensor systems using detec-tors with read-out noise as well, such asCCDs. Combined statistical PSF esti-mation and blind deconvolution post-processing has been discussed by Fuscoet al. (1998) and Christou et al. (1997).

6. Acknowledgements

We would like to thank the ESO staffat La Silla 3.6-m for the effective ob-serving support during the observingruns. We would also like to thank KeithHege, Matt Chesalka and Stuart Jefferiesfor the IDAC code and for useful discus-sions about its application. This work wassupported by the Adaptive Optics Groupof the Instrumentation Division, at theEuropean Southern Observatory.

Bibliography

Christou, J.C., Hege, E.K., Jefferies, S.M., &Keller, Ch.U., 1994, Proc SPIE, ed. J.B.Breckenridge, 2200, 433–444.

Christou, J.C., Bonaccini, D., & Ageorge, N.,1997, Proc. SPIE, ed. R.K. Tyson & R.Q.Fugate, 3126, 68–80.

Christou, J.C., Marchis, F., Ageorges, N., Bo-naccini, D., & Rigaut, F. 1998a, ed. D.Bonaccini & R.Q. Fugate, 3353, 984–993.

Christou, J.C., Hege, E.K., Jefferies, S.M., &Cheselka, M., 1998b, Proc. SPIE, ed. A.D.Devir, A. Kohnle, U. Schreiber, & C. Werner,3494, 175–186.

Le Mignant, D., Marchis, F., Bonaccini, D., etal., 1998, Proc ESO/OSA, ed. D. Bonaccini,56, 287–301.

Marchis, F., Prangé, R., Christou, J.C., 1999,Icarus, submitted.

Fusco, T., Véran, J.-P., Conan, J.-M., &Mugnier, L.M., 1999, Astron. Astrophys.,134, 193.

Hartkopf, W.I., Mason, B.D., Barry, D.J.,McAlister, H.A., Bagnuolo, W.G., & Prieto,C.M., 1993, Astron. J., 106, 352.

Hollis, J.M., Kafatos, M., Michalitsianos, A.G.,& McAlister, H.A., 185, Astrophys. J., 289,765.

Jefferies, S.M. & Christou, J.C., Astrophys. J.,1993, 415, 862.

Lucy, L., 1974, Astrophys. J., 79, 745.Sheppard, D.G., Hunt, B.R., & Marcellin,

M.W., 1998, J. Opt. Soc. Am. A., 15,978–992.

Thiébaut, E., & Conan., J.-M., J. Opt. Soc. Am.A., 1994, 12, 485.

Véran, J.P., Rigaut, F., Maître, H., & Rouan,D., 1997, J. Opt. Soc. Am. A., 14, 3057.

L A T E S T N E W S

“First Light” for VLT High-Resolution SpectrographUVES(Excerpt from ESO Press Release 15/99, 5th October 1999)

A major new astronomical instrumentfor the ESO Very Large Telescope atParanal (Chile), the UVES high-resolu-tion spectrograph, made its first obser-vations of astronomical objects onSeptember 27, 1999. The astronomersare delighted with the quality of thespectra obtained at this moment of “FirstLight”. Although much fine-tuning stillhas to be done, this early success prom-ises well for new and exciting science pro-jects with this large European researchfacility.

Astronomical Instruments at VLT KUEYEN

The second VLT 8.2-m Unit Tele-scope, KUEYEN (“The Moon” in the Ma-puche language), is in the process of be-ing tuned to perfection before it will be“handed” over to the astronomers onApril 1, 2000.

The testing of the new giant tele-scope has been successfully completed.The latest pointing tests were very pos-itive and, from real performance mea-surements covering the entire operating

range of the telescope, the overall ac-curacy on the sky was found to be 0.85arcsec (the RMS-value). This is an excel-lent result for any telescope and impliesthat KUEYEN (as is already the case forANTU) will be able to acquire its futuretarget objects securely and efficiently,thus saving precious observing time.

The three instruments foreseen atKUEYEN are UVES, FORS2 andFLAMES. They are all dedicated to theinvestigation of the spectroscopic prop-erties of faint stars and galaxies in theUniverse.

The UVES Instrument

The Ultraviolet Visual Echelle Spectro-graph (UVES) is the first instrument onKueyen. It was built by ESO, with thecollaboration of the Trieste Observatory(Italy) for the control software. Completetests of its optical and mechanical com-ponents, as well as of its CCD detectorsand of the complex control system, weremade in the laboratories of the ESOHeadquarters in Garching (Germany)before it was fully dismounted and

shipped to the ESO Paranal Observatory,130 km south of Antofagasta (Chile).There, the different pieces of UVES (witha total weight of 8 tons) were carefully re-assembled on the Nasmyth platform ofKUEYEN and made ready for real ob-servations.

UVES is a complex two-channelspectrograph that has been built aroundtwo giant optical (echelle diffraction)gratings, each ruled on a 84 cm × 21 cm× 12 cm block of the ceramic materialZerodur (the same that is used forthe VLT 8.2-m main mirrors) and weigh-ing more than 60 kg. These echellegratings disperse the light from celestialobjects collected by the telescope into itsconstituent wavelengths (colours).

UVES’ resolving power (an opticalterm that indicates the ratio betweena given wavelength and the smallestwavelength difference between two spec-tral lines that are clearly separated bythe spectrograph) may reach 110,000,a very high value for an astronomi-cal instrument of such a large size. Thismeans for instance that even com-paratively small changes in radial ve-

Figure 1: The powerof UVES is demonstrat-ed by this two-hourtest exposure of thesouthern quasar QSOHE2217-2818 with U-magnitude = 16.5 anda redshift of z = 2.4.This UVES echellespectrum is recordedin different orders (theindividual horizontallines) and altogethercovers the wavelengthinterval between 330–450 nm (from the bot-tom to the top). It illus-trates the excellent ca-pability of UVES to

work in the UV-band on even faint targets. Simultaneously with this observation, UVES alsorecorded the adjacent spectral region 465–660 nm in its other channel. The broad Lyman-alphaemission from ionised hydrogen associated with the powerful energy source of the QSO is seenin the upper half of the spectrum at wavelength 413 nm. At shorter wavelengths, the dark re-gions in the spectrum are Lyman-alpha absorption lines from intervening, neutral hydrogen gaslocated along the line-of-sight at different redshifts (the so-called Lyman-alpha forest) in the red-shift interval z = 1.7–2.4. Note that since this exposure was done with the nearly Full Moon abovethe horizon, an underlying, faint absorption-line spectrum of reflected sunlight is also visible.

This false-colour image has been extracted from an UVES echelle spectrum of SN 1987A, witha slit width of 1 arcsec only. The upper part shows the emission lines of nitrogen, sulfur and hy-drogen, as recorded in some of the spectral orders. The pixel coordinates (X,Y) in the originalframe are indicated; the red colour indicates the highest intensities. Below is a more detailedview of the complex H-alpha emission line, with the corresponding velocities and the positionalong the spectrograph slit indicated. Several components of this line can be distinguished. Thebulk of the emission (here shown in red colour) comes from the ring surrounding the supernova;the elongated shape here is due to the differential velocity exhibited by the near (to us) and farsides of the ring. The two bright spots on either side are emission from two outer rings. The ex-tended emission in the velocity direction originates from material inside the ring upon which thefastest moving ejecta from the supernova have impacted. Finally, there is a broad emission ex-tending all along the spectrograph slit (here mostly yellow) upon which the ring emission is su-perimposed. This is not associated with the supernova itself, but is H-alpha emission by diffusegas in the Large Magellanic Cloud (LMC) in which SN 1987A is located.

23

locity (a few km/sec only) can be accu-rately measured and also that it is pos-sible to detect the faint spectral sig-natures of very rare elements in celestialobjects.

One UVES channel is optimised for theultraviolet and blue, the other for visualand red light. The spectra are digitallyrecorded by two highly efficient CCD de-

tectors for subsequent analysis and as-trophysical interpretation. By optimisingthe transmission of the various opticalcomponents in its two channels, UVEShas a very high efficiency all the way fromthe UV (wavelength about 300 nm) to thenear-infrared (1000 nm or 1 µm). Thisguarantees that only a minimum of theprecious light that is collected by KUEYEN

is lost and that detailed spectra can beobtained of even quite faint objects,down to about magnitude 20 (corre-sponding to nearly one million timesfainter than what can be perceived withthe unaided eye). The possibility of do-ing simultaneous observations in thetwo channels (with a dichroic mirror) en-sures a further gain in data gathering ef-ficiency.

First Observations with UVES

In the evening of September 27, 1999,the ESO astronomers turned theKUEYEN telescope and – for the first time– focussed the light of stars and galax-ies on the entrance aperture of the UVESinstrument. This is the crucial momentof “First Light” for a new astronomi-cal facility.

Much of the time during the first ob-serving nights was spent by functionaltests of the various observation modesand by targeting “standard stars” withwell-known properties in order to mea-sure the performance of the new instru-ment. They showed that it is behavingvery well. This marks the beginning of aperiod of progressive fine-tuning thatwill ultimately bring UVES to peak per-formance.

The astronomers also did a few “sci-entific” observations during these nights,aimed at exploring the capabilities of theirnew spectrograph. They were eager todo so, also because UVES is the firstspectrograph of this type installed at atelescope of large diameter in the south-ern hemisphere.

Many exciting research possibilitiesare now opening with UVES. They in-clude a study of the chemical histo-ry of many galaxies in the Local Group,e.g. by observing the most metal-poor(oldest) stars in the Milky Way Galaxyand by obtaining the first, extremelydetailed spectra of their brightest starsin the Magellanic Clouds. Quasars anddistant compact galaxies will also beamong the most favoured targets of thefirst UVES observers, not least becausetheir spectra carry crucial informationabout the density, physical state andchemical composition of the earlyUniverse.

ErratumIn the June issue of The Messenger

(No. 96), page 26 (article by H. Lamyand D. Hutsemékers), formula 2 shouldread:

24

The editors of the La Silla News Page would like to welcome read-ers of the fourteenth edition of a page devoted to reporting on tech-nical updates and observational achievements at La Silla. We wouldlike this page to inform the astronomical community of changesmade to telescopes, instruments, operations, and of instrumentalperformances that cannot be reported conveniently elsewhere.Contributions and inquiries to this page from the community aremost welcome.

News from the NTTO. HAINAUT and the NTT Team

During the past months, many thingshave changed at the NTT. In June, weclosed the telescope during 10 days, fora major maintenance period. During thattime, our computer system was com-pletely upgraded: the 99FEB version ofthe VLT software was installed, togeth-er with the latest version of the Data FlowSoftware. In parallel, several machineswere either changed or received addi-tional memory, making the whole systemmuch faster. Visiting astronomers will bepleased to learn that the observer’s work-station has been replaced by an HP-C360 (a very powerful machine) equippedwith 50 Gb of disk for data storage andanalysis. The latest versions of MIDAS,IRAF, IDL and eclipse are installed andrunning on this machine. A consequenceof the workstation upgrade is that P2PP(the “Phase II Preparation Programme”used to prepare the observations) isrunning much faster: one no longer hasto wait several seconds for a pop-up toappear. The La Silla software team dida great job at installing this softwarewith the help of colleagues from Garchingand Paranal. At this time, there are stilla few details to be ironed out (e.g. a prob-lem of compatibility between the new soft-ware and the old time distribution cardson the axes computers), but they shouldbe solved soon, and have no major im-pact on the operation.

During the June technical time, all thetelescope optics were maintained: AlainGilliotte washed the main and tertiarymirrors with soap and water (cf. Fig. 1).The result is excellent: M1’s reflectiv-ity is back to 89.5%, while it was 90.1%right after its last aluminisation two yearsago. The micro-roughness of the mirrorswas not fully restored by the washing: itis now 37Å, instead of 10Å after alumin-isation. We are investigating the conse-quences of this change on the diffusedlight. The secondary mirror was also alu-minised.

EMMI and SUSI2 received a completeoptomechanics maintenance: all the op-tics were cleaned, and some motorswere changed. The dichroic used inEMMI-DIMD was inspected, as it wassuspected to cause some problems, butwas found in good shape. SUSI’s M4 (thepick-up mirror) was found to have adamaged coating; the whole mirror is be-ing replaced. Various narrow band filtersjust arrived for SUSI2: Hα, Hα (6000km/s), Hβ, Hβ (6000 km/s), OIII and He II.They will be tested and hopefully offeredsoon to the observers.

SOFI has been misbehaving over thepast months: the liquid Helium closed-cycle cooler failed just before the tech-nical time (during which it was supposedto be maintained!), causing the completeloss of three nights. After its replacement,

it was discovered that part of the instru-ment was not properly cooled. This newproblem is being investigated, and willforce us to re-open the instrument soon.In the meantime observations are not dis-rupted as we use the LN2-based pre-cool-ing system.

A new GPS board arrived to replacethe one we are currently using for our timereference. In August the week counter ofthe general GPS system will reach thecritical value of 1023, then restart at 0. OurGPS, as most of those more than a fewyears old, cannot deal with this changeand will find itself back in the late 1970s.After changing a chip in our old board, wewill keep it as a spare.

From the operational point of view, thepast months have seen the implementa-tion of a detailed plan of preventive main-tenance. Every day, a scheduler providesus with a list of items to be verified and,in most cases, we just have to run a “tem-plate” which performs a series of mea-surements, and generates a plot and di-agnostics.

The NTT is still operated with a teamthat is short of one astronomer, whoshould replace Chris Lidman (who left forthe VLT) as SOFI instrument scientist.During the first six months of the year, thefellows were asked to perform extra dutydays at the telescope to ensure fullsupport of each SOFI run, but in August,we will have to offer this instrument withminimum support. We hope to recruit anadditional astronomer in the comingmonths.

Ismo Kastinen recently joined theteam as our 6th Telescope and Instru-ment Operator. This will allow us to runthe telescope most of the time with a day,a night, and a day/night operator. Thisconfiguration provides the necessarymanpower to perform the operations re-quired by the calibration and maintenanceplans.

During the coming months, we plan toupdate the EMMI and SUSI observationtemplates. As these were written at thetime of the “Big Bang” (the 1996–1997one, not the cosmological one), and asthey were the first observation templatesever written, they are far from ideal froman operation and maintenance point ofview. We hope to be able to offer the newones early next millennium.

The La Silla News Page

Figure 1: Alain Gilliotte, optician engineer at La Silla, is starting to wash the NTT main mirror,with a solution of soap and water and a sponge.

25

flow system (DFS) was implemented. Allframes obtained with the EFOSC2 in-strument are transferred via standardVLT/DMD data handling tools to a cen-tralised La Silla archive, where compactdisks are produced for each observingnight. The use of the DFS also allows usto implement a data reduction pipeline,which we plan to offer for the most com-monly used modes of EFOSC2 in thenear future. It also allows us to enhanceour instrument calibration database, andto ease data quality control checks.

Maybe the most important improve-ment in the 3.6-m that comes with the up-grade is its gain in operational stabil-ity and in the reduction of technicaldowntime. Beside the more reliable elec-tronics used, the strict implementation offormalised procedures and checklistsfor daily start-ups, instrument set-ups andchange-overs lead to a constant and sig-nificant decrease in downtime, compar-ing the 7% in 1998 (corresponds to 140hours of total available observing time lostdue to technical problems) with the 4%in 1999 (January–July, 60 h). In view ofthese numbers we ask the visiting as-tronomer to forgive us the inconvenienceof finding the 3.6-m telescope buildinglocked (like the NTT), which requires tocall the operator on duty to grant accessto the building. But also this action helpsus to keep configuration control.

Finally we are happy to announce anew, very detailed EFOSC2 manual, in-cluding a description of the “neo-classi-cal” observing strategy. It is available fromthe 3.6 Web pages (http://www.ls.eso.org.lasilla/Telescopes/360cat/efosc/html/doc/EFOSC2manula.ps.gz), and we invite

3.6-m Telescope Control System Upgrade CompletedM. STERZIK, U. WEILENMANN, and the 3.6-m Upgrade Team

New highlights can be reported fromthe 3.6-m upgrade project: the full VLT-compliant Telescope Control System(TCS) has been finally tested and com-missioned during recent technical-timeperiods. On August 9th it has been suc-cessfully put into routine operation. Let usbriefly recall the milestones passed. The"heart" of the telescope, the servo con-trol system for the axis, was completelymodified. The telescope control is nowbased on so-called “Local Control Units”(LCUs), an identical hardware platform tothat used at the VLT. It substitutes theover 20 years old TCS running onHP1000 computers. In contrast to the VLTand NTT, the 3.6-m is an equatorialmounted telescope, and the entire con-trol SW had to be modified and adaptedfor controlling the alpha and delta axis.With the broad usage of standardised ar-chitectures and software concepts at the3.6-m, which were originally designed anddeveloped for the VLT, we intend to gainoperational stability, quality, and main-tainability. However, it also requires to re-place proven components like the old, in-cremental encoders with optical strip en-coders which are required to read thetelescope position. The new encoders aresupposed to deliver a five times higherresolution, but once we tested the per-formance of them in practice, we en-countered an unexpected problem: in ex-treme telescope positions, implying sev-eral tens of tons of weight on the axis, thetight tolerances (±100 mm) required forreliable readings of the position were ex-ceeded, thus loosing control over the tele-scope. Through careful adjustment of theheads we were able to restrict the “dead

zone” to a very small region in theNorthwest. In September, a dynamicalmount for the reading heads will be in-stalled, which is expected to completelysolve this problem.

Work is also proceeding on other sys-tem components: The manual telescopecontrol and the interlock system is beingupgraded and adapted to the new con-trol environment. The Infrared configu-ration (f/35) is being prepared for the ar-rival of TIMMI2 early next year and theCES refurbishment and upgrade is duein October. There are still minor issuesto be clarified in order to fully understandthe behaviour of the telescope. To main-tain the already excellent image qualityof less than 0.6 arcsec, both, operationaland maintenance procedures have to beimplemented and also an improvementof the f/8 top-end is presently being stud-ied. All in all, close to 40 FTEs will havebeen invested into the upgrade of the 3.6-m telescope, and La Silla has gained vastexperience in operating and maintainingVLT-era telescopes.

What do all these upgrades mean tothe user, the visiting astronomer? First ofall, the pointing and tracking performanceis dramatically improved: a remarkablepointing accuracy of 5 arcsec rms is nowachieved, and the drift rate is well below1 arcsec per 10 minutes during freetracking, a value that was out of reach forthe old system. In combination with thenew TCS, the entire VLT-compliant soft-ware presently running in the 3.6-m wasupgraded to the VLT-FEB99 CommonControl Software release, implying fullY2K compliance of the system. With thehelp of DMD the archiving part of the data

3.6-m Telescope Blind Tracking PerformanceFirst Detection of the Methane Field Brown Dwarf SDSS 162414.37+ 002915.6 in i-Band

all EFOSC2 ob-servers to con-sult this manualin advance oftheir observa-tions.

Without theconstant enthu-siasm and per-sonal commit-ment of all thepeople involved,in particular thetechnical sup-port teams at LaSilla, the 3.6-mupgrade projectwould neverhave reachedthe current levelof quality andsuccess.

26

R E P O R T S F R O M O B S E R V E R S

Deconvolving Spectra of Lensing Galaxies, QSO Hosts, and More . . . F. COURBIN1,2, P. MAGAIN2*, S. SOHY2, C. LIDMAN 3, G. MEYLAN 4

1Universidad Católica de Chile, Departamento de Astronomía y Astrofísica, Santiago, Chile2Institut d’Astrophysique et de Géophysique, Liège, Belgium3European Southern Observatory, Chile4European Southern Observatory, Garching*Also Maître de recherches au FNRS (Belgium)

1. Background and Motivation

High spatial resolution undoubtedlyplays a key role in most major advancesin observational astrophysics. In thiscontext, considerable effort has beendevoted to the development of numericalmethods aimed at improving the spatialresolution of astronomical images.However, the most commonly used tech-niques (e.g., Richardson 1972, Lucy1974, Skilling & Bryan 1984) tend to pro-duce the so-called “deconvolution arte-facts” (oscillations in the vicinity of highspatial frequency structures) which alterthe photometric and astrometric proper-ties of the original data. Recently, Magain,Courbin & Sohy (1998ab; hereafter MCS)proposed and implemented a new de-convolution algorithm which overcomessuch drawbacks. Its success is mainly theconsequence of a deliberate choice toachieve an improved resolution ratherthan an infinite one, hence avoiding re-trieving spatial frequencies forbidden bythe sampling theorem.

Many successful applications of theMCS algorithm have been carried out in

the framework of an intensive effortto obtain detailed light/mass maps forlensing galaxies (e.g., Courbin et al.1997, 1998ab, Burud et al. 1998abc).These results, when compared with re-cent Hubble Space Telescope (HST)images, clearly demonstrate the effec-tiveness of the method in producing re-liable high-resolution images.

High angular resolution is certainlya must in imaging, but it is also ofgreat interest in spectroscopy. Fromrealistic simulations, we illustrate herea promising spectroscopic version ofthe MCS algorithm. We demonstratethat flux-calibrated spectra of severelyblended objects can be accurately ex-tracted and show how the algorithmcan be used to decontaminate thespectra of extended objects from thelight of very nearby bright point sources.The preliminary deconvolution of near-IRNTT (SOFI) spectra of the lensed QSOHE 1104-1805 are presented as an ex-ample of an application to real data. Thefaint spectrum of the lensing galaxy isclearly unveiled by the deconvolution al-gorithm.

2. Spatial Deconvolution of Spectra

MCS have shown that sampled imagesshould not be deconvolved with the ob-served Point Spread Function (PSF), butwith a narrower function, chosen so thatthe final deconvolved image can be prop-erly sampled, whatever sampling step isadopted to represent it. For this purpose,one chooses the final (well-sampled)PSF of the deconvolved image and com-putes the PSF which should be used toperform the deconvolution and which re-places the observed total PSF.

Depending on the resolution which isaimed at, the width of such a profile maybe anything between “slightly narrower”than the observed PSF or seeing (this re-sults in a very high but not infinite gain inresolution), and critically sampled width,which would result in a modest gain in res-olution. A straightforward consequence ofchoosing the shape that the PSF will havein the deconvolved image is that it is, in-deed, exactly known. Such prior knowl-edge can be used to decompose the data(image or spectrum) into a sum of point

Figure 1: Two possible instrumental set-ups to obtain simultaneously the spectra of the object of scientific interest (a lensed QSO in the presentexample) and that of one or more PSF stars. Left: long slit spectroscopy. Right: Multi Object Spectroscopy. The latter allows one to observe sev-eral PSF stars, whatever Position Angle is selected to observe the target.

27

sources with known analytical spatialprofile, and deconvolved numerical back-ground. This decomposition was suc-cessfully used in the MCS image decon-volution algorithm. We applied the samefundamental rule to construct an algorithmfor the spatial deconvolution of spectra,as described in detail in Courbin, Magain,Kirkove & Sohy (1999).

As with the image deconvolution al-gorithm, precise knowledge of the in-strumental profile is required. This profilevaries with wavelength and time. It maybe obtained in practice by aligning si-multaneously on the slit, both the objectof scientific interest and one or severalpoint sources. Whenever the target is mul-tiple (this is indeed likely if deconvolutionis required), no reference point sourcemay be observable on the same slit, asthe position angle of the spectrograph isin such a case imposed by the geometryof the target. This is no more true whenobserving in MOS mode, where severalPSF stars can be observed at the sametime and used to improve the spatial sam-pling of the spectra, to reach even high-er spatial resolution and to achieve moreaccurate point/extended sources de-composition. Then, finding suitable PSFstars is not more a limitation in spec-troscopy than in imaging. Figure 1 showstwo examples of instrumental set-upswhich may be adopted to yield observa-tions well suited to the problem of spec-tra deconvolution.

In order to minimise the effect of thePSF variation across the field, PSF starsshould be selected as close to the maintarget as possible. Observing severalsuch point sources will not only allow over-sampling of the deconvolved spectra, butalso to check whether a given referencestar is multiple or not, and to check thePSF stability over the field. Even ifthe algorithm can cope with criticallysampled data, detectors with small pixelsizes should be preferred (e.g., the high-resolution collimator of FORS1/FORS2provides 0.1″ pixel, which is excep-tionally small for a spectrograph), andhigh signal-to-noise ratio aimed at. Theabove conditions are in fact the same asin imaging. Good sampling, PSF stabili-ty and high signal-to-noise are definitelythe main requirements when performingphotometrically (spectrophotometrically inthe present case) accurate deconvolu-tions.

In practise, the deconvolved PSF(i.e., shape of the point sources in thedeconvolved spectra) can be as narrowas one may wish, the only theoretical lim-itation to the gain in resolution being thesampling of the deconvolved data. Inour implementation of the algorithm,one can improve the pixel size (in thespatial direction) by a factor 2, so thatthe maximum resolution will be 2 (half)CCD pixels FWHM. This limit can inprinciple be reached, whatever seeingis available in the data spectra, although,

as one can expect, the quality of theresult decreases as the seeing in-creases.

3. Simulations

The algorithm has two obvious appli-cations: firstly, to extract the individualspectra of strongly blended point sourcesand, secondly, to unveil the spectra of ex-tended objects hidden by much brighterpoint sources. While de-blending pointsources is relatively easy (see Courbin,Magain, Kirkove & Sohy 1999 for an ex-ample), the most interesting applicationsof our spectrum deconvolution algorithmmay consist in its capability to decomposespectra into the components from pointsources and extended sources. The twosimulated spectra presented hereafter il-lustrate the results that can be expectedfrom high signal-to-noise data.Our simulations include slit misalign-ment, seeing variation as a function ofwavelength as well as – exaggerated –atmospheric refraction. Gaussian photonnoise and readout noise are also addedto the data. A typical signal-to-noise ra-tio is 200–300 per spectral-resolutionelement. The PSF is derived from thesimulated spectrum of a point sourcewhose S/N ratio is similar to the one ofthe scientific data. The deconvolutions areperformed as in imaging and the optimalchoice of the different Lagrange mul-tipliers to be used is guided by the visu-al inspection of the residual maps (RM)(see Burud et al. 1998b for an exam-ple of a RM). The RM is the absolutevalue of the difference between the rawdata and the deconvolved image re-convolved by the (narrower) PSF, inunits of the noise. An accurate de-convolution should therefore correspondto a flat RM with a mean value of 1.

3.1 Simulated lens and QSO host galaxies

Figure 2 presents the simulated spec-tra of two lensed images of the samequasar, separated by 6 pixels in the spa-tial direction. The spatial resolution is 4pixels FWHM and the signal-to-noise ra-tio of the brightest spectrum is about200–300 per spectral resolution element,depending on the wavelength. The flux ra-tio between the two QSO images is 3 (1.2magnitudes). The spectrum of the – ex-tended – lensing galaxy is also incorpo-rated in the simulated data. It is locatedonly 2 pixels away from the centroid of thefaintest QSO image and is about 3 to 5magnitudes fainter than the QSO images(depending on wavelength) and thereforecompletely invisible in the raw data (seeleft panel of Figure 2). The second pan-el shows the result of the deconvolution,the third panel displays the 2-D decon-volved spectrum of the lensing galaxyalone, which compares very well with theoriginal simulated spectrum shown inthe fourth panel. The RM, as defined atthe beginning of this section, is displayed

(a) (b) (c) (d) (e)

Figure 2: From left to right: (a) Simulated spectrum of a lensed quasar (~ 4000–8000 Å), (b) itsdeconvolution, (c) the deconvolved background (here, the lensing galaxy alone), (d) the galaxyused to construct the simulation, and (e) the residual image (data minus model) in units of thephoton noise. The simulation considers seeing variation with wavelength and (exaggerated) at-mospheric refraction.

28

in the last panel and does not show anysignificant structure, as expected for agood deconvolution.

Figure 3 confirms the good results ob-tained in Figure 2. The 1-D spectra of the2 QSO images as well as the spectrumof the lensing galaxy are in very goodagreement with the input spectra (solidline), in spite of the blending and highluminosity contrast. The emission line inthe spectrum of the lensing galaxy is wellrecovered and its (spectral) position isretrieved with an accuracy of 0.1 pixel.

A more difficult application may con-sist of the spectroscopic study of QSOhost galaxies. In this case, a bright QSOis centred almost exactly on the bulge ofa much fainter host galaxy, which cansometimes be imaged with standardtechniques. We have tested the spec-troscopic version of the MCS algorithm onsuch a difficult case, and show from theresults in Figure 3 that the spectrum of ahost galaxy can be accurately deconta-minated from light pollution by its centralAGN. The host galaxy in our example is

about 2 magnitudes fainter than the QSOand has the spatial extension of a low red-shift galaxy (z = 0.1–0.3).

3.2 A near-IR spectrum of the lens-ing galaxy in HE 1104–1805

We recently obtained with SOFI,mounted on the NTT, near-IR spectra ofthe so-called “double Hamburger”, i.e., thedoubly imaged gravitationally lensed QSOHE 1104–1805 (Wisotzki et al. 1993), dis-covered in the Hamburg survey for bright

Figure 3: Left: The top panel shows the 1-D deconvolved spectra of the two simulated QSO images. In the middle and bottom panels are dis-played the flux ratios between the two spectra and the original QSO spectrum used in the simulation. Right: the spectrum of the lensing galaxyalone (top) and its division by the input spectrum (bottom). The position of the emission line is retrieved with an accuracy of 0.1 pixel.

Figure 4: Left: The two left panels display a 2-D simulated spectrumof a QSO and its host galaxy. The two right panels compare the 2-D deconvolved spectrum of the host galaxy alone, with the spectrumused in the simulation. Right: 1-D deconvolved spectra of the QSO(top) and of the host galaxy (bottom). In each case, the solid line in-dicates the 1-D original spectrum used to make the simulated data.

29

QSOs and for which the time delay hasbeen reported (Wisotzki et al. 1998).However, the redshift of the main lens-ing galaxy remains unknown, which stillhampers any attempt to estimate H0from this system. As the lensing galaxyis very red (e.g., Courbin et al. 1998b),near-IR spectroscopy appeared an ob-vious way of determining its redshift.

SOFI spectra were obtained at the NTTbetween 0.9 and 2.5 microns under goodseeing conditions (about 0.6–0.7″) anddeconvolved down to a resolution of0.3″. As no PSF star was observed, thePSF was computed directly from thespectra of the two QSO images (more de-tails about this procedure are given inCourbin et al. 1999). Figure 4 shows theresults of the deconvolution performed be-tween 1.5 and 2.5 microns, since the lensis too faint in the 0.9–1.5 micron region.The original data are shown in the toppanel of Figure 4. The three other pan-els, from top to bottom, show the de-convolved spectra with improved sam-pling (0.145″/pixel) in the spatial direction(this is why the width of the deconvolved2-D images is twice that of the raw data),the spectrum of the lensing galaxy, as wellas the RM (flat with a mean value of 1).The data presented here have a much

lower S/N than in the simulations but aresufficient to unveil the spectrum of the faintlens. Since no emission line is detected,we are still left without precise measure-ment of the lens redshift. However, theprospect of measuring this redshift withoptical/near-IR spectroscopy is good:with the improved S/N, sampling, andspectral resolution of future VLT data!

4. Future Prospects

Our new extension to spectroscopy ofthe MCS image deconvolution algorithmobviously has a wide field of applications.The most original and promising onesmay consist in spectroscopic studies in-volving extended objects hidden by – of-ten brighter – point sources. We have pre-sented an application of the method togravitationally lensed quasars, where thespectrum of a very faint lensing galaxy canbe extracted. A similarly interesting ap-plication will be to take full advantage ofthe ability of the algorithm to decomposespectra, in order to carry out the first sys-tematic spectroscopic study of quasarhost galaxies. With current instrumenta-tion mounted on 8–10-m-class tele-scopes, sufficiently high signal-to-noisespectra can be obtained, at least for low-

redshift quasars, in order to derive pre-cise rotation curves of their host galaxy,provided the spectrum of the bright QSOnucleus can be removed accurately. Thepresent spectrum deconvolution algo-rithm is very well suited for such a pur-pose and may therefore allow significantprogress, not only towards the mea-surement of the mass of the centralblack hole in QSO host galaxies, but alsoin astrophysics in general. We intend torelease in the near future a documentedpublic version of our image deconvolutionalgorithms, followed latter by the spec-troscopic version of code illustrated in thepresent article.

Acknowledgements

F. Courbin acknowledges financialsupport through the Chilean grantFONDECYT/3990024. S. Sohy is sup-ported by contracts ARC 94/99–178“Action de Recherche Concertée de laCommunauté Française (Belgium)” and“Pôle d’Attraction Interuniversitaire” P4/05(SSTC Belgium).

References

Burud, I., Courbin, F., Lidman, C., et al.1998a, ApJ, 501, L5.

Burud, I., Courbin, F., Lidman, C., et al.1998b, The Messenger 92, 29.

Burud, I., Stabell, R., Magain, P., et al. 1998c,A&A, 339, 701.

Courbin, F., Magain, P., Keeton, C.R., et al.1997, A&A, 324, L1.

Courbin, F., Lidman, C., Frye, B., et al. 1998a,ApJ, 499, L119.

Courbin, F., Lidman, C., Magain, P., 1998b,A&A 330, 57.

Courbin, F., Magain, P., Kirkove, M., Sohy, S.,1999, ApJ, in press.

Courbin, F., Lidman, C., Meylan, G., Magain,P., 1999, in preparation.

Lucy, L. 1974, AJ, 79, 745.Magain, P., Courbin, F., & Sohy, S. 1998a, ApJ,

494, 472.Magain, P., Courbin, F., & Sohy, S. 1998b, The

Messenger 88, 28.Richardson, W.H.J. 1972, J. Opt. Soc. Am.,

62, 55.Skilling, J., & Bryan, R.K. 1984, MNRAS, 211,

111.Wisotzki, L., Koehler, T., Kayser, R., Reimers,

D., 1993, A&A 278, L15.Wisotzki, L., Wucknitz, O., Lopez, S., Soren-

sen, A, 1998, A&A 339, L73.

Figure 5: From top to bottom: A near-IR spectrum of the lensed QSO HE 1104–1805, between1.5 µm (left) and 2.5 µm (right). The seeing is 0.6–0.7 arcsec and the lensing galaxy is situatedabout 0.5″ away from the brightest QSO image. The spectrum is deconvolved down to a reso-lution of 0.3″ with improvement of the sampling in the spatial direction. The third panel displaysthe spectrum of the lens alone and the last panel shows the residual map (RM, see text) in unitof the photon noise, to illustrate the good agreement between the data and their deconvolution.The dark areas are not used, because of strong atmospheric absorption.

Emission-Line-Object Survey in the LMC with theWFI: New Faint Planetary Nebulae P. LEISY1, P. FRANCOIS1,2, and P. FOUQUÉ1,3

1European Southern Observatory, Santiago, Chile; 2Observatoire Meudon DASGAL, Paris, France3Observatoire Meudon DESPA, Paris, France

1. Introduction

Until now, all emission-line-object sur-veys in the two Magellanic Clouds (MCs)

have been made with Schmidt plates, ei-ther with prism objective or with the 2-fil-ter (ON, OFF) method. The only deep sur-vey, dedicated to the central part of the

MCs, dates from 1980, and suffers fromlarge biases because it was done usingonly two filters to select the candidates(about 30% of false candidates). No

30

large emission-line survey has been pub-lished with CCD devices until now.

Our final goal is a large survey (10square degrees in total) of two regions inthe SMC and the LMC, to identify newfainter emission-line objects (as PNe, HIIregions, Be stars, symbiotic stars, youngstars, etc.).

We concentrate our efforts mostly onPNe and HII regions. The new candidateswill allow us to derive accurately the fullPNe luminosity function in the MagellanicClouds (MCs), which is still poorly knownfor faint objects.

Imaging with the WFI will hopefully befollowed up with EFOSC2 and/or FORSto take, for the first time, spectra of thenew very faint PNe but also the close-byHII regions, in order to determine thephysical properties and accurate abun-dances of these emission-line objects.

2. Overview

LMC and SMC are ideal targets for atleast three reasons:

– they are close enough so that we canhave a good spatial resolution and verygood S/N spectra

– their PNe and HII regions are brightenough so that a detailed analysis of theirchemical composition is feasible with thecurrent instrumentations.

– the metallicities are lower than ourGalaxy

The study of all types of PNe (with dif-ferent brightness, age and metallicity) isone of the cornerstone of our under-standing of Intermediate-Mass Stars andthe chemical evolution of galaxies. PNeare used as probes to determine the pastabundances of the progenitor stars andalso the current abundances of the lightelements released at the end of theAGB.

The faintest PNe are also of interestbecause they contain two important class-es of objects. Either they are old objects(low mass with longer lifetime) which gives

information on the chemical compositionat the early stages of the galaxy (to tracethe chemical evolution over the past 10billion years) or they are objects at thevery end of their evolution in a state whichallows us to look into deeper layers of theirenvelope.

To improve our understanding of thechemical evolution of the nearby galax-ies, it is crucial to use the largest spatial,temporal and metallicity coverages. ThePNe will help us in our understanding ofthe enrichment processes in He, C, N inthe Intermediate-Mass stars, becausethey tell us about the dredge-up mecha-nisms occurring during the last stages ofevolution of intermediate-mass stars andhow they depend on the metallicity.Helium, Nitrogen, and Carbon are, directlymanufactured by the PNe's star and theirabundances thus provide the yield ofthose elements in the correspondinggalaxy. In particular, we have shown inthe Magellanic Clouds (Leisy and Den-nefeld,1996), in a sample of 70 objects,that enrichment or destruction are strong-ly metallicity dependent. We obtain metal-licity relations, but they are still hamperedby the too few studied objects, particularlyin the very low metallicity range. TheNitrogen and Carbon production are byfar more effective when the initial metal-licity is low. Apparently, in the SMC, thelower-mass stars produce also Oxygen,but our results are based on a small sam-ple (8 objects).

These abundances have also been di-rectly compared with those of adjacent HIIregions which provide “recent” abun-dances. Until now, only very old, inho-mogeneous, data are available in somegalaxies, and moreover, the number ofstudied HII regions is always too small fora detailed comparative study. They comefrom various authors, are of poor quality(the forbidden emission line [OIII] λ4363Åis often missing or its uncertainty isgreater than 50%). Moreover, in a givengalaxy, the metallicity determined from HIIregions shows a large dispersion. Themetallicity determinations have mostlybeen based on the abundances ofOxygen because it is easily measured byoptical spectroscopy. However, it may notbe representative of the metallicity of theobject since the ejection of various ele-ments from stellar wind, Supernovaeand Planetary Nebulae involves differenttimes and length scales. The existenceof spatial inhomogeneities in the distrib-ution of various chemical elements with-in a HII region is still a matter of debate.

As the information coming from PNeand HI regions is by far uncomplete andbiased to the brightest objects, a surveyof PNe's and HII regions in nearby galax-ies is necessary to derive proper chem-ical abundances and check the differentelements as metallicity indicators.

3. Observations

The main advantage of using WFI overexisting “uncomplete surveys” is that itprovides a high spatial resolution (pixelsize 0.24″) which is mandatory in suchcrowed regions. Moreover a large deepsurvey of emission-line objects in the MCscan be done covering several degrees.Our data provide at the same time: ac-curate positions (~0.2″) catalogue andfinding charts, and the absolute fluxes.Nevertheless, this observation step isneeded to prepare further observationswith bigger or space telescopes, which re-quire accurate positions and photometryin order to save as much observing timeas possible.

Three nights were allocated at the be-ginning of January 1999, but because the2.2-m WFI was offered in February withdelay to the community, we got twonights in March and could only observethe LMC.

The two nights (March 3rd and 4th,1999) were photometric with an averageseeing of about 1.0″ (from 0.8″ to 1.3″)at airmasses between 1.3 to 1.9.

We observed three spectrophotomet-ric standard stars per night in order to cal-ibrate the images in absolute fluxes.

We were able to observe two fields pernight, and therefore cover a full squaredegree on the sky. We observe each LMCfield in Hα, [OIII], Continuum and B forcentring purposes (and further findingcharts). Each exposure was divided in 5jittered exposures (total time between 20and 30 min) in each filter (1 min in B). Thisprocedure was adopted to avoid spuriousidentifications, and to fully cover thegaps between the 8 CCDs. Therefore, foreach field, taking into account the over-heads of focusing, pointing and reading(20 ×) of the CCD, an amount of ~ 2 h 30min was needed.

4. Reductions

For interferencial imaging, the moonbackground is not crucial because aftera 5-min exposure time, even with fullmoon, the sky level is of the order of150–200 ADUs.

Figure 1: PN1 in Hα and [O III]. Note the highexcitation object (ratio [O III]/Hα = 10).

Figure 2: PN 2 in Hα and [OIII]: extended ob-ject ~ 2.5″ (or 0.6 pc).

Figure 3: PN 3 in Hα and [OIII] Low excitationobject.

Figure 4: PN 4 in Hα and [O III]: Bipolar ex-tended object ~ 5″ in Hα, but more compactin [OIII].

31

Preliminary reductions of the WFI datahave been made, but a powerful computerwas still missing at that time. Rememberthat a full frame is 300 Mb!

Here are the different reduction stagesadopted:

– Bias substraction and FF division– Cosmic-event removing– Combine the 5 images per filter (not

yet done)– Frame recentring (Continuum as

reference)– Flux calibration– Continuum frame substraction– ([OIII]-Continuum) divided by (Hα-

Continuum)– Accurate astrometry using the ESO

Starcat package facility (not yet done)

5. Results

These observations demonstrate that,as expected, we still can find a lot of newemission-line objects. Moreover, theseobservations provide, in the survey area,an unbiased and complete survey (rangegreater than 8 magnitudes).

As a first product this study, preliminaryto further spectroscopic study, will lead tothe publication of a catalogue with precisecoordinates and good finding charts ofmore than 1000 emission objects (PNe,HII regions, SNR, Hα-emission stars,etc.) per square degree.

The performance of the WFI is very en-couraging:

– Firstly, we were able to (easily) findthe 6 known PNe present in our 4 fields,and moreover our photometry agreed

within 10–20% with the fluxes derivedspectroscopicaly.

– Secondly, with the calibration in flux-es, we confirm that we easily reach flux-es down to Fα = 10–16 erg.cm–2 s–1, andtherefore that we are able to find objects20–50 times fainter than the last deepsurvey (Jacoby 1980).

With a bigger area covered and bet-ter statistics, we will derive an unbiasedPNe luminosity function.

In each of the two first 30′ by 30′ re-duced fields we discovered more than100 emission-line objects. In one field wefound 2 new PNe, 8 compact HII regionsand ~ 80 Hα-emission-line stars. In theother field, also 2 new PNe, 4 compactHII regions and ~140 Hα- emission linestars.

All 4 PNe are very faint and 2 of themare fully resolved (diameter > 1.0″). ThePNe have fluxes of the order of 10 10–16

erg.cm–2.s–1, which one can compare tobrigthest (3000 10–16 erg.cm–2.s–1) or thefaintest (~ 200 10–16 erg.cm–2.s–1) PNealready known.

Figures 1 to 4 show the finding chartsin both Hα and [OIII] filters.

5.1. Expected PNe number in theLMC

In 1978 Sanduleak et al. have esti-mated the total PNe number to 400 in theLMC (and 100 in the SMC accordingly tothe LMC/SMC mass ratio equal to 4). In1980 Jacoby determined these num-bers to be 996 and 285, respectively.

The current number of known PNe inthe Magellanic Clouds are respectively

282 and ~ 85 for the LMC and the SMC.This study demonstrates that many

PNe have been missed by the previoussurveys and that probably a few hundredcan be discovered with the current in-strumentation.

In the studied area we double theknown sample of PNe, therefore, wecan extrapolate to find more than 250 newPNe in the LMC. Statistically, we can atleast expect to find about 50–100 newfaint and extended PNe in the Bar. ThesePNe are of particular interest becausethey are old, allowing to determine galaxyabundances at epoch up to 10 billionsyears.

6. Conclusion

The WFI has proved with this study thatit is an excellent efficient instrumentwhich can be used to survey large areas(like nearby galaxies) and still discovermany faint interesting objects.

We hope that we will soon be able tocover several other square degree anddetermine an accurate PNe luminosityfunctions over this complete and unbiasedsample.

The next step, with such faint objects,will be a follow-up with VLT telescopes(FORS) to obtain very good S/N spectraand derive accurate abundances.

References

Jacoby, G.H., 1980 ApJS 42, 1.P. Leisy, M. Dennefeld, 1996 A&AS 116, 95.Sanduleak, N., MacConnell, D.J., & Phillip,

A.G.D., 1978, PASP 90, 621.

A N N O U N C E M E N T S

International Staff (1st July 1999 – 30th September 1999)

ARRIVALS

EUROPE

ALEXOV, Anastasia (USA), Science Data Analyst/ProgrammerBALESTRA, Andrea (I), Software EngineerCESARSKY, Catherine (F), Director GeneralCHRISTENSEN, Lars (DK), HST Public Outreach ScientistDELPLANCKE, Françoise (B), PhysicistDÜHRING, Margit (DK), Associate ST-ECFFEDRIGO, Enrico (I), Software EngineerGENNAI, Alberto (I), Control/Hardware EngineerGORSKI, Krzyszof (PL), Archive AstronomerIZZO, Carlo (I), Scientific Applications DeveloperKIM, Tae-Sun (ROK), FellowMENARDI, Serge (F), Opto-Mechanical EngineerPALSA, Ralf (D), Astronomical Data Analysis SpecialistPIRANI, Werther (I), VLT Software System ManagerPIRARD, Jean-François (B), Infrared Instrumentation Engineer

POZNA, Eszter (I), Software EngineerTORWIE-SCHMER, Claudia (D), Accounting AssistantWILHELM, Rainer (D), Engineer/Physicist

CHILE

CHADID, Merieme (MA), FellowGILLET, Gordon (D), Electronics EngineerJONES, Heath (AUS), FellowLE FLOC’H, Emeric (F), Coopérant La Silla NTT

DEPARTURES

EUROPE

CONTARDO, Gertrud (D), StudentIBATA, Rodrigo (GB), FellowJANDER, Georg (D), Mechanical Engineer

CHILE

PIZZELLA, Alessandro (I), FellowSBAIHI, Marc (F), Mechanical Engineer

PERSONNEL MOVEMENTS

32

ESO, the European Southern Observatory,was created in 1962 to "… establish andoperate an astronomical observatory in thesouthern hemisphere, equipped with pow-erful instruments, with the aim of further-ing and organising collaboration in as-tronomy …" It is supported by eight coun-tries: Belgium, Denmark, France, Germany,Italy, the Netherlands, Sweden andSwitzerland. ESO operates at two sites. Itoperates the La Silla observatory in theAtacama desert, 600 km north of Santiagode Chile, at 2,400 m altitude, where sev-eral optical telescopes with diameters upto 3.6 m and a 15-m submillimetre radiotelescope (SEST) are now in operation. Inaddition, ESO is in the process of buildingthe Very Large Telescope (VLT) on Para-nal, a 2,600 m high mountain approximately130 km south of Antofagasta, in the driestpart of the Atacama desert. The VLT con-sists of four 8.2-metre and three 1.8-me-tre telescopes. These telescopes can alsobe used in combination as a giant inter-ferometer (VLTI). The first 8.2-metre tele-scope (called ANTU) started regular sci-entific operations in April 1999 and also thesecond one (KUEYEN) has already deliv-ered pictures of excellent quality. Over 1200proposals are made each yea r for theuse of the ESO telescopes. The ESOHeadquarters are located in Garching,near Munich, Germany. This is the scien-tific, technical and administrative centre ofESO where technical development pro-grammes are carried out to provide the LaSilla and Paranal observatories with themost advanced instruments. There arealso extensive astronomical data facili-ties. In Europe ESO employs about 200 in-ternational staff members, Fellows andAssociates; in Chile about 70 and, in ad-dition, about 130 local staff members.

The ESO MESSENGER is published fourtimes a year: normally in March, June,September and December. ESO also pub-lishes Conference Proceedings, Preprints,Technical Notes and other material con-nected to its activities. Press Releases in-form the media about particular events. Forfurther information, contact the ESOEducation and Public Relations Departmentat the following address:

EUROPEAN SOUTHERN OBSERVATORY Karl-Schwarzschild-Str. 2 D-85748 Garching bei München Germany Tel. (089) 320 06-0 Telefax (089) 3202362ips@eso.org (internet) URL: http://www.eso.org

The ESO Messenger:Editor: Marie-Hélène DemoulinTechnical editor: Kurt Kjär

Printed by J. Gotteswinter GmbHBuch- und OffsetdruckJoseph-Dollinger-Bogen 22D-80807 MünchenGermany

ISSN 0722-6691

Local Staff (Chile)(Until 30th September 1999)

ARRIVALSNAVARRETE, Julio, Telescope Instrument Operator, Paranal (transferred from La Silla)CARRASO, Cecilia, Personnel Assistant, SantiagoZAPATA, Joel, Warehouse Administrative Assistant, ParanalCARVAJAL, Marta, Administrative Secretary

DEPARTURESMUÑOZ, Nelson, Electrónico, La SillaPARADA, María, Recepcionista, SantiagoVILLANUEVA, Raúl, Ingeniero Civil, Paranal

ContentsTELESCOPES AND INSTRUMENTATIONM. Tarenghi: News from the VLT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1E. Ettlinger, P. Giordano, and M. Schneermann: Performance of the VLT Mirror

Coating Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4M. Sarazin and J. Navarrete: Climate Variability and Ground-Based Astronomy:

The VLT Site Fights Against La Niña . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8C. Castillo F. and L. Serrano G.: Analysis of the Anomalous Atmospheric

Circulation in Northern Chile During 1998 . . . . . . . . . . . . . . . . . . . . . . . . . . . 10M. Ferrari and F. Derie: Variable Curvature Mirrors . . . . . . . . . . . . . . . . . . . . . . 11F. Derie, E. Brunetto, and M. Ferrari: The VLTI Test Siderostats Are Ready for

First Light . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12J.C. Christou, D. Bonaccini, N. Ageorges, and F. Marchis: Myopic Deconvolution

of Adaptive Optics Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14Latest News: “First Light” for VLT High-Resolution Spectrograph UVES . . . . . . 22Erratum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

THE LA SILLA NEWS PAGE

O. Hainaut and the NTT Team: News from the NTT . . . . . . . . . . . . . . . . . . . . . 24M. Sterzik, U. Weilenmann and the 3.6-m Upgrade Team: 3.6-m Telescope

Control System Upgrade Completed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

REPORTS FROM OBSERVERSF. Courbin, P. Magain, S. Sohy, C. Lidman, G. Meylan: Deconvolving Spectra

of Lensing Galaxies, QSO Hosts, and More… . . . . . . . . . . . . . . . . . . . . . . . 26P. Leisy, P. Francois, and P. Fouqué: Emission-Line Object Survey in the LMC

with the WFI: New Faint Planetary Nebulae . . . . . . . . . . . . . . . . . . . . . . . . . 29

ANNOUNCEMENTSPersonnel Movements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31List of Scientific Preprints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

List of Scientific Preprints(July–September 1999)

1331. C. Mouton, K. Sellgren, L. Verstraete, A. Léger: Upper Limit on C60 and C+60 Features in

the ISO-SWS Spectrum of the Reflection Nebula NGC 7023. A&A.1332. V. Doublier, D. Kunth, F. Courbin, P. Magain: POX 186: the Ultracompact Blue Compact

Dwarf Galaxy Reveals its Nature. A&A.1333. V. Doublier, A. Caulet, G. Comte: Multi-Spectral Study of a New Sample of Blue Compact

Dwarf Galaxies: II.– B and R Surface Photometry of 22 Southern Objects. A&A.1334. Th. Rivinius, S. Štefl, D. Baade: Central Quasi-Emission Peaks in Shell Spectra and the

Rotation of Disks of Be Stars. A&A.1335. F. Marchis, H. Boehnhardt, O.R. Hainaut, D. Le Mignant: Adaptive Optics Observations

of the Innermost Coma of C/1995 O1. Is there a “Hale” and a “Bopp” in Comet Hale-Bopp? A&A.

1336. P. François, D. Briot, F. Spite, J. Schneider: Line Profile Variation and Planets Around51 Peg and υ And. A&A.

1337. F. Comerón and J. Torra: A Near Infrared Study of the HII/Photodissociation Region DR18 in Cygnus. A&A.

1338. S. Arnouts et al.: Measuring and Modelling the Redshift Evolution of Clustering: the HubbleDeep Field North. M.N.R.A.S.

1339. G. De Marchi, F. Paresce, L. Pulone: The Mass Function of Main Sequence Stars in NGC6397 from Near IR and Optical High Resolution HST Observations. ApJ.

1340. F. Comerón: Vertical Motion and Expansion of the Gould Belt. A&A.