The Messenger 11420

TTHE HARPS PROJECT wasborn in May 1998 when ESOissued an Announcement ofOpportunity asking for thedesign, construction, and pro-

curement of an instrument dedicated tothe search for extrasolar planets and aim-ing at an unequalled precision of 1 m/s. Inresponse to ESO’s call the Observatoire deGeneve together with the PhysikalischesInstitut der Universität Bern, the Observa-toire de Haute-Provence, and the Serviced’Aéronomie du CNRS have formed aConsortium which realized, in collabora-tion with ESO, this ambitious project inonly three and a half years. The instrumentwas installed by the Consortium on ESO’s3.6-m Telescope at La Silla in January 2003and first light took place on February 11th.HARPS was commissioned during theperiods 70 and 71. During period 71 a firstGTO run of seven nights had been allocat-ed for the Consortium as well. Theseobservations have already produced manyexciting results.

The instrument was handed over toESO La Silla by the end of September2003 and made available to the Communi-ty for period 72. Together with the 3.6-mtelescope HARPS is now delivering fan-tastic scientific frames tothe observers. But this isnot the only data productof HARPS. The powerfulHARPS pipeline providesany observer with extract-ed and wavelength cali-brated high-resolutionspectra, as well as the pre-cise radial velocity of theobserved star. The observ-er will thus leave theObservatory with fullyreduced spectra. Addi-tional information on howto observe with HARPScan be found athttp://www.ls.eso.org/lasilla/sciops/harps/.

AN UNEQUALLED FACILITY FORRADIAL-VELOCITY MEASUREMENTS

HARPS is a fibre-fed, cross-dispersedechelle spectrograph. Two fibres, an objectand a reference fibre, feed the spectro-graph with the light from the telescope andthe calibration lamps. The fibres are re-imaged by the spectrograph optics onto amosaic of two 2k4 CCDs, where twoechelle spectra of 72 orders are formed foreach fibre. The covered spectral domainranges from 380 nm to 690 nm. The resolu-tion of the spectrograph is given by thefibre diameter and attains a value of aboutR=115,000.At this resolution each spectralelement is still sampled by 3.2 CCD pixels.A summary of the main HARPS featuresis given in Table 1. For a detailed descrip-tion of the instrument we refer to the arti-cle published in Pepe et al. (2002) and tothe HARPS Users Manual.

HARPS is an ordinary spectrographwith outstanding efficiency and spectralresolution. The main characteristics ofHARPS is however its extraordinary sta-bility. The ThAr-reference technique isable to measure and correct the tiniestinstrumental drifts; nevertheless a lot ofeffort was put into making the spectro-graph intrinsically stable, in order to avoid

SSETTINGETTING NNEWEW SSTTANDANDARDSARDS

WITHWITH HARPSHARPSBY OCTOBER 1ST, 2003, ESO'S NEW AND UNIQUE PLANET-HUNTING MACHINE HARPS (HIGH-ACCURACY

RADIAL VELOCITY PLANETARY SEARCHER) HAS BECOME OPERATIONAL. THE MEASUREMENTS MADE DURING THE

COMMISSIONING PHASE AND THE FIRST WEEKS OF OPERATION ARE OF OUTSTANDING QUALITY. IN THIS ARTICLE WE

REPORT AMONG OTHER EXAMPLES ON THE FIRST EXTRA-SOLAR PLANET DISCOVERED WITH HARPS AND ON THE

DETECTION OF TINY STELLAR OSCILLATIONS. THE RESULTS PRESENTED DEMONSTRATE THAT HARPS IS CURRENT-LY THE MOST PRECISE DOPPLER-MEASUREMENTS MACHINE IN THE WORLD. WITH THIS ACQUISITION ESO PLACES

ITSELF AT THE HEAD OF A SCIENTIFIC DOMAIN, WHOSE INTEREST HAS CONTINUED TO GROW DURING THE PAST

YEARS.

Table 1: HARPS spectrograph characteristics

Optical design fibre-fed, cross-dispersed echelle spectrograph

Technique simultaneous ThAr Reference

Number of fibres 2

Fibre aperture on sky 1 arcsec

Collimated beam diameter 208 mm

Covered spectral range 380 nm to 690 nm

Spectral resolution R=115,000Spectral format 72 echelle orders

61.44 x 62.74 mmCCD chip mosaic, 2xEEV2k4

pixel size=15µmSampling 3.2 pixels/SE Min. inter-order 33 pixels

M. MM. MAAYYOROR11, F, F. P. PEPEEPE11, , DD. Q. QUELOZUELOZ11, F, F. B. BOUCHYOUCHY22, , G. RG. RUPPRECHTUPPRECHT33, G. L, G. LOO

CCURURTTOO44, G. A, G. AVILVIL AA33, , WW. B. BENZENZ55, J, J.-L. B.-L. BERERTTAAUXUX66, , X. BX. BONFILSONFILS11, , TTHH. D. DALLALL44, , H. DH. DEKKEREKKER33, B. D, B. DELEL ABREABRE33,,WW. E. ECKERCKERTT44, M. F, M. FLEURLEURYY11, , A. GA. GILLIOILLIOTTETTE44, D, D. G. GOJOJAKAK33,,JJ.C. G.C. GUZMANUZMAN44, D, D. K. KOHLEROHLER77,,JJ.-L. L.-L. LIZONIZON33, A. L, A. LONGINOONGINOTTITTI33, , C. LC. LOOVISVIS11, D, D. M. MÉGEVÉGEVANDAND11, , L. PL. PASQUINIASQUINI33, J, J. R. REYESEYES33, , JJ.-P.-P. S. SIVIVANAN77, D, D.. SSOSNOOSNOWSKAWSKA11, , R. SR. SOOTTOO44, S. U, S. UDRDRYY11, , A. A. VVANAN KKESESTERENTEREN33, , L. L. WWEBEREBER11, U. , U. WWEILENMANNEILENMANN44

1OBSERVATOIRE DE GENÈVE, SWITZERLAND; 2LABORATOIRE D'ASTROPHYSIQUE

DE MARSEILLE, FRANCE; 3ESO GARCHING; 4ESO LA SILLA; 5PHYSIKALISCHES INSTITUT DER

UNIVERSITÄT BERN, SWITZERLAND; 6SERVICE D'AÉRONOMIE, VERRIÈRES LE BUISSON, FRANCE; 7OBSERVATOIRE DE HAUTE-PROVENCE, FRANCE

© ESO - December 2003 21

as far as possible any kind of second-orderinstrumental errors. As a consequence, wehave decided to operate the spectrographin vacuum (see Figure 1), since ambientpressure variations would have producedhuge drifts (typically 100 m/s per mbar).The operating pressure is always keptbelow 0.01 mbar such that the drifts neverwill exceed the equivalent of 1 m/s per day.

Not only the pressure but also the tem-perature is strictly controlled. In fact, atwo-stage air-conditioning around the vac-

uum vessel controls the air temperature to17!C with a long-term stability of the orderof 0.01!C. Because of the huge thermalinertia of the vacuum vessel and the excel-lent thermal insulation between spectro-graph and vessel provided by the vacuum,the short-term temperature stability obtai-ned is even better. Over one day we havemeasured variations of the order of 0.001Krms.An example of the excellent control isgiven by the behaviour of the echelle grat-ing temperature plotted in Figure 2a.

The extraordinary stability translatesdirectly into the stability of the radialvelocity measurements. This is mostimpressively shown by the measurementseries presented in Figure 2b. Both fibreshave been exposed repetitively with theThAr calibration light and the drift –expressed in m/s – was computed as a func-tion of time. During several hours the totaldrift remains well below 1 m/s and themeasurement is dominated by “noise”. Thisnoise is introduced by the CCD whosetemperature varies by ±0.02 K and pro-duces microscopic dilatation of the chip. Infact, both fibres show the same behaviourand when subtracting one fibre from theother this “noise” disappears and weobtain the dispersion values expected fromthe photon-noise level of the simultaneousThAr reference. These measurementsprove that the simultaneous ThAr refer-ence technique is perfectly able to trackinstrumental drift at a level of 0.1 m/s rms.

HARPS has been optimised for radial-velocity efficiency, i.e. to obtain in theshortest possible exposure time the mostprecise radial-velocity measurement. How-ever, this requirement is not necessarilysynonymous with optical efficiency. Forexample, spectral resolution is an impor-tant factor in reducing the photon noise ofthe radial-velocity measurement but isoften in competition with the optical effi-ciency of the spectrograph or with the sizeof the instrument and the telescope. Onthe other hand the instrumental errorsmust also be low, at least as low as the goalfor the photon-noise equivalent. The

Figure 1:The HARPS spectrograph inside the air-conditioned room at La Silla just before beingclosed and evacuated.

Figure 2: a) Tempera-ture of the echellegrating over 24 hoursrecorded at two differ-ent days. b) Drift ofthe spectrograph dur-ing a ThAr series ofseveral hours provingthat the spectrographis extremely stableand that the simulta-neous ThAr referenceis able to track andcorrect tiniest instru-mental drifts.

Figure 3: a) Total effi-ciency of HARPS(right-hand ordinate)and efficiency of singlecomponents (left-handordinate). b) Compari-son of the measuredSNR with the valuescalculated by means ofthe Exposure TimeCalculator. For com-parison we performeda 71s exposure onHD10700 (mv=3.5) atan airmass of 1.1. TheDIMM monitor wasindicating a seeing of0.9 arcsec.

The Messenger 11422

design of HARPS is consequently theresult of an accurate trade-off between allthe relevant factors. Despite various com-promises regarding for example the fibrediameter (1 arcsec on the sky) and thefibre-feed itself, the optical efficiencyobtained is remarkable. Figure 3a showsthe calculated total optical efficiency ofthe instrument and some of its subsys-tems. These values have been adopted inthe current version of the HARPS-Expo-sure Time Calculator (ETC) which servesto estimate the SNR obtained during anexposure. In Figure 3b we compare theresults obtained using the ETC with realmeasurements on the star HD 10700.

HARPS DELIVERING PRECISIONNEVER REACHED BEFORE

During the HARPS commissioning wemonitored the star α Centauri B, which isa K-type star substantially smaller thanour Sun. During 7 hours we collected atotal of 420 spectra with typical SNR of500 each at λ=550 nm. The radial-velocitymeasurement sequence plotted in Figure4a indicates a dispersion of 51 cm/s butthe zoom shown in Figure 4b shows thatthis dispersion is completely dominatedby 4-minutes stellar oscillations. In fact,the power spectrum of this sequenceshown in Figure 5 clearly exhibits a seriesof peaks around 4 mHz, corresponding toindividual acoustic modes of the star, withamplitude in the range 10-20 cm/s. Thepositive interference of several oscillationmodes may lead to amplitudes much larg-er than the amplitude of single modes.From the power spectrum we estimate

that the average contribution of modes tothe total dispersion reaches 0.44 m/s. Byquadratic subtraction of this value fromthe measured dispersion we obtain anoise level of 0.26 m/s, which is in goodagreement with the value extrapolatedfrom the mean white noise level meas-ured in the high frequency range between6 and 8 mHz in the power spectrum. Thephoton noise on a single measurement isof 0.17 m/s, which leaves less than 0.2 m/sfor all other possible error sources (ThArnoise, guiding errors, influence of theatmosphere, instrumental errors).

The long-term precision of the instru-ment cannot be checked easily because itrequires a long time base on one hand,and the knowledge of stable stellarsources on the other hand. Especially thelatter point represents a new challengesince the intrinsic stability of the stars hasnever been studied at this level of preci-sion before. Nevertheless we havebeen able to gain some indication ofthe long-term precision of HARPSby observing the star HD 83443which has a planetary companionwith well known orbital parameters.Figure 6 shows a total of nine radial-velocity data points, some of themcollected during the first commis-sioning period in February 2003, andsome others about 4 months laterduring the second commissioningheld in June 2003. All the data fitwell the calculated orbit whoseparameters are fully consistent withthe previously known solution. Theobtained dispersion (O−C) is only

1.7 m/s and partly due to photon noise.The residual noise is a combination of cal-ibration errors (< 0.5 m/s) and effects dueto the star itself, which can be due to pul-sations, activity and jitter. It is also worthnoting that the HARPS vaccum has beenbroken between the two commissioningruns.

We conclude that the short-term pre-cision (1 night) of HARPS is well below 1m/s. Even on the time scale of a commis-sioning period (2 weeks) we have beenable to measure standard stars whichshowed radial-velocity dispersions of1 m/s including photon noise. The samelevel of precision is obtained onHD 83443 over a time scale of 4 months.Apart from photon noise a major contri-bution to the residuals probably comesfrom the star itself (pulsations, activity, jit-ter). Only an in-depth, long-term studywill allow us to identify the relativeimportance of the various error sources.

CATCHING THE TINY MELODYOF A SOLAR-LIKE STAR

Stars, which are spheres of hot gas, prop-agate very well in their interiors acousticwaves which are generated by turbulentconvection near their surfaces. Frequen-cies and amplitudes of these acousticwaves, also called oscillation modes or p-modes, depend on the physical conditionsprevailing in the layers crossed by thewaves and provide a powerful seismolog-ical tool. Helioseismology, which moni-tors the oscillation modes of our Sun, hasbeen used since the 1970’s and led tomajor revisions in the “standard model”of the Sun and provided, for instance,measures of the Sun’s inner rotation, thesize of the convective zone and the struc-ture and composition of the external lay-ers. Solar-like oscillation modes generateperiodic motions of the stellar surfacewith periods in the range 3 − 30 min., butwith extremely small amplitudes. The cor-responding amplitudes of the stellar sur-face velocity modulations are in the range10 − 100 cm/s.

Figure 4: a) Series of 7 hours and 420 exposures on α Centauri B proving the extraordinaryshort-term precision of HARPS. b) Zoom of figure a) to illustrate the presence of a periodic sig-nal produced by the stellar pulsation.

Figure 5: a) Power spectrum of α Cen B. The acoustic modes corresponding to the 4-minutesoscillation are clearly identified and emerge well above the noise. b) Autocorrelation of thepower spectrum of α Centauri B.

Figure 6: Radial-velocity data of HD 83443,which harbours a known extra-solar planet. Thedata fit well the calculated orbit with a weighteddispersion of 1.7 m/s.

© ESO - December 2003 23Mayor M. et al., Setting New Standards with HARPS

Two years ago, we were able to detectwith the CORALIE spectrograph anunambiguous oscillation signal of 31 cm/samplitude of the star α Centauri A, anearby solar twin (Bouchy & Carrier,2001 & 2002). With these measurementsseveral oscillation modes were separatedand clearly identified.

The measurements made on α Cen-tauri B with HARPS have lower frequen-cy resolution because of the limited dura-tion of the tests. To obtain the same reso-lution we would have had to observe thestar for 13 consecutive nights. On theother hand the measurement obtainedwith HARPS are impressive mostlybecause of efficiency, cycle time and pre-cision. Figure 4 illustrates clearly thedetectivity obtained with HARPS:despite the small amplitude the 4-minutesstellar oscillation can be detected “byeye” directly in the time series, and theybecome even more evident in the FourierTransform space shown in Figure 5a. Fig-ure 5b shows the autocorrelation of thepower spectrum. The plot clearly indi-cates that the peaks have a comb-likestructure, with a main peak separation ofabout 160 µHz. This is the typical signa-ture of oscillation modes of identicalangular degree expected for solar-likestars. The intermediate peaks appearingin the figure correspond to the correlationbetween modes with angular degree l=1and l=0 or 2.

Such a level of accuracy in the detec-tion of solar-like oscillation modes hasnever been reached before. This resultobtained with HARPS on a very shortsequence (7 hours), clearly shows theunique potential of this instrument in thedomain of asteroseismology. Long

sequences (several nights are needed toreach a sufficient frequency resolution tocharacterize individual p-modes) of highsampling radial-velocity measurementswill permit to measure acoustic waveswith amplitudes as tiny as a few cm/s.

DOWN TO THE STELLAR LIMITAfter the amazing results obtained on αCentauri B we have decided to monitor asmall set of solar type stars. Figure 7shows short sequences of radial velocitymeasurements obtained on these objects.On each of these sequences the stellaroscillations are clearly visible. Stellarparameters and observed oscillationmodes are listed in the Table 2.As expect-ed, these measurements clearly show thatperiod and amplitude of the oscillationmodes are directly related to the stellarproperties. They demonstrate the fullcapability of HARPS for asteroseismolo-gy, not only on very bright stars.

These results show however also thatthe main limitation for HARPS will comefrom the stellar “noise” itself. The individ-ual mode amplitudes of these stars are inthe range 10−50 cm/s but their additive

interference leads to modulation of theDoppler time series of up to 10 timestheir value. This is clearly a limitation forhigh precision exoplanet surveys at thelevel of 1 m/s and requires us to define aprecise strategy of target selection andobservation. Short-period oscillations (4−8 min.) of small amplitudes, which occurin G and K dwarfs, can be easily averagedout with typical exposures times of 15minutes. Sub-giants and giants, which canshow larger amplitudes and longer peri-ods, should however be carefullyremoved from the sample. This kind ofprecaution should permit us to detectexoplanets at the level of 1 m/s.

OBSERVING FAINT OBJECTSThe OGLE-III programme, involved inan extensive ground-based photometricsurvey for planetary and low luminosityobject transits, has recently announcedthe detection of transiting candidatesaround stars located in the direction ofthe Galactic centre. The radii of theseobjects were estimated to be in the rangebetween 0.5 and 5 Jupiter radii, meaningthat some of these objects are probably inthe planetary domain. In order to meas-ure the mass of these objects, and thendetermine their real nature, a radialvelocity follow-up is necessary.

In order to test the limiting magnitudeof HARPS, we observed the faint candi-date OGLE-TR-56 (mv=16.6). One hourexposure allowed us to reach a signal-to-noise ratio between 3 and 4 correspon-ding to a photon noise uncertainty on theradial velocity of about 30 m/s. The fiveRV measurements presented in Figure 8demonstrate the full capability ofHARPS to find and characterize hotJupiters around stars as faint as OGLE-TR-56.We have fitted to these data pointsa sinusoidal curve with fixed period asgiven by the photometric study.The meas-ured amplitude is however much differentand in contrast to the results published byKonacki et al. (2003).A complete analysisof these measurements will be required inorder to determine whether this RV mod-ulation is due to a planetary companionor to a binary system blended and

Table 2: Stellar parameters and oscillation modes properties

Star Spectraltype Period Peak to Peak

modulation RMS Dispersion

α Cen B K1V 4 min. 1.5 m/s 0.5 m/s

HD20794 G8V 5 min. 2.5 m/s 0.8 m/s

HD160691 G5IV-V 9 min. 5 m/s 1.7 m/s

β Hydri G2IV 16 min. 5 m/s 1.7 m/s

Figure 7:Summary plotof the radial-velocity ofstars for whichwe have madelong observa-tions series.All of themshow clearindications ofstar pulsation.

The Messenger 11424

masked by the main star. This kind ofstudy has a huge interest for the futureradial-velocity follow-up of exoplanetsdetected by the satellite COROT.

THE HARPS EXOPLANET PROGRAMME

The HARPS Consortium GTO pro-gramme will be devoted exclusively to thestudy and characterization of exoplanets,in continuation of a planet-search pro-gramme initiated 10 years ago (Queloz &Mayor, 2001). For a large, volume-limitedsample we will do a first screening inorder to identify new “Hot Jupiters” andother Jovian-type planets. Increasing thelist of “Hot Jupiters” will offer a chance tofind a second star with a planetary transitamong relatively bright stars. Better sta-tistics are needed to search for new prop-erties of the distribution of exoplanetparameters.

This part of the programme alreadystarted in July 2003 within the first GTOperiod assigned to the Consortium. In thisshort run of only 9 nights HARPSunveiled all its characteristics, its opticalefficiency, the unique precision and, inparticular, the outstanding efficiency ofthe reduction pipeline. In fact, we havecollected about 500 spectra and radial

velocity measurements of several hun-dred stars. Many of them show a varyingradial velocity, sometimes with a clearperiodicity. In the case of HD 330075 thesignal is most probably due to an exo-planet orbiting this star in 3.37 days. Theradial-velocity curve of HD 330075 isshown in Figure 9. Precise photometricmeasurements carried out on the SATDanish Telescope and by the Swiss-EulerTelescope have shown the absence of aphotometric transit, which would haveallowed us to determine the radius andthe orbital inclination of the planet. Nev-ertheless, these measurements have alsodemonstrated that the photometric vari-ability is low, and that we can excludetherefore the radial-velocity variation tobe produced by stellar activity.This makesHD 330075 the first HARPS extra-solarplanet candidate.

Only a few of the hundred detectedplanets have masses less than the mass ofSaturn, and due to the present precisionof radial velocity surveys the distributionof planetary masses is heavily biased (orcompletely unknown) for masses lessthan half the mass of Jupiter. We will takeadvantage of the very high precision ofHARPS to search for very low mass plan-ets. For a sample of pre-selected non-active solar-type stars we will be able toexplore the domain of the mass-functionfor planetary masses less than the mass ofSaturn down to a few Earth masses forshort periods.

A systematic search for planets will bemade for a volume-limited sample of M-dwarfs closer than 11 parsecs. Such a sur-vey of very low mass stars will give us achance to derive the frequency of planetsas a function of the stellar mass. Theseobjects are of prime importance forfuture astrometric studies to be carriedout with the VLTI or space-mission likeSIM.

Stars with detected giant planetsexhibit an impressive excess of metallicityin contrast to stellar samples withoutgiant planets (Santos et al., 2001). Theexcess of metallicity does not seem relat-

ed to the mass of the convective zone andprobably finds its origin in the chemicalcomposition of the primordial molecularcloud. To add new constraints to the linkbetween star chemical composition andfrequency (or properties) of exoplanets,we will carry out two additional pro-grammes: The first programme is a searchfor exoplanets orbiting solar-type starswith notable deficiency (for most of them[Fe/H] between −0.5 and −1.0). Amongthe existing detections of exoplanets onlytwo or three have been found with metal-licity in that range. We aim at being ableof estimating the frequency of exoplanetsin that domain of metallicity and, if possi-ble, to compare their characteristics(masses, orbits) to planets orbiting metalrich stars. The second programme aims atexploring the link between stellar metal-licity and properties of exoplanets. Visualbinaries with solar-type stars of almostidentical magnitudes have been selected.We will search for exoplanets orbitingone of the components of these systems.For those including giant planets adetailed chemical analysis will be donefor both stellar components to search forpossible differences in their chemicalcompositions.

Follow-up radial velocity measure-ments for stars with planetary transitsdetected by the COROT space missionwill be made with HARPS. Photometrictransits provide an estimate of the radiusof the transiting planet as well as theorbital period and phase. Complementaryground-based spectroscopic measure-ments with HARPS will constrain theplanetary mass and then the planet meandensity. The main scientific return for theplanetary programme of the COROTmission will come from the combinationof the photometric and radial velocitydata.

ACKNOWLEDGEMENTSWe would like to acknowledge the many peo-ple whose names do not appear among theauthors but have contributed to the great suc-cess of the HARPS project through their valu-able and dedicated work. This project has beenfinanced by the Swiss National Science Foun-dation, the French region “Provence, Alpes etCôte d’Azur”, the Institut National des Sciencesde l’Univers INSU (F), the Université deGeneve (CH), the Observatoire de Haute-Provence (F), the Physikalisches Institut derUniversität Bern (CH), and the Serviced’Aéronomie du CNRS (F). The HARPS Pro-ject is a collaboration between the HARPSConsortium and the European SouthernObservatory (ESO).

REFERENCESBouchy F. & Carrier F., 2001, ESO Messenger,

106, 32Bouchy F. & Carrier F., 2002, A&A, 390, 205Konacki M. et al., 2003, Nature, 421, 507Pepe F. et al., 2002, ESO Messenger, 110, 9Queloz D. & Mayor M., 2001, ESO Messenger,

105, 1Santos N. et al., 2001, A&A, 373, 1019

Figure 8: Radial-velocity of OGLE-TR-56measured with HARPS. Despite the faint mag-nitude of the object, the dispersion of the datapoints around the fitted curve is only 30 m/s.

Figure 9: First HARPS extra-solar planet candidateHD 330075. The planet has aminimum mass of 0.8 MJ andorbits the star in 3.37 days ata distance of 0.044 AU fromits parent star, and belongstherefore to the category ofHot Jupiters. The inducedradial-velocity variation has asemi-amplitude of 107 m/s.The total dispersion withregard to the orbital solutionis (O-C)=2.5 m/s, and includesphoton noise, stellar pulsa-tion and possible jitter.