Telescopes and Instrumentation
Astronomical Spectrograph Calibration at the Exo-Earth Detection Limit
Gaspare Lo Curto1 length (such as in Å), as a fractional For the measurement of frequencies, Luca Pasquini 1 wavelength, e.g. 10–7, or by scaling by the LFCs represent the ultimate level of preci- Antonio Manescau1 speed of light to express it in m/s. The sion, as they are locked to the energy Ronald Holzwarth 2, 3 limitation on Th-Ar wavelength precision level of a well-known atomic transition via Tilo Steinmetz 3 is due not only to the measurement pro- an atomic clock. They are the most pre- Tobias Wilken 3 cess (Palmer Engleman, 1983), but also cise time-keeping devices, and the unit of Rafael Probst 2 to the production method (contamination) measurement of time in the International Thomas Udem 2 and aging of the lamps. Even when aver- System (SI), the second, is defined by the Theodor W. Hänsch 2 aging over a wide spectrum with say, caesium transition via a caesium atomic Jonay González Hernández 4, 5 10 000 lines, measurements are limited clock. LFCs have many applications, from Massimiliano Esposito 4, 5 to an overall precision of 10–9 at best metrology and precise time-keeping to Rafael Rebolo4, 5 (the achievable precision is furthermore laboratory atomic and molecular spec- Bruno Canto Martins 6 degraded by line blending, and non- troscopy. Now the period is beginning Jose Renan de Medeiros 6 uniform density of the Th lines across the when LFCs will “look at the sky”. spectrum).
1 ESO This precision in the measurement of the The demonstrator programme 2 Max-Planck-Institut für Quantenoptik, positions of spectral lines is not sufficient Garching, Germany for various compelling science cases: In 2006 ESO approached the group of 3 Menlo Systems GmbH, Martinsried, – the measurement of the variation of the Prof. Theodor Hänsch at the Max Planck Germany fundamental constants, which requires Institute for Quantum Optics (MPQ), to 4 Instituto de Astrofisica de Canarias, a precision at least as good as the study the possibility of using an LFC for La Laguna, Spain precision with which the constants are the calibration of high-resolution astro- 5 Departamento de Astrofísica, determined; for the best known (the nomical spectrographs. Due to its unsur- Universidad de La Laguna, Tenerife, proton to electron mass ratio and the passed stability, HARPS at the 3.6-metre Spain fine structure constant) this is telescope in La Silla provided the best 6 Universidade Federal do Rio Grande de ~ 3–4 × 10–10 (Beringer et al., 2012); candidate to validate the performance of Norte, Natal, Brazil – the amplitude of the recoil motion this technique (Mayor et al., 2003). A fruit- imprinted by the Earth on the Sun ful collaboration was initiated between is ~ 9 cm/s. The radial velocity detec- ESO, MPQ and Menlo GmbH, a spin-off Following the development of the laser tion of an extrasolar planet with company from MPQ which markets LFCs, frequency comb which led to the 2005 the mass of the Earth in a 1 AU orbit with the goal of demonstrating the feasi- Nobel Prize in Physics, we began in around its solar-type star therefore bility of operating an LFC to calibrate an vestigating the possibility of using this requires a measurement precision of astronomical spectrograph, thus opening novel technology for precise and accu- about 3 cm/s, or 10–10; up a new horizon for the current and next rate spectrograph calibration. A pro- – the direct measurement of the expan- generation high-precision, high-resolution gramme was begun, aimed at demon- sion rate of the Universe, which can spectrographs. strating the capabilities of laser fre- constrain the cosmological parameters quency combs (LFC) when coupled to that define the metric of the theory The demonstrator programme started in an astronomical spectrograph. In the of gravity, requires a measurement pre- 2007 (Araujo-Hauck et al., 2007) and last three years we have tested an LFC cision at the level of cm/s (10–10) for concluded in 2011. In this period a proto- connected to HARPS at the 3.6-metre over ten years (Liske et al., 2008). type LFC dedicated to astronomy (or telescope in La Silla, the most precise astro-comb) was developed and refined. spectrograph available. Here we show The new class of giant telescopes with The astro-comb was tested in the labora- the very promising results obtained so greatly increased light-collecting power tory; in the infrared (IR) regime on a tele- far, and outline future activities, including will naturally ease the photon noise limi scope for the first time (Steinmetz, 2008); the provision of an LFC system for rou- tation to high precision spectroscopy by to validate technical solutions; and in four tine operation with HARPS, to be offered a factor of five to ten. A new calibration campaigns in the visible, with HARPS to the community in the near future. source is therefore needed, which ena- at the 3.6-metre telescope in La Silla, to bles these science cases and capitalises test the global performance (Lo Curto on the great opportunity that giant tele- et al., 2010; Wilken et al., 2010; Wilken et The most widely used wavelength refer- scopes open up for high precision spec- al., 2012). ence in astronomical spectroscopy, the troscopy. The ideal calibration source thorium–argon (Th-Ar) lamp, can achieve would be at least ten times more precise An LFC consists of thousands of equally a precision on the determination of indi- than Th-Ar lamps allow and would have spaced frequencies over a bandwidth vidual line positions of several tens of many unblended lines, with approximately of several THz. It is based on the proper- metres per second (m/s). The uncertainty the same intensity and equally spaced ties of femtosecond (fs) mode-locked in the wavelength of a spectral line can across the spectrum. The obvious choice lasers. The frequency difference between either be expressed as an absolute wave- is the laser frequency comb. two neighbouring lines corresponds to
2 The Messenger 149 – September 2012 Figure 1. The spectrum of the LFC dispersed by a Although an LFC based on fibre lasers are detrimental to nonlinear conversion low-resolution grating and projected on a wall. The is essentially an off-the-shelf product, its processes. The approach to overcome individual lines are not resolved and only the contin- uum is visible: its spectral structure is due to the adaptation for use in an instrument like this dilemma employs two key compo- spectral broadening stage and can be flattened out HARPS requires major developments. nents. First, a comb system based on by a spatial light modulator (SLM). The basic turn-key LFC systems available an Yb-fibre laser enables the use of Yb- on the market today deliver a comb of fibre high-power amplifiers to reduce the repetition frequency (frep) of the lines centred at 1025–1050 nm, with a re the problem of low pulse energies. Sec- pulsed laser and is therefore constant petition frequency of 250 MHz and gen- ond, when using specially designed across the comb spectrum. The entire erally cover only few nanometres in wave- photonic crystal fibres (PCFs), relatively LFC spectrum can be described by the length. Resolving spectral lines 250 MHz low pulse energies are sufficient to obtain simple equation f = foff + n∙frep where frep apart in the visible range requires a spec- spectral broadening. is the repetition frequency, i.e. the dis- tral resolution of more than six million, tance in frequency between two adjacent which is not practical for the typical use Another challenge comes at the FPC fil- lines, and foff is the “offset frequency” that of an astronomical spectrograph. An LFC tering stage and arises from the fact that can be interpreted as a “zero point”; n with such a line spacing would appear the nonlinear processes (interactions is a large integer which projects the radio as a continuum source to instruments between photons) in the PCF can amplify frequencies foff and frep into the visible such as HARPS or UVES (see Figure 1). spectral lines that were intended to be domain. Since all frequencies used in the suppressed. We saw this effect during system are radio frequencies, they can New developments are needed to oper- our first test run in La Silla. The solution be stabilised to an atomic clock using ate an LFC on an astronomical spectro- was to employ more FPC cavities, essen- well-established electronic phase-locking graph in the visible with a spectral resolu- tially one of them replicating the sup techniques. In this way, each optical tion of ~ 105: pression of the unwanted modes, and frequency obtains the accuracy and long- 1) increase the repetition frequency to with a higher finesse. Finally in our system term stability of the atomic clock. ~ 18 GHz; we used three FPCs in series to guar 2) double the frequency of the spectrum antee a sufficient suppression of the An LFC acts like a gear, transferring the to have it centred in the visible at intermediate lines. A series of four FPCs precision of atomic clocks from the micro- ~ 520 nm; was also tested, but no improvements wave regime to the optical. LFCs are 3) broaden the spectrum to increase were noticed from the fourth cavity. The the ideal calibrator for astronomical spec- wavelength coverage. line spacing is increased to 18 GHz, trographs if they can cover the spectral These are the three steps where most of which is well resolved by HARPS (the in bandwidth of the spectrograph with a the efforts of the programme have been strument has an intrinsic resolution of sufficiently flat spectrum and if their line focussed. ~ 5 GHz). spacing is adapted to the spectrograph’s resolution. While there are several pro- After the last FPC a fibre amplifier is posed frequency comb systems that will Towards an LFC for astronomy employed that can deliver up to 10 W to match the criteria for a spectrograph cali- facilitate the subsequent nonlinear pro- brator, our choice has been to use a High finesse Fabry-Pérot cavities (FPCs) cesses. The spectral bandwidth after fibre-laser-based LFC. Fibre lasers are acting as periodic, high-resolution spec- these amplifiers suffers from re-absorp- technically mature and turn-key systems tral filters can be used to increase the tion and gain narrowing. The initial 50 nm that are commercially available. When line spacing by transmitting only modes wide spectrum reduces to a 3 dB band- using Yb-fibre lasers, high-power amplifi- (lines) that are phase-shifted by an inte- width of only 20 nm after the high-power ers can be employed and the second ger number of wavelengths and sup- amplifier. After re-amplification, the fre- harmonic of the central wavelength pressing the intermediate ones. After this quency of the light is doubled by a sec- (1030 nm) is in the centre of the desired first step, the spectrum of the LFC under- ond harmonic generator (SHG), and wavelength range for a spectrograph goes two nonlinear processes: one for finally injected into the PCF for spectral in the visible. However, the required fibre frequency doubling and one for spectral broadening. The broadening is a highly length limits the round-trip time in the broadening. The challenge is to combine nonlinear process (four-wave mixing), and oscillator, and thus the line spacing of the the frequency conversion and the high so the intensity distribution in the final laser (its pulse repetition rate) is currently pulse repetition rate. A high repetition rate spectrum can vary. For this reason, in the limited to roughly below 1 GHz. corresponds to low pulse energies that last test run, a spatial light modulator
The Messenger 149 – September 2012 3 Telescopes and Instrumentation Lo Curto G. et al., Spectrograph Calibration at the Exo-Earth Detection Limit
Figure 2. Representa- Rb atomic tion of the HARPS-LFC clock Fabry-Pérot setup as described in 18 GHz filter cavities the text. The SLM after 250 MHz the PCF is missing in the figure because it Laser frequency comb was not always used during data acquisition (see text for details). High power From Wilken et al., 2012. Dynamic fibre amplifier scrambler PCF HARPS A 450 nm–590 nm Fibre 1040 nm–1060 nm SHG spectrograph coupling B Multimode fibre Single mode fibre
Figure 3. A comparison LFC spectrum of a portion of one order of the LFC and the Th-Ar lamp as seen by HARPS. Th-Ar lamp spectrum
(SLM) was sometimes inserted as a last The wavelength range of the LFC for are considering a test with two independ- stage of the LFC, yielding a flat spectrum HARPS now covers ~ 100 nm, and work ent LFCs feeding light to HARPS in the envelope at the cost of an attenuation of is in progress to extend this to at least future. Currently we are focussing instead 12 dB. 200 nm. Owing to the high line density of on effects due to mode (mis)matching, the LFC spectra, we are now able to coupling and injection. To characterise Light from the LFC is finally coupled to a better characterise the detectors. For these, we inject light into both HARPS multimode fibre that comprises a dynamic example, Figure 4 shows the effect of the calibration fibres and illuminate both the mode scrambler. Multimode fibres are stitching pattern of the CCD fabrication object and the simultaneous reference mandatory, because starlight cannot yet on the wavelength calibration. Every orders (from here onward referred to be coupled efficiently to single mode 512 pixels the wavelength solution en as channels) in the CCD. We then moni- fibres. The scrambler serves to reduce counters a discontinuity as the intra-pixel tor the relative drifts measured between multi-pass interference effects and the distance is slightly different along the the two channels, with the assumption sensitivity to the light injection by aver stitching borders. The effect reaches over that instrumental drifts are subtracted out aging over a large number of fibre modes. 60 m/s, well above the attainable stability when looking at the relative variations The light is then coupled, within the of HARPS. Although the effect is clearly between them (an assumption which is HARPS calibration unit, to the two HARPS visible when using the LFC calibration, it the basis of the simultaneous reference calibration fibres, which simultaneously goes unnoticed when Th-Ar calibration method). illuminate the telescope focal plane, is used, due to the low number of lines injecting light into the “object” fibre and present in the orders and their highly non- In Figure 5 the measured drift of all the the “simultaneous reference” fibre. The uniform distribution. This effect clearly spectra collected in the two test runs of system is sketched in Figure 2. underlines the need for a detailed char- November 2010 and January 2011 is acterisation of the detector when the shown. During the test run we stressed highest radial velocity precision and ac the system in many ways, well beyond Characterising the LFC on HARPS curacy are aimed for. Such in-depth char- what could be considered “standard acterisation can be performed with the operation”: we completely disassembled Some of the advantages of LFCs over use of the LFC, by modifying the param the FPC chain and reassembled it again Th-Ar lamps are immediately noticeable eters that define the spectrum, i.e. the in January to test a fourth FPC; changed from Figure 3, where a portion of the offset frequency and the repetition fre- several PCFs; inserted and removed order of the two sources is compared. quency, effectively scanning the detector neutral density filters; misaligned the The LFC has a very dense, non-blending and probing all the pixels which lie within injection fibres; introduced polarimetric line pattern (~ 350 lines per order), which the spectral profile. plates, integrating spheres, etc.... We also is also very uniform in intensity. In con- switched off the CCD cooling for few trast the Th-Ar lamp has fewer lines The characterisation of the stability of minutes. Despite this, each individual (~ 100 lines per order), some of which the LFC system when coupled to HARPS series of data shows a relative drift con- are saturated. The spectra in Figure 3 has been clearly the first priority during sistent with, or very close to, the esti- were obtained using both the object and the tests. However with only one comb mated photon noise, and the relative drift the simultaneous reference fibres of available we have to rely on its intrinsic between the two channels never HARPS, and by illuminating the spectro- stability as verified up to now on very exceeded 1 m/s, and the root mean graph simultaneously with both sources. many laser physics experiments. We square (RMS) of the relative drift was
4 The Messenger 149 – September 2012 Figure 4. Evidence of well beyond what happens under stand- 1DRHCT@KRSGNQCDQjS the CCD stitching 1DRHCT@KRSGNQCDQjS CHRBNMSHMTHSHDR ard operating conditions. pattern as revealed by #HEEDQDMBD3G Ql+%"B@KHAQ@SHNM
B the LFC wavelength R U
l calibration. Due to the When looking at sequences of data dur- L
D paucity of spectral lines,
U ing which we did not intervene to modify D U TQ the Th-Ar calibration the system (orange and violet vertical TQ B was unable to identify B this effect. From Wilken stripes in Figure 5), the stability was seen et al., 2010. to be superb. As in almost all series, the KHAQ@SHNM @ KHAQ@SHNM
@ RMS of the relative drift between the two
RNEB l l channels is consistent with the photon RNEB noise of ~ 7 cm/s. In order to decrease l l the contribution of the photon noise, we 1DRHCT@K
1DRHCT@K added an increasing number of expo- l l sures and computed the Allan (or two- sample) deviation (Figure 6). This estima- l l tor is capable of distinguishing statistical ""#OHWDKHMMCDBGDKKDNQCDQ noise processes from systematics, e.g., drifts. In Figure 6 we see that the orange and violet points, corresponding to two ~ 34 cm/s over the whole run. When the 5 cm, was used to couple the LFC outgo- quiet LFC series, follow well the curve CCD cooling was stopped, and then ing fibre to the HARPS calibration fibres. of the estimated (not fitted) photon noise, restarted, we could measure a drift of We do not yet understand the cause of up to a level of ~ 2.5 cm/s, where the 13 m/s on the individual channels, but this drift, but speculate it could be related data flatten. At this point systematics limit the relative drift between channels was to the strong intensity variation (close the stability of the system. Short-term below 40 cm/s peak to valley, an indi to two orders of magnitude) or to a differ- (~ 2 hours) repeatability of the LFC+ cation that the method of simultaneous ent occupation of the modes after the HARPS system is at the level of 2.5 cm/s. reference in HARPS is capable of reduc- transition through the integrating sphere. Although we cannot say the same for ing systematic effects by more than a Possible alternative explanations point the long-term stability due to the various factor of 30, at least when they originate towards detector effects that could dis- drifts exemplified in Figure 5, we stress from the detectors. tort the instrument profile depending on once more that these instabilities in the light intensity, or data reduction effects, long term were most likely generated Two large deviations in the relative drift also related to a change of the instrument from our interventions in the system. In between the two channels are seen in profile for very faint signals. Work is still Figure 6 we display a similar measure- the data (Figure 5). These correspond being done to understand these measure- ment for Th-Ar calibration, and the limita- to the series of spectra acquired when ments, but it must be stressed that this is tion of Th-Ar lamps with respect to the an integrating sphere, with a diameter of LFC is clearly visible.
November 2010 January 2011 8 Channel A Series 1 Series 2 Channel B A–B 6 Figure 5. Drift of the two spectrograph chan- nels for all the LFC 4 data collected between November 2010 and
ence (m/s) January 2011 (almost
fer 1300 spectra) is shown. 2 The drift of the individual channels is measured with respect to an arbi- 0 trarily chosen spectrum. The relative drift between them is shown in black. The orange
Computed velocity dif –2 and violet vertical stripes indicate two long Integrating sphere CCD cooling off series (> 200 spectra in total) during which –4 time the system was left unperturbed. From 0 200 400 600 800 1000 1200 Wilken et al., 2012.
The Messenger 149 – September 2012 5 Telescopes and Instrumentation Lo Curto G. et al., Spectrograph Calibration at the Exo-Earth Detection Limit
During our test runs in La Silla we moni- 25 Figure 6. Two-sample tored the radial velocity variation of the variance of the LFC and the Th-Ar calibration. solar-type star HD 75289, known to host The black solid line is an a Jupiter-mass planet with a period of estimation of the photon 10 3.5 days. Using the LFC for wavelength noise. The dashed hori- calibration and simultaneous reference, zontal lines at 10 cm/s and 2.5 cm/s indicate, we could reconstruct the orbit of the exo- for the Th-Ar and the planet. The measurements calibrated LFC simultaneously, the with the LFC nicely agree with previous level at which the data data and the deviations from the fit could 2.5 no longer follow the esti- mated noise and sys- be due to stellar activity or pulsations. tematic effects become To our knowledge this is the first time that LFC series 1 important. From Wilken the orbit of an exoplanet has been recon- 2-sample deviation (cm/s) LFC series 2 et al., 2012. structed using LFC calibrations. 1 Thorium lamp Photon noise
During the four years of the programme 0.5 the LFC system has improved much, 108 109 1010 1011 not only in its performance, but also in Accumulated photons robustness and operational stability. The laboratory test and the campaigns with refine the observation strategy for and the E-ELT. The experience collected HARPS have achieved successful results, ESPRESSO planet searches. with the HARPS LFC will be invaluable for which have consolidated the concept of ESPRESSO and beyond. the system. The achieved repeatability of The radial velocity (RV) precision of 2.5 cm/s is sufficient to detect an Earth- HARPS is estimated to be around While delivering top RV precision data like planet in a one-year orbit around a 60 cm/s (Lovis et al. 2006). Knowledge of with HARPS, this project would permit an star like our Sun. the instrument has permitted the instru- understanding of the long-term system mental causes that limit the RV precision atics (if any), early optimisation of comb to be identified: the light injection system operations and of the reduction software Laser frequency comb for HARPS and scrambling, the wavelength calibra- before ESPRESSO goes online and well tion system and, to a lesser extent, the in advance of the preliminary design of After the successful completion of the temperature variations of the detector. any high resolution spectrograph for the demonstrator programme, beginning in After commissioning a new injection sys- E-ELT. February 2012, in partnership with the tem in 2009, which improved the image Instituto de Astrofisica de Canarias and stability at the fibre entrance, the Th-Ar the Universidade Federal do Rio Grande wavelength calibration system is the Acknowledgements de Norte, we started a project aimed at strongest limiting factor to the long-term We wish to thank Gerado Avila from ESO for prepar- the acquisition of a turn-key LFC system RV precision of HARPS. With the LFC, ing the injection and the scrambling unit, for HARPS. The system will be offered which has been shown to have a stability and Francesco Pepe, Christophe Lovis and Bruno to the astronomical community after suc- as good as a few cm/s with HARPS, the Chazelas from the Geneva Observatory for their help cessful commissioning at the telescope. instrument is capable of reaching long- with the reduction software and their advice. This project is not only directed towards term RV precision below 30 cm/s, giving improvement of HARPS, but also towards access to the detection of Earth-mass References future instruments like ESPRESSO. The planets in close-in orbits, tracing out goals of this programme are to: the path towards the detection of Earth Araujo-Hauck, C. et al. 2007, The Messenger, 129, 24 – gain experience in the use of the LFC, twins. This RV precision is midway be- Beringer, J. et al. 2012, Phys. Rev., D86, 010001 study its long-term behaviour and its tween the current HARPS performance Liske, J. et al. 2008, MNRAS, 386, 1192 reliability in operations, optimise the and the 10 cm/s precision expected for Lo Curto, G. et al. 2010, Proc. SPIE, 7735, 77350Z reduction tools to achieve the best per- ESPRESSO at the VLT, planned to start Lovis, C. et al. 2006, Nature, 441, 305 Mayor, M. et al. 2003, The Messenger, 114, 20 formance with the LFC, with a view operations in about four years. The LFC Murphy, M. et al. 2007, MNRAS, 380, 839 to its use on VLT and E-ELT instrumen- calibration system will be one of the key Palmer, B. & Engleman, R. 1983, Atlas of the tation; components of ESPRESSO. Thorium spectrum, Los Alamos National – bridge the gap between HARPS and Laboratory Steinmetz, T. et al. 2008, Science, 321, 1335 ESPRESSO and acquire the potential of The step of moving from a laboratory Wilken, T. et al. 2010, Proc. SPIE, 7735, 77350T detecting a population of super-Earths prototype to a device to be operated at Wilken, T. et al. 2010, MNRAS, 405, L16 in the habitable zone of solar-type stars; the telescope on an existing instrument Wilken, T. et al. 2012, Nature, 485, 611W – improve HARPS precision, which in will move us to the production phase, turn will improve our capability for find- and will drive the development of an op ing low-mass planets, further the erational comb for the next high-precision understanding of stellar activity and radial velocity instruments for the VLT
6 The Messenger 149 – September 2012