Astronomical Science

Hunting for Frozen Super-Earths via Microlensing

Jean-Philippe Beaulieu1,3 8 Boyden , University of of cold telluric planets. Its detection Michael Albrow1,4 the Free State, Department of Physics, confirms the power of this method and, Dave Bennett1,5 , South Africa given our detection efficiency, suggests Stéphane Brillant 1,6 9 Astronomisches Rechen-Institut (ARI), that these recently-detected planets John A. R. Caldwell1,7 Zentrum für Astronomie, Universität may be quite common around M stars, Johannes J. Calitz1,8 Heidelberg, Germany as confirmed by subsequent detection Arnaud Cassan1,9 10 Lawrence Livermore National Labora- of a ~ 13 Earth-mass planet. Using a Kem H. Cook1,10 tory, Livermore, California, USA network of dedicated 1–2-m-class tele- Christian Coutures1,3 11 University of Tasmania, School of scopes, we have entered a new phase Stefan Dieters 1,11 Mathematics and Physics, Hobart, of planet discovery, and will be able to Martin Dominik1,2,12,13 Tasmania, Australia provide constraints on the abundance Dijana Dominis-Prester1,14 12 Scottish Universities Physics Alliance, of frozen Super-Earths in the near future. Jadzia Donatowicz1,15 University of St Andrews, School of Pascal Fouqué1,16 Physics and , St Andrews, John Greenhill1,11 United Kingdom The discovery of extrasolar planets is ar- Kym Hill1,11 13 Royal Society University Research guably the most exciting development Matie Hoffman1,8 Fellow in during the past decade, Uffe G. Jørgensen1,17 14 University of Rijeka, Physics Depart- rivalled only by the discovery of the cos- Stephen Kane1,18 ment, Faculty of Arts and Sciences, mic acceleration. The unexpected variety Daniel Kubas1,6 Rijeka, Croatia of giant exoplanets, some very close Jean-Baptiste Marquette1,3 15 Technische Universität Wien, Austria to their stars, many with high orbital ec- Ralph Martin1,19 16 Observatoire Midi-Pyrénées, Labora- centricity, has sparked a new generation Pieter Meintjes1,8 toire d’Astrophysique, Université Paul of observers and theorists to address John Menzies1,20 Sabatier, Toulouse, France the question of how planets form in the Karen Pollard1,4 17 Niels Bohr Institutet, Astronomisk context of protostellar accretion discs. Kailash Sahu1,21 Observatorium, København, Denmark Planets are now known to migrate and Christian Vinter1,17 18 Department of Astronomy, University of maybe even be ejected, via planet-disc Joachim Wambsganss1,9 Florida, Gainesville, Florida, USA and planet-planet interactions. We are Andrew Williams1,19 19 Perth Observatory, Perth, Australia beginning to discover how our Solar Sys- Kristian Woller1,17 20 South African Astronomical Observa- tem fits into a broader community of Marta Zub1,9 tory planetary systems, many with very differ- Keith Horne1,2,12 21 Space Telescope Science Institute, ent properties. Microlensing-based Alasdair Allan 2,22 Baltimore, Maryland, USA searches play a critical role by probing for Mike Bode 2,23 22 eSTAR Project, School of Physics, cool planets with masses down to that Daniel M. Bramich1,2,24 University of Exeter, Devon, United of Earth. Of key interest is how planets Martin Burgdorf 2,23 Kingdom are distributed according to mass and or- Stephen Fraser 2,23 23 Astrophysics Research Institute, bital distance (Figure 1) as this information Chris Mottram 2,23 Liverpool John Moores University, provides a crucial test for theories of Nicholas Rattenbury 2,25 Birkenhead, United Kingdom planet formation. Core accretion models Colin Snodgrass 2,6 24 The Isaac Newton Group of Tele- (Ida and Lin, 2005) are today the best Iain Steele 2,23 scopes, Santa Cruz de La Palma, description for the formation of planetary Yiannis Tsapras 2,23 Canary Islands systems: the accretion of planetesimals 25 The University of Manchester, Jodrell leads to the formation of cores, which Bank Observatory, Macclesfield, then start to accrete gas from the primi- 1 PLANET Collaboration Cheshire, United Kingdom tive nebula. This scenario predicts that for 2 RoboNet Collaboration M dwarf stars there is a preferential for- 3 Institut d’Astrophysique de Paris, mation of Earth- to Neptune-mass planets CNRS, Université Pierre et Marie Curie, In order to obtain a census of planets in 1–10 AU orbits. These planets are ex- Paris, France with masses in the range of Earth to pected to form within a few million years. 4 University of Canterbury, Department Jupiter, eight telescopes are being used More massive planet (Jupiter) formation of Physics and Astronomy, by the combined microlensing cam- is hampered by a longer formation time ­Christchurch, New Zealand paign of the PLANET and RoboNet col- (10 Myr) during which the gas evaporates 5 University of Notre Dame, Department laborations for high-cadence photo- and is no longer available to be accreted. of Physics, Notre Dame, Indiana, USA metric round-the-clock follow-up of 6 ESO ongoing events, alerted by the OGLE There is a wide variety of planets and at 7 McDonald Observatory, Fort Davis, and MOA surveys. In 2005 we detected first sight it appears that our system Texas, USA a planet of 5.5 Earth masses at 2.6 AU is very special. However, our view of the

from its parent 0.22 MA M star. This whole picture is still blurred by obser- object is the first member of a new class vational biases inherent to the detection

The Messenger 128 – June 2007 33 Astronomical Science Beaulieu J.-P. et al., Hunting for Frozen Super-Earths via Microlensing

techniques using transits and radial ve- Known exoplanets: 11 + 180 + 4 = 195 (January 2007) locities. Both methods are more sensitive Orbit period in years (for Solar-mass star) to massive planets close to their parent 3 d 30 d 1yr 10 yr 100 yr star. Doppler measurements and the 4 Doppler: 180 space transit missions, such as COROT, 10 can already, or will shortly, be able to de- 1000 tect Neptune-mass planets close to their t, µFUN parent star. Direct detections fill the other J extreme of very large separations which 100 S are unknown in our Solar System. It is ET/RoboNe Kec therefore necessary to use different tech- k, VL N 10 TI U niques, each probing different areas of , HAT SIM the planet-mass versus orbital distance OGLE, TrES, WASP parameter space. 1 V E MPF OGLE, MOA, PLAN Planet mass in Earths Kepler Already with ground-based observations, 0.1 M the microlensing technique is sensitive M to cool planets with masses down to that 0.01 Transits: 11 (+ 3) Microlensing: 4 of the Earth orbiting 0.1–1 MA stars, the Astrometry: 0 most common stars of our Galaxy, in or- P bits of 1–10 AU. Currently, over 700 micro- 0.01 0.1 1 10 100 lensing events towards the Galactic Bulge Orbit radius in Astronomical Units (AU) are alerted in real time by the OGLE and MOA surveys each year. During these Figure 1 (above): Exoplanet discovery space (planet Figure 2 (below): The left panel shows the caustic events, a source star is temporarily mag- mass versus orbit size) showing the 8 planets from structure of a star/planet lens, with two possible tra- our Solar System (labelled as letters), 180 Doppler jectories of a source star. The right panel shows nified by the gravitational potential of wobble planets (black triangles), 11 transit planets the corresponding observed light curves. Hitting the an intervening lens star passing near the (blue circled crosses), and 4 microlensing planets planetary caustic, or passing close to it, induces a line of sight, with an impact parameter (red circled crosses). Also outlined are the regions short-lived but clearly detectable photometric signal. smaller than the Einstein ring radius RE, a that can be probed by different methods: Doppler, transits; astrometry; and microlensing from the quantity which depends on the mass of ground and space. Microlensing is a cost-efficient the lens, and the geometry of the align- way to measure the mass function of cool planets ment. For a source star in the Bulge, with down to the mass of the Earth. a 0.3 MA lens, RE ~ 2 AU, the projected angular Einstein ring radius is ~ 1 mas, and 0.10 Mass ratio q = 0.002 Impact angle = 271.0 (deg) Impact angle = 50.0 (deg) the time to transit RE is typically 20–30 Separation d = 1.2 days, but can be in the range 5–100 days. 17.5 0.05

A planet orbiting the lens star generates a Planetary caustic

E Central caustic 18.0 caustic structure in the source plane, with θ /

2 0.00 one small caustic around the centre of θ -Band magnitude mass of the system, the central caustic, I –0.05 and one or two larger caustics further 18.5 away, the planetary caustics. If the source star happens to reach the vicinity of one –0.10 of the caustics, its magnification is signifi- –0.10 –0.05 0.00 0.05 0.10 2520 2540 2560 2580 HJD - 2.450.000 cantly altered as compared to a single θ1/θE lens, resulting in a brief peak or dip in the observed light curve. the detection of planets in such events space to derive the planet/star mass ratio becomes highly efficient (Griest and q, and the projected separation d in units The duration of such planetary lensing Safizadeh 1998). In contrast, planetary of RE. In general, model distributions for anomalies scales with the square root of caustics are only approached for a spe- the spatial mass density of the Milky Way, the planet’s mass, lasting typically a cific range of orientations of the source the velocities of potential lens and source few hours (for an Earth) to 2–3 days (for a trajectory, but the characterisation of a stars, and a mass function of the lens Jupiter). These two caustics (Figure 2) planetary signal is much easier for such stars are required in order to derive prob- provide two modes for detection. When configurations. ability distributions for the masses of the the central caustic is approached for all planet and the lens star, their distance, as events with a small impact angle between The inverse problem, finding the prop- well as the orbital radius and period of source and lens star, corresponding to erties of the lensing system, is a complex the planet, by means of Bayesian analysis. a large peak magnification of the event, nonlinear one within a wide parameter

34 The Messenger 128 – June 2007 The observational challenge is to monitor In the case of OGLE, which has accumu- ongoing microlensing events, detected Liverpool Telescope 2.0-m lated many images of a given field before by the OGLE and MOA survey telescopes, a microlensing event is detected there, with a fleet of telescopes to achieve SAAO 1.0-m the template is built from a set of the best round-the-clock monitoring and detect Rockefeller 1.5-m images and is not held fixed throughout real-time deviations in the photometric the season. But in our case, we start ob- signal. The telescopes belonging to our serving an event when receiving the network together with their locations are Danish 1.5-m OGLE or MOA alert, so we have to build shown in Figure 3. During the coming the template ‘on the fly’. This generates three Galactic Bulge seasons (from May problems when new images appear after Perth 0.6-m to September 2007, 2008, 2009 in the Faulkes North 2.0-m a few nights, which are better than the southern hemisphere) we are planning to Canopus 1.0-m first template. We then have to re-run the use the eight telescopes of the PLANET/ process routine on all images of the event. RoboNET networks: Danish 1.5-m at A different image subtraction pipeline Faulkes South 2.0-m La Silla (Chile), Canopus 1.0-m at Hobart is used on the RoboNet telescopes, but it and Bickley 0.6-m at Perth (Australia), follows the same philosophy. Rockefeller 1.5-m at Bloemfontein and SAAO 1.0-m at Sutherland (South Africa). Figure 3: The different telescopes of The typical uncertainty of the on-line These are the standard telescopes of the the PLANET/RoboNet network. photometry is 1.2% for an I = 17.8 mag PLANET network, to which were added Galactic Bulge star at the Danish 1.5-m in 2004 two robotic telescopes of the instructing observers at the PLANET tele- telescope, and allows an on-line detec- UK RoboNet network, North Faulkes 2-m scopes by means of a web page. We tion of a deviating signal. When this ap- in Hawaii and Liverpool 2-m in Canary is- plan to embed our experience into future pears, excitement grows and an alert to lands, joined in 2006 by the South Faulkes advanced versions of the priority algo- homebase is issued. Homebase then 2-m in Australia. rithm and further automate this process. prompts an off-line reduction of the event images, which are regularly uploaded to At the beginning of the night, the observ- the Paris central archive. The off-line re- Observing strategy, and description of er finds on the PLANET web pages the duction is done with our other image sub- the reduction pipelines list of targets with sampling intervals set traction pipeline, pySIS, which facilitates up by the homebase. He then defines the ‘fine-tuning’, so as to get the best possi- A typical observing season of the Galac- exposure times for each target and re- ble photometry but is more difficult to tic Bulge starts at the beginning of May ports them on our private web page, so automate for real-time use. If the off-line every year and lasts four months. Among that the homebase can estimate the ob- reduction confirms the deviation, an alert the 691 alerts available in 2006 (579 from serving load at each telescope. Typical is issued to the microlensing community OGLE-III and 112 additional from MOA-II), sampling intervals are 0.5, 1, and 2 hours, to intensify observations and maximise about 180 are available every night in the according to the priority of each event. the chances of a good characterisation middle of the season. Of these, around However, in case of a high magnification of the deviation, which is absolutely nec- 20 targets can be monitored by 1-m-class event, when the sensitivity to a planet is essary for future modelling of the event. telescopes, whereas the Danish 1.54-m maximal, the sampling interval can be re- Moreover, all photometric data are made and the 2-m telescopes can follow more duced to a few minutes, to the exclusion public immediately, as assistance to events. Therefore, we must apply some of all other candidate objects. all teams in order to maximise the planet criteria to select our 20 targets for every hunting community’s success. observing night. This is done by one At the end of the exposure, the image is member of the collaboration acting as a pre-processed (bias, dark removed and coordinator, the so-called ‘homebase’. flatfielded), gets a standard name and The discovery of the frozen superEarth Depending upon the current magnifica- is passed to an on-line pipeline. Starting OGLE-2005-BLG-390Lb tion, the source brightness, and the in 2006, on all the PLANET telescopes, time of the last observation, a priority al- we shifted from a DoPhot-based on-line On 11 July 2005, the OGLE Early Warning gorithm assigns a worth to each of the pipeline to an image subtraction pipeline System announced the microlensing events and suggests sampling rates, with based on ISIS (Alard 2000). This robust event OGLE 2005-BLG-390, with a rela- the goal to maximise the planet detection implementation, named WISIS, has two tively bright G4III giant as the source star. efficiency. If the magnification of one main tasks: process takes all available PLANET/RoboNet included it in its list event becomes very high, it may become images of a given event, chooses the best of targets and started to monitor it on the sole designated target during that template and subtracts all images from 25 July. The microlens peaked at a mag- night. While these suggestions are direct- that template after convolving the refer- nification Amax = 3 on 31 July. We were ly submitted to intelligent agents steering ence point spread function (PSF) with the planning to continue to monitor it until the the robotic telescopes of the RoboNet kernel to mimic the current PSF. The source exited the Einstein ring, when on network, the homebase currently tunes update task only processes new images 10 August observers at the Danish tele- them using our experience gained, before using a previously chosen template. scope noticed a measurement deviating

The Messenger 128 – June 2007 35 Astronomical Science Beaulieu J.-P. et al., Hunting for Frozen Super-Earths via Microlensing

by 0.06 mag from the point source point lens prediction. They then took a second 33 1.6 measurement, deviating by 0.12 mag. OGLE OGLE data became available, confirming 1.5 the deviation seen in Chile. In order to 2 check the nature of the deviation, home- 2.5 1.4 base increased the proposed sampling Planetary 1 rate at the automated Perth telescope. 1.3 deviation Perth started to observe this event con- tinuously as soon as the target was within 2 –1 0 9.5 10 10.5 11 reach. South Africa was clouded out, and when observations resumed in Chile, Magnification OGLE Danish it was clear that the anomaly was over. RoboNet Perth Different telescopes continued to ob- 1.5 Canopus MOA serve the microlensing event. Perth data – which were received only with some delay – finally confirmed the short-dura- tion deviation with a good coverage of six 1 additional data points. Combined with –20 0 20 two additional independent data points Days since 31 July 2005 UT from the MOA team (Mt. John, New Zea- land), the evidence of a well-covered Figure 4: Light curve of OGLE-2005-BLG-390Lb, period of ~ 10 years (plotted as solid line). The best short-term deviation from a point-lens showing a brief planetary anomaly lasting for less alternative model is a binary source star but is re- than a day observed by four telescopes. The lens jected by the data. The dashed line is the point source light curve was on record (see Figure 4). star is a ~ 0.22 MA Galactic Bulge M dwarf orbited point lens model without the planetary deviation by a ~ 5.5 Earth-mass planet at ~ 2.6 AU with a (Beaulieu et al. 2006). Frenetic modelling activities started and it became clear very quickly that we had around the probed stars. The first at- tected by microlensing, two of which discovered a low-mass planet. The analy- tempts were done on individual high- are Jupiter analogues (Bond et al. 2004, sis has proven to be rather straightfor- magnification events using a point-source Udalski et al. 2005) and one Neptune ward for this event involving the transit of approximation, and then were applied (Gould et al. 2006), where the perturba- a large source star over a planetary caus- to a sample of the 42 well-covered micro- tion is due to the central caustic, and tic (Figure 5). The modelling of the photo- lensing events acquired by PLANET in one rocky/icy planet of 5.5 Earth mass metric data yields the mass ratio q = 7.6 1995–1999 (Gaudi et al. 2002). Less than named OGLE-2005-BLG-390Lb (Beau- ± 0.7 10–5, and the projected planet sepa- 1/3 of the lenses are orbited by Jupiters lieu et al., 2006) via a planetary caustic. ration d = 1.61 ± 0.008 (in units of RE, with orbits in the range 1–5 AU. We are They are overplotted on Figure 7. So the the Einstein ring radius). We performed a currently working on an analysis combin- era of discovery of frozen Super-Earths Bayesian analysis using Galactic mod- ing 11 years of data (1995–2005). We has been opened by the microlensing els and a mass function in order to derive calculate the detection efficiency of each technique. One of the consequences of probability distributions for the lens pa- microlensing light curve to lensing com- the discovery of such a small planet, is rameters (see Figure 2 from Beaulieu et panions as a function of the mass ratio that these small rock/ice planets should al. 2006) and a constraint on the nature and projected separation of the two com- be common. of the lens (low-mass main-sequence star ponents, now taking into account ex- or stellar remnant). The median values tended source effects. We use the same +0.21 yield a host star of mass 0.22 –0.11 MA lo- Bayesian analysis as for determining Obtaining more information about these cated at a distance of 6.6 ± 1.1 kpc within probability densities for the lens star and planets +5.5 the Galactic Bulge, orbited by a 5.5 –2.7 planet properties (Dominik 2006). Fig- Earth mass planet at an orbital separation ure 7 gives the mean detection efficiency Unlike other techniques, microlensing +1.5 of 2.6 –0.6 AU (Figure 6, and the artist of PLANET combining 14 well-sampled does not offer much chance to study the view of the planet by Herbert Zodet in events from 2004. For Jupiter-mass planetary system in more detail because ESO Press Release 02/06). planets, the detection efficiency reaches the phenomenon only occurs once for 50%, while it decreases only with the each star. Only a significant statistical square-root of the planet mass until the sample will allow us to reach firm conclu- Planet detection efficiency detection of planets is further suppressed sions and finally answer the question of by the finite size of the source stars for how special is our own Solar System. Ad- The detection efficiency of the experi- planets with a few Earth masses. Never- ditional information about a specific event ment can be determined from all collect- theless, the detection efficiency still re- can be obtained once the lens star is di- ed data, and comparison with the detec- mains a few per cent for planets below rectly detected. Here, we must wait many tions (or the absence of such) allows 10 Earth masses, made of rock and ice. years till the relative motion of the source conclusions about the planet abundance As of today, four planets have been de- and lens stars separate them on the sky.

36 The Messenger 128 – June 2007 Figure 6: Perturbation of the image of the source star by the planetary caustic observed with an ideal telescope. The time interval between the images is 0.1 day. The image scale is 30 microarcsec (Bennett and Williams).

0.10

0.05 E θ /

2 0.00 θ

–0.05

–0.10

0.90 0.95 1.00 1.05 1.10

θ1/θE

Figure 5: The source star and the planetary caustic of OGLE-2005-BLG6390. Notice the small size of the planetary caustic compared to the source star.

%

50 % 5 2 Bennett et al. (2006) using HST images 1000 have detected the lens star in the micro- lensing event OGLE-2003-BLG-235/ ) 25%

MOA-2003-BLG-53, and therefore the Earth 5 % uncertainty on the planetary parameters % have been greatly reduced. This could 100 5 be achieved too with HST or adaptive op- tics for OGLE-2005-BLG-169. In the case 2% of the lens OGLE-2005-BLG-390La, the Planet mass (M observation is much more difficult since 2 % we would need to detect a K ~ 22 mag 10 object at about 40 mas (in five years) from a star that is 10 mag brighter. In the com- ing years, statistics about frozen Super- 0.1 1.0 10.0 Earth planets orbiting M and K dwarfs will Semi-major axis (AU) be obtained, complementing the param- eter space explored by space transit mis- is supported by the Agence Nationale de La Recher- Figure 7: Average detection efficiency of planets as a che. KHC work performed under auspices of US function of mass and orbital separation (assuming sions like COROT and KEPLER or ag- DOE, by UC, LLNL under contract, W 7405-Eng-48. circular orbits) in the 14 favourable events monitored gressive ground-based Doppler search, NJR acknowledges financial support by a PPARC by PLANET in 2004 (preliminary analysis) along like those using the CORALIE, SOPHIE PDRA fellowship and is partially supported by with the planets detected by microlensing as of 2006 and HARPS instruments. the European Community’s Sixth Framework Marie marked as points. Curie Research Training Network Programme, Con- tract No. MRTN-CT-2004-505183 ANGLES.

Acknowledgements References Relevant websites Microlensing follow-up observations would not hap- pen without the alert systems of the OGLE and MOA Alard C. 2000, A&AS 144, 363 PLANET: http://planet.iap.fr/ collaborations. We are very grateful to OGLE and Beaulieu J. P. et al. 2006, Nature 439, 437 RoboNet: http://www.astro.livjm.ac.uk/RoboNet/ MOA, and to the that support our sci- Bennett D. P. et al. 2006, ApJ 647, 171 MicroFUN: http://www.astronomy.ohio-state. ence (ESO, Canopus, CTIO, Perth, SAAO, Boyden) Bond I. A. et al. 2005, ApJ 606, 155 edu/~microfun/ via the generous allocations of time that make this Dominik M. 2006, MNRAS 367, 669 MOA website: http://www.phys.canterbury.ac.nz/ work possible. The operation of Canopus Obser- Gaudi B. S. et al. 2002, ApJ 566, 463 moa/ vatory is in part supported by a financial contribution Gould A. et al. 2006, ApJ 644, L 37 OGLE website: http://www.astrouw.edu.pl/~ogle/ from David Warren, and the Danish telescope at Griest K. and Safizadeh N. 1998, ApJ 500, 37 La Silla is operated by IDA financed by SNF. The Ida S. and Lin D. N. C. 2005, ApJ 626, 1045 RoboNet project is funded by the UK STFC. HOLMES Udalski A. et al. 2005, ApJ 628, 109

The Messenger 128 – June 2007 37