The Gaia mission. Objectives, description, data processing F. Mignard. Observatory of the Côte d'azur, Nice.

Transcription

1 The Gaia mission Objectives, description, data processing ---- Observatory of the Côte d'azur, Nice. 1

2 Outline Introduction to Space Astrometry context The Gaia Mission overall objectives instruments & observations performances Gaia Data Processing Overall principles Challenges Organisation Conclusions 2

3 What is Astrometry? Astrometry deals with the measurement of the positions and motions of astronomical objects on the celestial sphere. Global or wide field astrometry Local or small field astrometry Astrometry relies on specialized instrumentation and observational and analysis techniques. It is fundamental to all other fields of astronomy. 3

4 Limitations of the classical approach System defined with equator and equinox precession and nutation modelling solution: fixed frame not linked to solar system ICRS Observations from the ground many stations needed to cover the sky disturbances from the atmosphere solution: go to space for gloabal astrometry Hipparcos Reference system based on stars problems with proper motions, multiplicity solution: distant sources already considered by W. Herschel & Laplace Adopted in ~ 1990 with ICRS and ICRF in 1998 and

5 Goals of Space Astrometry Primary Objectives not achievable from Earth Ascertain the distances of the stars stellar parallaxes for astronomers Define and materialise the inertial frame now based on extragalactic sources Secondary objectives Astrophysics with astrometry, photometry, spectroscopy stellar and galactic physics detection of extrasolar planets solar system dynamics Tests of fundamental physics in space based on light path geometry 5

6 Space Astrometry: Past & Present 1 mas = 5 nrad 10 µas = 50 prad A successful forerunner: HIPPARCOS (ESA) accuracy of 1 mas ~ a 1000 km The unfortunate followers accuracy of 0.1 mas ~ a 1000 km Roemer, FAME-1, FAME-2, DIVA, Lomonossov, AMEX ESA US US DE RU US Study phase JASMINE (Japan) in the IR Delayed > 2015 SIM (US) with 1 µas accuracy Funded launch NanoJasmine [4 mas], J-MAPS (US) [ 1mas] Gaia (ESA) : 25 µas ( a 1000 km) 6

7 G A I A 10 9 stars 10 V < 3 25 V = 15 mag ESA mission Launch: summer 2012 Mission : 5 yrs Photometry ( ~ 25 bands) Radial velocity Low resolution spectroscopy 7

8 10 µas: Incredibly small! 0.3 mm displacement on the Earth edge-on sheet of 2000 km 10 µas = 50 prad km displacement of a 100 mas/yr star in one hour 10 µas motion of a fast minor planet in 100 µs km 8

9 Assets of Gaia A single mission with three nearly synchronous data taking Astrometric, photometric and spectroscopic data GAIA is a scanning mission no pointing, no change in the schedule Uniform coverage of the sky Quasi regular time sampling over 5 years ~ 80 observations photometry, orbits of binaries, asteroids Survey mission sensitivity limited Internal and autonomous detection system to G = 20 Global astrometry of staggering precision Internal metrology, thermal and mechanical stability Experienced and motivated community in Europe after Hipparcos scientific and industrial 9

10 Which science with Gaia? Stellar physics Quasars & galaxies Exo-planets Reference frame 10

11 Parallax horizon for a G0V star (no extinction) 10 kpc 1% % Solar system Credit: ESA/NASA/CFHT/NOAO 11

12 Product summary 10 9 stars 10 6 V = 12, 30 x 10 6 V = 15, 250 x 10 6 v = 18 σ 8 µas V < 12, 25 µas V = 15, 250 µas V = / deg 2 ; max : 7.5 x 10 5 /deg x 10 6 radial velocities Stellar classification for all classes and types Variability analysis over ~ 10 8 stars stellar masses σ < 1 % Extra solar planets to 200 pc 3 x 10 5 minor bodies of the solar system, 100 masses ~ 5 x 10 5 QSOs + z + photometry, ICRF in the visible PPN γ to ~ 2x

13 How it works Instrument and observations 13

14 Gaia: The Spacecraft Thermal shielding Ø ~ 10 m Solar panels Telescopes focal plane Propulsion module Antenna (2-4 Mbit/s) 14

15 The Assembled Spacecraft 15

16 Gaia : telescopes and detector 2 off-axis telescope 1.45 x 0.5 m 2 aperture 35 m focal length M4M'4 beam recombiner single focal plane 106 CCDs 1 Gigapixel 0.93 x 0.42 m 2 16

17 RVS Instrument in the P/L 17

18 Focal Plane Assembly SM1-2 AF1-9 BP RP RVS WFS 420 mm 0.69 WFS BAM BAM FOV1 930 mm sec FOV2 0s sec 18

19 Multiplexing observations 106 CCDs, 938 million pixels, 2800 cm cm cm Wave Front Sensor Wave Front Sensor Blue Photometer CCDs Red Photometer CCDs Radial-Velocity Spectrometer CCDs Basic Angle Monitor Basic Angle Monitor Sky Mapper CCDs Astrometric Field CCDs Image motion Star motion in 10 s animation W. O Mullane 19

20 Sky coverage The Scanning law is optimized to explore the same area at more or less regular intervals: parallax, proper motion, variability, orbits The scan direction must allow alpha and dec measurements The along-scan speed must be constant Mathematically: a set of two differential equations Three independent rotational motions 20

21 Gaia : Scanning (diagrammes L. Lindegren) Motion of the spin axis Sky covered over 4 days Trajet de la direction de visée sur 4 jours Axe de rotation 4 rev/jour Trajet de l'axe de rotation sur 4 jours 45 Trajet du soleil sur 4 mois soleil Trajet de l'axe de rotation sur 4 mois Crédit : L. Lindegren 21

22 Sky coverage: Equatorial coordinates # Transits during the mission 22

23 Sky coverage: Galactic coordinates # Transits during the mission 23

24 Orbit and sun-aspect angle 150x10 6 km ~1.5x10 6 km L2 45º

25 25 Gaia on the shelves

26 Hard stuff ' already manufactured spectro mirror M1 Rb clock 26

27 Torus: key structural element The torus was completed in June segments brazed together in a 3.5 m diameter structure 27

28 Soyouz Launchpad near Kuru

29 29 Expected Performances

30 Astrometric accuracy: single observation Single observation accuracy orbit, solar system one field transit point source 30

31 Astrometric accuracy: end of mission in µas Sky-averaged standard errors for G0V stars (single stars, no extinction) V magnitude mag Parallax μas Proper motion μas / an μas Notes: Estimates calculated with the Gaia Accuracy Tool (courtesy J. de Bruijne, ESA) Radiation-damage effects on CCDs not fully taken into account Estimates include a 20% margin (factor 1.2) for unmodelled errors 31

32 Photometric accuracy Illustration with the full light band (G-band) highest time resolution 32

33 Radial velocity accuracy (EOM, km/s) V * K1 III <1 <1 <1 < K0V <1 <1 < G0V <1 <1 < F0V <1 < A0V < B5V RAVE : <V r > ~ 2 km/s, 9<I<12 P. Sartoretti et al.,

34 Data Processing Relevant figures 34

35 Few relevant numbers Data volume compressed telemetry raw data processed data and archives 250 Tb 100 TB ~0.5 to 1 PB Computational size 1.5x10 21 FLOPs crude estimate Computational power expected in the DPAC in ~ 2012 > 10 TFLOP/s 2 yr CPU for FLOPs Data transfer Downlink 50 GB/day Data exchange between ESAC and DPCs : challenging could rely to physical shipment! 35

36 Sources of data Gaia has three instruments with three data flow Astrometric CCDs Photometric CCDs in the BP/RP bands Spectroscopic data Data is organised in form of telemetry packets one must also add house-keeping data and orbit data DPAC provides also auxiliary data from the ground 36

37 Data volume Compressed telemetry down-linked 5 years, 2.0 Mb/s = 250 Tb = 30 TB Usable raw data on ground x ~ 5 = 100 TB (65 : SM + AF + phot, 35 : RVS) ~ 50 GB/day Real volume with intermediate processed data x ~ 5 10 = ~ 1 PB Numbers of image per second average typical max 19, , x

38 How big is 1 PB? 1 GB 1 TB 20 TB - 1 hour of video Text of all the books published per year (1 million) nearly storable on your HD Text of all the catalogued books (~ 30 the Library of Congress or British Library [from wikipedia] 50 TB Amazon DB 3 PB Digitized Library of Congress with images and audio documents 1 PB remains very big, not so much in storage but in data management 38

39 Who is producing large datasets? Observational data Earth and space sciences Astronomy and Astrophysics Space Physics Atmospheric Science Geoscience Ocean Science Experimental Laboratory Data CERN 39

40 Examples of Scientific Databases SDSS: 20TB of observations; Final Catalogue ~ 1 TB Planck : telemetry 100kbps, 10 Gb/day, Data Base ~1 TB VLT: 100 TB over the last 6 years CDS : VizieR (Catalogues) = 0.5 TB Aladin (images) = 6 TB daily flow = 4 GB/days CP violation at CERN ~ 50 TB/yr Human Genome Database: ~ 25 TB (for only 3x10 9 bases) SLAC: ~ 1TB/jour ; 1 PB stored and accessible CERN/LHC ~ 5 PB/yr, ~ 100 PB in 2020 SKA (Radio Astronomy) should generate > 10 PB/day of raw data in 2020!!! 40

41 Different complexities in the DP Data volume Computational volume Data entanglement Institutional constraints The sheer complexity of Gaia DP results from the combination of these four elements 41

42 Computational Size F = FLOPs Total FLOPs Initial treatment 2x10 15 F/days 4x10 18 Iterative astrometry 2x10 19 F/cycle 2x10 20 Image update 1x10 20 F/cycle 1x10 21 Spectroscopy 4x10 17 F/cycle 4x10 18 Photometry 5x10 18 F/cycle 5x10 19 Non single sources ~ F/cycle 1x10 17 Total for a 5-year data processing: 1.5x10 21 FLOPs 42

43 How big are FLOPs? Largest prime number found M 47 = 2 43,112,609 1 ~ 10 7 digits ~ 3x10 19 FLOPs Computing power of GIMPS ~ 25 TFLOP/s SETI : largest computation ever FLOPs, 1.4x10 6 computers, ~ 265 TFLOP/s available Output : 1 bit ( they are there ) ZetaGrid zeros of the Riemann ζ(s) FLOPs, 10 4 old x86 PCs, ~ 7 TFLOP/s, FAH (Folding at Home:: protein structure computation) 2x10 5 PCs, 1 PFLOP/s (1000 TFLOP/s) peaked in Sept 2007 VIRGO (Gravit. waves) annual processing ~ FLOPs for 10 6 sources > 2-3 DP centres 43

44 How big are FLOPs? FLOPs is big, but achievable with a good organisation, but 44

45 Comparison to Gaia Today big computations ~ FLOPs all in distributed computing over thousands units all with virtually no data handling or big storage needs all can be broken down into small independent pieces Gaia - big computation for today standards - cannot be fully setup into many parallel computations - involves a large data handling - data must be accessed chronologically or per source - computing power must be available in few centres 45

46 Computing power for Gaia Straight evaluation of the power required 1.5x10 21 /3years ~ 10 TFLOP/s Gaia number crunching on six data processing centres two with large clusters with a total of ~ TFLOP/s in 2012 full time availability part-time availability of Barcelona Super Computer MareNostrum ~ 95 TFLOP/s today Other limiting factors local I/O for every processing data transfer between secondary data bases at DPCs 46

47 Trend in computer power Total 1st 500th Power in FLOP/s between 1993 to

48 Data Processing Architecture 48

49 Major steps in the data processing 1/2 Initial and global processing : Data reception, preliminary attitude, source identification Calibration, attitude, reference system a global iterative processing is performed in this step solution over ~ 10 8 primary stars Update of the Main Database astrometric solution on x months satellite attitude instrument parameters like CCD scales, Basic Angle 49

50 Major steps of the data processing 2/2 Scientific processing : Processing for well-behaved sources astrometric solution for secondary stars photometry, variability detection and analysis analysis of spectroscopic data partly iterative for the wavelength calibration Special sources double and multiple stars ( > 10 8 sources) unresolved galaxies quasars (~ 5x10 5 ) solar system objects (~ 3x10 5 ) Astrophysical parameters extraction 50

51 Data Processing: General Breakdown Relativistic formulation Multiple systems Reference system/qsos Exo-planets Secondary stars Astrometric solution Galactic rotation model radial velocity Variability Solar system Astrophysical Characterisatio L, T eff, Fe/H.. Spectroscopy Photometry Calibration Attitude Iterative Astrometric Solution Astrometry Object Matching First Look best instrument param. orbit data SS ephemeris Cycle Intermediate catalogue on-board attitude Image param. PSF/LSF CCD data Daily 51

52 daily ~ 6 months irregular Raw data Unpacking Telemetry packets Science ESAC Telemetry Gaia S/C Ground station Telemetry First Look Reports Command file Telecommands ESA Telemetry, House Keeping Mission ESOC Raw data base F/L Results Orbit data, S/C data DP: Functional Architecture Image update Reference data Initial data treatment First look Image parameters Main data base version n GASS Simulation GOG Inst. Calibrations AGIS Ast. Valid. Objects Photom. Spectro. Variability Ast. Param. Gaia products version n-1 DPAC 52

53 Main Database concept and versioning DPAC system structured with hub-and-spokes A central Database hosted at ESAC Send data version n to Processing centres Received results versions n+1 from these centres Each Processing centre in charge of part of the processing 53

54 Nature of the algorithms The processing comprises hundreds of algorithms Some are genuine numerical procedure Inverse computations: typically model fitting to observed data Astrometric solution, orbit determination, calibration, attitude determination Direction computation Prediction of an observation, ephemeris computation, astrometric model Statistical validation, plotting Synthetic spectral libraries Many are closer to data handling with more combinatoric than numeric Data compression, automated classification, Object identification Match an observation to a catalogue star Detect any previously observed solar system object 54

55 One Example The Astrometric Solution 55

56 Basic astrometric model Absolute motion of Vega non rotating reference frame

57 Astrometric Core Solution Central Problem: For each of 10 9 observed celestial objects we want to determine six astrophysical parameters: Position on celestial sphere: α,δ Parallax (distance): π Proper motion: μ α, μ δ Radial velocity: v R at the µas level (π: <25µas@V=15, <7µas@V<10) using (in theory) no a priori knowledge of these quantities but derive them from observation data alone in a self-consistent manner 57

58 Astrometric Core Solution Each observed object entering the FOV transits 9 AF CCDs 9 elemental observations per object per field transit one observation: a CCD centroid of the image in pixel units ~10 9 objects in total In 5 years we will have ~80 FOV transits per sources transits elementary data 1 trillion observations 58

59 Astrometric Core Solution Need to determine > 5 x 1 billion unknowns from the incorporates observations using an observation model that Satellite attitude 40 x 10 6 unknowns Calibration parameters 1 million unknowns Global astrophysical parameters ~ 10 Could set up a system of equations that solves directly for the unknowns system is manifold over-determined Observation model not too complex Problem: Connectivity of the parameters 59

60 Astrometry: Block Iterative System inner iteration (non-linearity, outliers) GIS outer iteration (interaction with all other processes) Every 6 months S G new data (IDT) Improved old data (IDU) GIS iteration (S-A-C-G cross-terms) A C improved selection of primaries S = Sources A = Attitude C = Calibration G = System, relativity 60

61 one 5 5 matrix per source N = one band matrix per attitude interval source attitude calibration s 1 s 2 s 3... a 1 a 2 a 3... c [S-A] [S-C] K Δx = b [A-C] [C-A] source ( ) attitude ( ) calibration (~10 6 ) Filled Sparse Zeroes the non-zero elements here connect the s, a, c parameters and make the system hard to solve one matrix per cal. unit Credit: U. Lammers/ L. Lindegren 61

62 AGIS Convergence Parallax Errors in µas noiseless data µas Conjugate Gradient -2.0 Credit: U. Lammers

63 Data Processing Organization The DPAC 63

64 Boundary conditions for the Data Processing The DP is a task shared between the community and the project The project supports : the spacecraft, launch, operations data reception and archiving initial treatment and part of the core processing ESA does not fund the overall scientific processing but ESAC is significantly involved Community implications are nationally funded visibility needed by every partners national interests and priorities must be matched to the DP needs scientific results are the real drive for the scientific community 64

65 Formation of the DPAC: Context ESA has issued an Announcement of Opportunity released on Nov 2006 it deals with the Gaia Data Processing A Consortium has been formed to answer this AO DPAC = Data Processing & Analysis Consortium Forms the Science Ground Segment for Gaia must transform the telemetry data into science products a large catalogue of astrometry, photometry, spectroscopy stellar sources, QSOs, Solar system objects response compiled in a 700-page proposal Formally selected by ESA Science Program Com. in May

66 The founding [and funding] documents 09/11/

67 ESA & DPAC Responsibilities Mission operations centre Data acquisition Telemetry data SOC Data processing Final results positions proper motions parallaxes radial velocities magnitudes variability orbits, masses T eff, log g catalogue access 67

68 Main features of the Data Processing No duplication of the processing: a single pipeline overall methodology learnt from Hipparcos resources not available for a full duplication specific parts could be duplicated if needed this is also part of the validation activities No way to set up a dedicated institute S/W development distributed in many places strong overall coordination between thee developments H/W equipment in a small number of DP centres this is where the data is received and processed 68

69 DPAC Membership February members 24 Funding agencies 93% in the 10 largest

70 The DPAC in the mission overall chart European Space Agency Agences Nationales ESA Project Manager/Team ESA Project Scientist Data Processing Research Training Network Industriel Gaia Science Team Sous-traitants Scientific community Gaia People Finder recense environ 500 personnes (excluant l'industrie) 70

71 Steering Committee Gaia Project Team DPACE Gaia Project Scientist CU1 System Architecture CU: Coordination Unit Gaia Science Team CU2 Simulation CU3 Core Processing CU4 Object Processing CU5 Photometric Processing CU6 Spectroscopic Processing CU7 Variability Processing CU8 Astrophysical parameters DPC ESAC DPC Barcelona DPC Torino DPC Cambridge DPC Geneva DPC CNES DPC: Data Processing Centre 71

72 Simulation activities A single group (CU2) in charge of the overall simulation Goal: cover the simulation needs for the whole DPAC to assist in the development of the DP algorithms to validate the DPAC SW systems to help evaluate the ultimate performances of the mission The Simulation is coordinated with the rest of the DPAC: The development of the different modules of the simulator is done in consultation with the other CUs it takes into account their needs and their global priorities specific error models are also provided by the CU experts Requests for simulated data are sent in each cycle by the different CUs in DPAC 72

73 Conclusion Gaia is on its way for a launch in 2012 It will be injected to an L2 orbit after few weeks A commissioning phase will follow over two months Then the data acquisition will start for five years After mission completion three more years needed to finish processing Catalogue and documentation to be available to the community in 2020 One or two partial intermediate data releases are planned during operations