Gaia VS-SA-ICAP working groups meeting 15/16 April 2004 Cambridge (These are rough notes written by Dafydd Wyn Evans and are intended as an aide memoire to those that attended. Also the main discussion points and actions have been listed. The VSWG web-site hosts many of the presentations given and further detail can be found there: http://obswww.unige.ch/~eyer/VSWG/Meeting2/Programme.html). 15/04/04 Coryn Bailer-Jones ICAP Overview Blind testing of photometry Optimization of Photometric Systems (latest) Heuristically designed filter system Figure of merit for filter systems Classification algorithms for GDAAS Binary stars ICAP-CBJ-007,-011 have improvements to algorithms Calibration stars Mass-Radius-Temperature calibrations - feed in parallax Heuristic Filter Design Described A&A paper will be out shortly Future work/priorities Photometric System optimization (1 July) Classification algorithms (core data processing) On-board detection (star/galaxy/binary/etc.) Selection of Global Iterative Solution stars Improved classification algorithm Recognition of degeneracies Coping with weak Astrophysical Parameters Template mismatch problem Parallax tool What is overlap with VS and SA working groups? Organizational GDAAS Data processing Announcement of Opportunity Manpower, budget, etc. Q: Wyn - the sample of stars used in the tests will influence the filters. Anthony - sample selected is based on Gaia priorities so that filters will be appropriate. Wyn Evans SA Overview Aim is that ground-based telescopes can do immediate follow-up. Supernovae (SN), microlensing, near Earth asteroids, etc. By September must deliver SN algorithm Correlate brightening objects with variable stars or nearby galaxies Simulations: 1 SN identification from MBP data alone 2 SN in galaxies (how does galaxy affect the detection?). 3 How much data is needed? 4 Periodograms What else needed? - catalogue of variable objects (stars and galaxies) Q: Floor suggested tackling problem in 3 stages 1 G band (since it has the most signal) 2 BBP 3 MBP (since it suffers from crowding) Laurent Eyer VS Overview Goals Members (core/associate, active/dormant) Design - Astrometry Photometry Radial velocities Scanning law (article out shortly) Provide algorithm for GDAAS2 "Bibliography" work - forecast of numbers of variables etc. After Nice - tasks Web-site Studies of astrophysical observations, other surveys, forecasts Signal processing Feed into Universe Model Algorithm for GDAAS2 Listed some tasks Future work Q: Floor 1. Reminded everyone that the window function of Gaia is very poor 2. Eclipsing binaries - many of these require manual intervention to analyse properly, but there will be too many of them Coryn Bailer-Jones Gaia Update Name is now Gaia not GAIA Final presentations by Astrium and Alenia-Alcatel (March) Phase B1 started April 2004 Phase B2 will start ~April 2005 GDAAS implementation of complexities revision of algorithm implementation schedule GDAAS2 has funding to end of 2004 New design New focal plane layout and RVS design (Karen mentioned that there may be some more funding for GDAAS) livelink has been improved Parameter database Conventions document (GAIA-ARI-BAS-003) 4 broad CCDs for BBP - layout of CCDs in Astro has been changed Spectro redesigned - PSFs are improved and vignetting reduced GDAAS (University of Barcelona, GMV, CESCA) Considerable progress made Complexities and problems causing delay in completion of tasks and slip in algorithm implementation schedule Major GIS test still pending There has been some discussion of whether GDAAS and simulation activites should be separated for manpower and independence reasons Annie Robin Bayesian Classification Method Described the Besançon model Use of model as a priori information (?) (DWE - is this like Chiu? Not really. It uses lots more information since using all info from Galaxy model) Future work: 3d extinction Gaia filter systems Isochrones & adjustable Star Formation Rate If the classification fails then perhaps give a Science Alert Classification may evolve during the mission Laurent Eyer Proposed VSWG algorithms Will visit Barcelona in June to look at implementation of algorithms in GDAAS (this is yet to be confirmed) Basic algorithm to be written by Sept 2004 Focus on 1 Detection and 2 period determination Showed the expected precisions of single transit for G, BBP, MBP Expect 82 transits over 5 years Proposal I: Variable detection like Hipparcos (4 types) Moments (mean, width, skewness, kurtosis) Outliers Abbe test Trend Problem is false detections Suggestion from Floor - perhaps ignore faintest 0.5-1.0 mag for alerts What test on what data? per CCD? per transit? combine the filter data? Proposal II: Covariance matrix Method to combine data Eigenvalues Proposal III: Variogramme/Structure function Q: FvL said that there were lots of software tools developed at Cambridge for Hipparcos that would be suitable for the accumulation of the above statistics. (Discussion after meeting - perhaps get SDSS data? Some of it is multi-epoch and it has 5 filters - might be useful for algorithm testing). Dimitri Pourbaix CID binaries: from SDSS to Gaia CID = Colour-Induced Displacement astrometry different in different passbands SDSS has 5 passbands with astrometry in all of them (done relative to r') If a CID, the order of "movement of" the images should be with wavelength ie. u g r i z. If the movement is caused by the actual movement of the image eg. asteroids, the order will be by filter position within the camera. Compare SDSS Gaia large errors small errors (astrometric) 1 observation lots 2D measurement 1D measurement GAIA-FA-002 GAIA duplicity detection - photocentric binaries For this work it is very important to have a very accurate calibration of the chromaticity effects. This is a good addition to the photometric outlier algorithm for detecting binaries. Q: Will need astrometric calibration per BBP passband for this to work. Can probably detect from a single observation. Probably don't need full astrometric solution, do it relative to G astrometry. (Coryn asked if anyone was taking any minutes or notes of the meeting? Since DWE was already taking notes for his own use, he was volunteered. Laurent was also asked to acquire the presentations given so that they can be placed on the VSWG website). Philip Willemsen Classifying binaries from MBP (single shot) Methods needs training Simulated single and double stars 2500 training set 10000 validation set 65% correct at G=15 (50% at G=20 ie. random) Some biases - single very blue stars tend to be misclassified Q: How is the training set used? eg should IMF be flat or realistic? How should mass ratios be modelled? Karen O'Flaherty Gaia Parameter Database http://astro.estec.esa.nl/GAIA/paramdb/ Maintained at ESTEC Data is validated by Science Team or appropriate working group Web based or offline (eg. via wget) Using wget and cron could be used as a method to update your own database automatically. GAIA-JdB-007 Proposal GAIA-UL-001 User Manual Q: Dimitri asked if rsync could be used instead of wget. End result of the discussion was that it wouldn't work. 16/04/04 Caroline Soubiran Catalogue of calibration stars for Astrophysical parameterization. Programme based on Astrophysical parameters Stellar physics: mass, radius, etc. Galactic structure: distances, proper motions, etc. Cosmology: abs luminosity of variables, ages, etc. Many new instruments and surveys to help Stellar atmosphere theory is improving POLLUX database high resolution, high S/N spectra from echelle spectrographs Missing - cool dwarfs with Teff < 5000K Problems are that they are faint and that classical detailed analysis can't be done (broad molecular features). These will also be the largest class of stars within Gaia. Some progress has been made Improved models - Allard Elodie @ Observatoire de Haute-Provence V<9 - few hundred observations Problems There are systematic differences between methods of determining Teff, logg, [Fe/H] eg. Teff can be different by 200K between authors. Q: Important to get angular diameters (VLBI) so that can get good Teff Vasily Belokurov Self-organizing maps for Variable Stars Classification must be fast (including the training) since will want to do this during the mission The is a reduction of dimensionality problem In case of 2d map it is many->two Visualization is also an important part of this method Doesn't need a training set to be pre-classified by humans (effectively a clustering algorithm). Hipparcos tests: data points (dimensions) come from power spectra of variables binned on a logarithmic scale (30-40 bins) + colour + amplitude Once you have generated a map, you can then calibrate it using known types of variability If you train with a different resolution of map (different number of nodes) you tend to get the same shape of map (other comments indicated that this may not be true). Example shown at beginning (2d data being mapped onto a 1d line) was good Using a supernova template (?), these will be mapped onto a different part of the map (if well sampled). Probably not useful for an alert since the classification is only correct after 100s of days due to Gaia sampling. But if you have good sampling you can get the class change occurring during the maximum Q: Piotr - how about alerting on classification change - most people's feeling was that this would be too noisy. Dafydd - another dimension that could be added is angular distance to nearest galaxy. This should help with supernovae detection. Laurent Eyer Classification of Variables Bayesian method (article out shortly) Antoine Naud has also worked on self-organizing maps Will check up on classification anomalies - unclassified objects and islands within the maps Q: Vasily - another advantage if using self-organizing maps is that errors can be incorporated fairly easily. This is not the case with neural nets. General Discussion The deadline is 1 September for the algorithms Classification Goal: Photometric System definition (with PWG) Variable Stars Goal: Algorithms for 4 statistics, definition of a constant star, perhaps some alert stuff Special Alert Goal: Supernovae detection. Also a study of SN going off in a faint galaxy - how does classification/crossmatching etc. change? Clarification of task boundaries: Classification identifies what is a single star and the VSWG then analyses the time series data. Another task identified is for VS and SA to improve the universe model of GASS Next meeting: Probably no meeting over the summer. October was suggested so that we can set new priorities. The Friday (8th October) of the Gaia meeting in Paris was suggested (9am to 5pm). Coryn will contact Meudon to try and arrange a room for us. Actions ------- DWE write up notes of meeting and give to Laurent to place on VSWG web-site LE collect presentations and place on VSWG website LE/DWE write VS algorithms for GDAAS (4 statistics and their accumulation and definition of constant/variable) LE/DWE improve variable star part of Universe Model in GASS NWE/VB write SA algorithms for GDAAS (Supernova detection) NWE/VB improve supernova part of Universe Model in GASS CB-J contact Meudon to organize room for next meeting