Spectroscopic analysis

GSPspec and MATISSE Automated stellar parameters determination

 Automated techniques of spectral analysis and classification are needed, in order to perform a rapid and homogeneous processing of the data and to allow an efficient scientific return.

Within the Gaia DPAC, this task is performed by the Generalized Stellar Parametrizer – spectroscopy (GSPspec) algorithm. The GSPspec top level Work Package (GWP-S-823, coorditated by A. Recio-Blanco, Nice, France) is part of the Coordination Unit 8 “Astrophysical Parameters” within the Gaia Data Processing and Analysis Consortium. It has to deliver stellar atmospheric parameters (effective temperature, surface gravity, global metallicity) and individual chemical abundances, together with their uncertainties, from calibrated Radial Velocity Spectrograph data (with robust time average and known flux covariances) on single stars. Ten people (five from Gaia group of Nice), including researchers and engineers, are currently involved in the GSPspec group.

The parametrisation problem of stellar spectra is examined in detail by this group, trying to optimize scientific interpretability and computational times. For this purpose, already existing and newly developed techniques are being tested.

The present version of GSPspec, already integrated in the general Gaia processing pipeline (see the Software Design Description document on the DPAC livelink GAIA-C8-SP-OCA-ARB-004) uses the MATISSE algorithm (Recio-Blanco A., Bijaoui A. & de Laverny P., 2006, MNRAS, 370, 141) to perform the stellar parametrization. The MATISSE method, developed in Nice, is connected to Local MultiLinear Regression methods. Other methods, including Nelder-Mead minimum of distances, neural networks and dichotomy decision trees are being tested and their results compared to the MATISSE ones. In this latter case, parameters are estimated by projections on relevant functions derived from a multi-linear regression.

 

The MATrix Inversion for Spectral SynthEsis (MATISSE) algorithm 

MATISSE is connected to Local Multilinear Regression (LMLR) methods. Its basic idea is to determine the stellar parameters by a simple projection of a input spectrum on set of vectors that are previously determined from a grid of synthetic spectra. MATISSE is based on a two steps scheme (see figure 1). In the frst one, a vector is associated to each parameter for a selected mesh in the synthetic spectra grid. This vector is determined so that its scalar product with the modeled data is
the most correlated with the input chosen parameter, whatever the others are. In the second step, observational data are crossed with these vectors, carrying out the physical parameter values.

 In addition to its application to the Gaia/RVS data, the Galactic Archaeology group of Nice who maintains and develops the MATISSE algorithm, is currently involved on different projects requiring the automated analysis of real spectra. Those different applications of the MATISSE method include observed standard stars catalogues (for calibration purposes) and large samples of VLT/FLAMES spectra (Gazzano et al. 2009, in preparation; Recio-Blanco et al. 2009, oral presentation at « The Milky Way and the Local Group, now and in the Gaia era », http://www.ari.uniheidelberg.de/mee...). In addition, MATISSE is currently applied to the analysis of the ESO’s high and medium resolution public archive (AMBRE project, responsible P. de Laverny), including the data from the FEROS, HARPS, UVES and FLAMES spectrographs.

MATISSE is coded in java and has a user-friendly interface (Fig. 2 and 3).

 
img article :
  • img titre : Schema_Matisse_detail
  • img descriptif : 

    Fig. 1. Example of the MATISSE two steps scheme for the automated determination of the three atmospheric parameters from an observed spectrum.

  • img titre :  matisseJavaInterface
  • img descriptif : 

    Fig. 2: Matisse java user-friendly interface (implemented by C. Ordenovic), before the algorithm execution.

  • img titre : matisseJavaInterface2
  • img descriptif : 

    Fig. 3: Matisse java interface after the execution of the programme, The extracted stellar parameters and the associated internal errors can be seen.