Table information for 'gdr3spec.ssameta'

General

Table Description: SSA Metadata for the Monte Carlo-sampled Gaia DR3 XP spectra

This table is available for ADQL queries and through the TAP endpoint.

Resource Description:

This is a re-publication the Gaia DR3 RP/BP spectra in the IVOA Spectral Data Model. It presents the continous spectra in sampled form, using a Monte Carlo scheme to decorrelate errors, elaborated in this resource's reference URL. The underlying tables are also available for querying through TAP, which opens some powerful methods for mass-analysing the data.

For a list of all services and tables belonging to this table's resource, see Information on resource 'Gaia DR3 RP/BP (XP) Monte Carlo sampled spectra'

Citing this table

@MISC{vo:gdr3spec_ssameta,
  year=2022,
  title={Gaia DR3 RP/BP (XP) Monte Carlo sampled spectra},
  author={Demleitner, M. and Andrae, R.},
  url={https://dc.zah.uni-heidelberg.de/tableinfo/gdr3spec.ssameta},
  howpublished={{VO} resource provided by the {GAVO} Data Center},
  doi = {10.21938/W:hWbPah3wBabuqewQULTA}
}

Copyright, License, Acknowledgements

This data is derived from public Gaia DR3 data. Please take note of ESAC's guide on how to acknowledge and cite Gaia results.

Resource Documentation

In Gaia's DR3, most BP/RP spectra come in “continuous” form only, that is, as coefficients of Gauss-Hermite polynomials. These can be turned into a “sampled” representation using GaiaXPy; however, since the errors are given in the form of covariance matrices for the polynomial coefficients, the errors in the resulting spectra are strongly correleated, which can sometimes result in artefacts in the signal.

To get approximately decorrelated errors and hence sampled spectra usable with less caution, we apply a scheme of Monte Carlo-sampling different realisations from the error model of the coefficients. Specifically, given the covariance matrix C defined through the Xp_coefficient_errors and Xp_coefficient_correlations column in the DR3 xp_continuous_mean_spectrum table, and noting that for a unit normal-distributed random variable u

holds (⟨x⟩ denotes the expectation), we can re-write the covariance matrix using a Cholesky decomposition into LL^T as

Hence, x = Lu is a realisation of the errors satisfying the covariance matrix. To come up with a sampled spectrum, we now draw (in this case) 10 samples of the coefficients and have GaiaXPy convert them to a sampled spectrum.

The source code we used for that is dr3_to_mcsampled.py.

To be on the conservative side of the resolution and the bandwidth, and also to keep storage requirements modest, we have chosen a relatively rough grid over the optical band, that is, bins of 10 nm over the spectral range between 400 and 800 nm.

Columns

Sorted alphabetically. [Sort by DB column index]

Columns that are parts of indices are marked like this.

VOResource

VO nerds may sometimes need VOResource XML for this table.