||Towards a Next-Generation Catalogue Cross-Match Service
||495, Astronomical Data Analysis Software and Systems XXIV (ADASS XXIV)
||Pineau, F.; Boch, T.; Derriere, S.; ARCHES Consortium
||We have been developing in the past several catalogue cross-match tools.
On one hand the CDS XMatch service (Pineau et al. 2011), able to perform basic but very
efficient cross-matches, scalable to the largest catalogues on a single regular
On the other hand, as part of the European project ARCHES1,
we have been developing a generic and flexible tool which performs
potentially complex multi-catalogue cross-matches and which computes
probabilities of association based on a novel statistical framework.
Although the two approaches have been managed so far as different tracks, the need for
next generation cross-match services dealing with both efficiency and complexity is
becoming pressing with forthcoming projects which will produce huge
high quality catalogues. We are addressing this challenge which is
both theoretical and technical.
In ARCHES we generalize to N catalogues the candidate selection criteria
– based on the chi-square distribution – described in Pineau et al. (2011).
We formulate and test a number of Bayesian hypothesis which necessarily increases
dramatically with the number of catalogues.
To assign a probability to each hypotheses, we rely on estimated priors
which account for local densities of sources.
We validated our developments by comparing the theoretical curves we derived
with the results of Monte-Carlo simulations.
The current prototype is able to take into account heterogeneous positional errors,
object extension and proper motion.
The technical complexity is managed by OO programming design patterns and
Large tasks are split into smaller independent pieces for scalability.
Performances are achieved resorting to multi-threading, sequential reads
and several tree data-structures.
In addition to kd-trees, we account for heterogeneous positional errors and object's extension using M-trees.
Proper-motions are supported using a modified M-tree we developed, inspired from Time Parametrized R-trees (TPR-tree).
Quantitative tests in comparison with the basic cross-match will be presented.