In addition to the matching criteria listed in the previous subsections,
you can build your own by combining any of these.
To do this, take the two (or more)
matchers that you want to use, and separate their names with a
"+" character. The values* parameters
of the combined matcher should then hold the concatenation of the
values* entries of the constituent matchers, and the
same for the params parameter.
Two rows are then considered to match if the the match is successful
for all of their constituent matchers.
A variant form where the names are separated with a
"*" character instead of "+" may also be used.
In this case an additional constraint is applied requiring that
the distance measure (see below) is less than or equal to unity,
thus requiring the points in the notional scaled parameter coordinate space
to be within a unit hypersphere rather than a unit hypercube.
So for instance the matcher "sky+1d" could be used
with the following syntax:
matcher=sky+1d values*='<ra/deg> <dec/deg> <x>'
params='<max-error/arcsec> <error>'
tuning='<healpix-k> <bin-factor>'
This would compare positions on the sky with an additional scalar constraint. Rows are considered to match if both their
values*:
ra/deg: Right Ascensiondec/deg: Declinationx: Cartesian co-ordinate #1params:
max-error/arcsec: Maximum separation along a great circleerror: Maximum Cartesian separation for matchtuning:
healpix-k: Controls sky pixel size. Legal range 0 - 29. 0 is 60deg, 20 is 0.2".bin-factor: Scaling factor to adjust bin size; larger values mean larger bins
ra, dec positions are within
max-error arcseconds of each other along a great circle
(as for matcher=sky)
and
their x values differ by no more than error
(as for matcher=1d).
Using matcher=sky*1d instead would work the same way
but restrict the matches a bit further.
This example might be used for instance to identify objects from two catalogues which are within a couple of arcseconds and also 0.5 blue magnitudes of each other. Rolling your own matchers in this way can give you quite flexible match constraints.
When identifying the closest match
(e.g. find=best1 in tmatch2)
the "distance" measure is obtained by
scaling the distances from each of the constituent matchers
and adding these scaled distances in quadrature,
so that each element of the matcher has approximately equal weight.
Scaling is generally done using the maximum permissible match
radius (or equivalent), so the distance measure looks something like
d = sqrt([dA/max(dA)]2
+ [dB/max(dB)]2).
However the details are a bit dependent on which matchers you are combining.
If the "*" separator is used instead of "+"
in the matcher specification as described above,
this distance will always be <=1 for successful matches.
Note that in STILTS v3.0-9 and earlier, a linear unscaled distance measure was used here instead, which did not give very meaningful Best match results.