Comparing CUPID ClumpFind with other Implementations

The CUPID implementation of ClumpFind is a completely independent and total re-write, based on the description of the algorithm in the 1994 Williams, de Geus & Blitz paper. Consequently, it differs slightly from other implementations such as the IDL implementation available from Williams own web page at http://www.ifa.hawaii.edu/ jpw/). In particularly, several extra configuration parameters have been added by CUPID to provide more detailed control of the algorithm. If you want to do a direct comparison between the CUPID implementation of the ClumpFind algorithm and another implementation, then you should set these extra parameters to the following values:

• ClumpFind.FwhmBeam = 0
• ClumpFind.MaxBad = 100000 (or some other very large number)
• ClumpFind.MinPix = 5 (for 3D data - use 20 for 2D data)
• ClumpFind.VeloRes = 0
• ClumpFind.IDLAlg = 1

You should also note that the pixel co-ordinate system used by the two implementations differ, and consequently the positions reported for the clump peaks will also differ. The IDL implementation of ClumpFind uses a pixel coordinate system in which the first pixel (i.e the bottom left pixel of a 2D image displayed ``normally'') is centred at (0,0). This differs from both the FITS and NDF definition of pixel co-ordinates. In FITS, the centre of the first pixel in the array has a value of 1.0 on every pixel axis. In NDF, the centre of the first pixel has a value of $LBND(I) - 0.5$ on axis $I$, where $LBND(I)$ is an attribute of the NDF known as the ``pixel origin''. For an NDF which has been derived from a FITS file and which has been subjected to no geometric transformations, the pixel origin will be 1.0 on every pixel axis, resulting in the centre of the first pixel having a value of 0.5 on every pixel axis.

Some implementations of ClumpFind do not conform exactly to the description in the published paper. Specifically:

1. the way in which areas containing merged clumps are divided up between individual clumps can differ slightly

2. the restriction that all peaks must extend at least as far as the second contour level is sometimes omitted

3. the restriction on the minimum number of pixels contained within a clump is sometimes varied in value

CUPID provides an option to use either the published algorithm, or the algorithm implemented by the IDL package available from Jonathan Williams WWW site (as available at 28th April 2006). Setting the configuration parameter ClumpFind.IDLAlg to a non-zero value will cause the IDL implementation to be used. The default value of zero causes the published algorithm to be used.