[daip] Fwd: GOODS-N Number Counts

[Eric Greisen]

Glenn Morrison wrote:
> Hi Eric,
> 
> Could you review the email below and tell me if you agree with the 
> statements made about the GAIN parameter in SAD?
> 
> -Thanks
> 
> --Glenn
> 
> Begin forwarded message:
> 
>> *From: *Eduardo Ibar <ibar at roe.ac.uk <mailto:ibar at roe.ac.uk>>
>> *Date: *August 22, 2008 12:14:20 AM HST
>> *To: *Glenn Morrison <morrison at cfht.hawaii.edu 
>> <mailto:morrison at cfht.hawaii.edu>>
>> *Cc: *Mark Dickinson <med at noao.edu <mailto:med at noao.edu>>, Rob Ivison 
>> <rji at roe.ac.uk <mailto:rji at roe.ac.uk>>, Frazer Owen 
>> <fowen at aoc.nrao.edu <mailto:fowen at aoc.nrao.edu>>
>> *Subject: **Re: GOODS-N Number Counts*
>>
>>
>> Hi Glenn,
>>
>>>> I have just modified the factor GAIN to =1.0 (it was 0.1 before) in 
>>>> the IMMOD-SAD simulations I explained before. Basically, IMMOD is 
>>>> used to introduce point sources smeared by bandwidth smearing in a 
>>>> residual image and SAD is used to extract them under the same 
>>>> criteria of the catalogues. The GAIN factor is critical to extract 
>>>> faint sources and to not to overestimate the number of gaussian-fits 
>>>> per source. This implies completeness as a function of flux density 
>>>> is higher, which translates in lower LogN.
>>>
>>> Are you just inserting point sources or can you also insert a source 
>>> distribution size (which is more realistic)?
>>
>> Just point sources but bandwidth smeared as a function of distance 
>> from the pointing centre. You are completely right in saying it would 
>> be much better to assume a distribution in size, although which one? 
>> Windhorst et al. (1990) and Bondi et al. (2003) have provided an 
>> angular distribution before but they differ quite substantially. To 
>> assume a distribution from the GOODSN sources itself is tricky due to 
>> bandwidth smearing too. This is why I decided to introduce point 
>> sources only but then applying a resolution bias factor.
>>
>>>>> The GAIN factor is critical to extract faint sources and to not to 
>>>>> overestimate the number of gaussian-fits per source.
>>>
>>> What do you mean here about "overestimate the number of gaussian-fits 
>>> per source" and why GAIN is helpful in this respect?
>>
>> In previous source extractions I completely missed the importance of 
>> GAIN factor. It is one of those parameters by default I prefer not to 
>> touch but somehow it was set to 0.1 that time. Look at the attached 
>> images which compare different SAD extractions:
>>
>> top-left: ICUT = 1.0, GAIN = 1.0
>> top-right: ICUT = 1.0, GAIN = 0.1
>> bottom-left: ICUT = 0.1, GAIN = 1.0
>> bottom-right: ICUT = 0.1, GAIN = 0.1
>>
>> Clearly GAIN plays a key role in extended bright sources, but also 
>> (surprisingly) in some single sources that are completely rejected.
>>
>> On the other hand, it's also clear SAD is fitting significant numbers 
>> of sources by two Gaussians (the radio contours in the small image). 
>> To be honest, I do not really understand the reasons why it does it, but
>> basically GAIN = 1 does two things:
>>
>> 1) Less sources are split into multiples
>> 2) Less sources are rejected by SAD based on the rms of the fit
>>
>> cheers
>>
>> edo
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Daip mailing list
> Daip at listmgr.cv.nrao.edu
> http://listmgr.cv.nrao.edu/mailman/listinfo/daip

I agree that GAIN is important when computing with SAD in that decisions
that are based on error levels are adjusted in quadrature by GAIN times
flux to allow higher errors on strong sources as "normal".  But to say 
that the expected rms after the fit equals or exceeds the flux fit 
(which is what GAIN=1 says) is absurd.  The default GAIN is 0.1 for 
Clean and I thought that that would be about right in SAD as well.

Eric Greisen