Re: [ccp4bb]: I to F's

[Edward Berry <eaberry@lbl.gov>]

***  For details on how to be removed from this list visit the  ***
***    CCP4 home page http://www.dl.ac.uk/CCP/CCP4/main.html    ***



Bart Hazes wrote:
> 
> ***  For details on how to be removed from this list visit the  ***
> ***    CCP4 home page http://www.dl.ac.uk/CCP/CCP4/main.html    ***
> 
> > Scalepack2mtz does not convert I to F, it merely writes I, sigmaI into mtz
> > format.  Then TRUNCATE does the conversion.  Basically:  F = sqrt(I), but
> > the reference is in the truncate manual.
> 
> Well truncate is a bit more sophisticated than that. It considers the
> probability distribution of I as a normal distribution with an aveage of Imean
> and standard deviation of SigImean but with the restriction that I must be
> positive. Therefore it truncates (I guess that's where the name came from) the
> left tail of the distribution at I=0. 

I always felt that "truncate" was a misnomer- truncate is what one used 
to do before before French&Wilson- all reflections with negative intensity
were discarded, so the the negative tail of the distribution was chopped off, 
i.e. the distribution was suddenly "truncated" at zero. French and Wilson
do not chop the tail at all, but do some kind of histogram matching to a
theoretical distribution which Bart Hazes has described better than I 
understand how, and push the distribution to the positive side of zero.
By default scalepack truncates at -3sigma, and I believe scalepack2mtz
and "truncate" reject none of the reflections from scalepack. (Now if 
you get a high R-factor in the bin between 0 and 0.3 sigma, you have to
remember that most of these reflections were actually measured as 
negative intensities)

Perhaps the F&W algorithm was seen as the PROPER way to truncate your 
data, rather than cutting it off at zero, and so became known as truncate.
Still, whenever a program gives me an option to truncate or not,
my first inclination is to say no, do French and Wilson instead.

Ed