[ccp4bb]: I to F

[Dmitriy Alexeev <dima@holyrood.ed.ac.uk>]

What the weak F's are for? For better refinement, OK! What is the major
danger of including them? Artificially low sigF's - not to overweight
the arbitrary assigned poorly measured F's. The estimate sigF=sqrt(sigI)
is as fine as any other, except for few (but very important for ML
refinement) cases when sigI is accidently too small - it happens if the
data redundancy is low. I suggest to introduce a lower limit for sigF
and derive it from the well measured reflections.
Indeed, in absence of systematic errors  and for well measured
reflections

sigI->const*sqrt(I) and  sigF=0.5*sigI/F.  Then:
sigF->const*sqrt(I)/F=const,  which is the estimate for min(sigF)

I compute this constant  and top sigF's up to this level. For good data
(statistical noise only) this const is really constant throughout all
the intensisty range (I tested it). Actually, the variability of
min(sigF) might be used as an indicator of systematic errors.
Dmitriy Alexeev, Edinburgh.
begin:vcard 
n:Alexeev;Dmitriy 
tel;fax:(44)01316507055
tel;work:(44)01316507047
x-mozilla-html:TRUE
adr:;;;;;;
version:2.1
email;internet:dima@holyrood.ed.ac.uk
fn:Dmitriy Alexeev
end:vcard