[sdpd] Re: UPPW-5 solution - UPPW-6 problem

[Alan Coelho <alan.coelho...@attglobal.net>]

Thank you for your laissez-faire thoughts Armel,

I agree with around 90% of what you say which for the sake of discussion 
is a bit of a shame. In any case we have 10% to play with which I hope 
others in the list are content with.

>Successful indexing depends on so many parameters : ....
>compound crystallinity, cell complexity, data quality, software

I fully agree, but lets isolate the part commonly known as indexing; 
this typically comprises sitting down in front of a PC with a 
diffraction pattern (or d-spacings) and an indexing algorithm. If this 
part of the process can be improved then this would be progress.

>But life is not always so kind. Samples do not want to crystallize, .....

I have a feeling that the next life is going to be the same; but look on 
the bright side - deep down we known how to index better than those with 
"Grant-in-Aid" .
 
>About "exhaustivity", it does not mean "successful in all cases".
>It only means "considering all aspects of the problem" - eventually
>before failing ;-). In my opinion, not considering the possibility for
>an impurity presence would yet remove that "exhaustivity" label.
>Anyway, well-selected words are used for markettng and packing
>Science. Scientists are particularly careful with their publication
>titles... It is similar to the promotion of a new perfume.
 
This is my point, are the so-called old (but very relevant) algorithms  
"considering all aspects of the problem". And I am so glad that you 
brought up the perfume analogy; we all seem to be under pressure to 
promote something; as scientists its ideas and in my case I need to be 
careful. Much of my ideas end up in commercial software and in regards 
to being careful I ask you to consider the ideas that are always freely 
available.
 
>PS - If you wish some simulated pattern with known solution,
>this could be proposed also as one of the next UPPWs.
>Using some newly predicted structure, for instance... - even not yet
>synthesized ;-).
 
Lets get back to the issue; how best to determine if an indexing 
algorithm is performing well. Consider this, before someone writes an 
indexing algorithm they should first investigate the need to do so. I 
undertook such a task when I wrote driver programs for ITO, DICVOL and 
TREOR; and I mean drive them hard by feeding 100s of simulated data sets 
and rigorously changing their control parameters. The purpose of such a 
task is not to ridicule as I think these methods are brilliant but 
rather to see if there's a chink in their armor.
 
To unequivocally prove if other methods fill the gap is not easy in the 
context of UPPW; but I wouldl certainly particiapte if its feasible. To 
put and end to the indexing duel, dare I say - more ground breaking than 
the indexing Olympics hosted by Robin Shirley in Durbin, the following 
would be required:
 
Simulated data, 20 or more data sets say,
 
i) generated from difficult test lattice parameters which includes 
dominant zone problems with either one or two short lattice parameters,
 
ii) I would leave impurity lines out for starters,
 
iii) I would also not bother with proposed predicted lattices as they 
may be of a simple indexing nature.
 
iv) Random noise needs to be introduced into the patterns (or maybe 
d-spacings) and the peak half widths needs to be varied such that peaks 
overlap. This is the most difficult part as the true solution should 
show up in an unambguous manner by at least one algorithm.
 
I could easily generate such data but this maybe considered unfair as I 
would know the solutions.
 
In my opinion such a set of UPPWs would be more useful than an indexing 
round robin using "real" data designed to investigate the performance of 
algorithms. I am of course not advocating the abolition of round robins 
as they serve a broader purpose.
 
cheers
alan


------------------------ Yahoo! Groups Sponsor ---------------------~-->
Buy Ink Cartridges or Refill Kits for your HP, Epson, Canon or Lexmark
Printer at MyInks.com. Free s/h on orders $50 or more to the US & Canada.
http://www.c1tracking.com/l.asp?cid=5511
http://us.click.yahoo.com/mOAaAA/3exGAA/qnsNAA/UIYolB/TM
---------------------------------------------------------------------~->

 

Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/