Detwinning without a known structure, and  such a high twin factor must 
be very unreliable .

If you have only IT1 and IT2, the detwin algorithm  gives  I1 = (TF*It1 
+ (1-TF)*IT2) / (1-2*TF)

(Maybe signs wrong there!)  Since (1-2*TF)  is almost 0, this is pretty 
inaccurate..

But if you have a structure already and can calculate I1(calc) and 
I2(calc), the detwinning keeps the calculated ration of 
I1(calc)/I2(calc) and gets a much better estimate of I1 and I2.

So it isnt surprising you P6 twinned map looks better than one 
calculated against detwinned P3 data.

Re the gap - I dont know.
Did you test to see if there was a pseudo translation vector or any 
other strange features to the data.
Eleanor







Benirschke, Robert C. wrote:
>
> Dear All,
>
> I had written earlier about my twinned crystal that was giving me 
> problems......thanks to all who responded, it was very helpful.  I 
> have another couple questions about the same crystal that is confusing 
> us.  The twinned crystals space group is P3.  Twinning was found using 
> the Yeates twinning server and the twin fraction of different crystals 
> ranged from .32-.48.  Crystals with the lower twin fraction diffracted 
> to a much lower resolution and the more twinned crystals diffracted to 3. 
>
> Using the data from the .48 twinned crystal I refined into the twinned 
> P6 data (what we believed to be the space group originally) as well as 
> the detwinned P3 data.  The result of this is that both are about the 
> same as far as R/Rfree (.28/.36) and the P6 is better as far as 
> structure geometry and density fit.  Why is this the case?  Does the 
> detwinning process introduce more error than if one just refined into 
> the twinned data?  Is truncate and the twinning server ever wrong in 
> their diagnosis of twinning?
>
> Secondly, this structure still has an area of missing density parallel 
> to the a-b plane which never resolves (1/6 at top and bottom along 
> c).  This is the case for twinned, detwinned, P6, P3, and everything 
> else I have tried.  It is also seen as P1, so we are not averaging out 
> weak reflections when scaling in different groups.  I have also 
> searched these areas with FFEAR which yields nothing substantial.  
> This area is where many long disordered chains point (nearly 30 
> residues), so one of our thoughts is that there are many different 
> conformations of these tails that point into this area.  Is this 
> possible?  It seems to me that this would be a situation that promotes 
> twinning as well.  Our second idea is that there is a translation of 
> 1/6 a unit cell along c that would allow the rest of the cell to be 
> resolved as long as that plane of empty density is filled with same 
> contents.  Would this be possible?  Any ideas of how to test each of 
> these possibilities?  Can anyone think of any other ideas.?
>
> I´m sure I´ve said things that are unclear, so if clarification is 
> needed please let me know.
>
> Thanks,
>
> Robert
>