Crystals Worked Examples

Chapter 3: Poor Quality Data - Tetraphenylene

3.1: Background

3.2: Analysis and solution

3.3: Refinement

3.4: But what of the e.s.d's?

This is a hard, well crystalline organic material. Melting point 232-235 C. The crystals are in the form of prisms terminated with brilliant pyramidal faces. It was selected as a potential test crystal for analysing data collection and processing strategies, since its internal symmetry permits non-crystallographic statistical tests to be applied. A crystal 0.04 x 0.05 x 0.17 mm was Araldited to a fine glass capilliary. A hemisphere of data was collected using the parameter settings selected by the Nonius COLLECT software. Further equivalent data sets were collected with exposure times doubled, halved and quartered.
The data we have here is for the very fast data collection. The full hemisphere of data took 48 minutes to collect (15519 reflections). The average redundancy is 3, with 4000 reflections having a redundancy of 5 or more.

Chemical formula C24 H16, Space group monoclinic C2/c.

Get started as in Exercise 1, with the following differences:

This time:
       Choose the workshop structure "Quick"
       the SHELX format input file is called veryfast.ins 
       choose F2 refinement instead of F. (Click Yes).
       the reflection file is called veryfast.hkl
       when the filter dialogue opens, change the minimum
            I/sigma(I) from 3.0 to -10.0 (i.e. only reject
            very negative data).

Foolishly ignore the "Initial analysis" selection, then solve the structure with SIR92, using the default settings. It will not be very successful.

Quit SIR, and when asked do not use the solution from Sir92. Close the advice dialog that follows.

Now re-select the "Initial Analysis" option in the GUIDE. (OK)

There may be a few moments delay while the systematic absences are loaded.

Select the "Absences" tab - you should see that they are fairly symmetrically distributed about 0,0.

Select "Sigma freq." - you will see that there are only about 500 reflections with I>3sigma(I).

Select "Wilson Plot" - the high-angle data don't make any sense.

Right-click a blue cross on the Wilson plot somewhere near rho ((sin theta/lambda)**2) = 0.35. Then click the "Reject data" menu item that pops up.

In the filter dialogue, round the (sin theta/lambda)**2 upper limit to exactly 0.35.

Close this dialogue, close the Initial Analyses window and then re-run SIR92, but this time click the radio-button that says "Filter Reflections using List28 conditions".

The structure should now solve.

Note that setting a minimum I/sigma(I) threshold instead of a resolution threshold will not help solve the structure. This is because while high-angle weak reflections are just noise, low angle weak reflections have high information content and are needed for the negative quartets.

The Guide will now invite you to perform Isotropic refinement.

Try it.

Because of the high R factor, the Guide will not advance past isotropic refinement.

Force anisotropic refinement.

There is little improvement in R, but if you enable ellipses in the model window they look fair.

Now choose "Filter Reflections" from the "Refinement" menu.

Change the I/sigma(I) threshold from -10.0 to 3.0, and then try some aniso refinement.

The R factor will be quite small, but in the Refinement summary tab, you will see that there are less than 3 reflections per parameter!

Accept the GUIDE's invitation to Add Hydrogen atoms.

You should see that the ones found in the difference map are reasonably near to the predicted positions.

Select more anisotropic refinement, but in the dialog box, also enable hydrogen position refinement.

You now have a stable refinement with no restraints yet an observation:parameter ratio of little over two.

Type the following into the CRYSTALS command line box (below the information tabs):

\DIST
E.S.D. YES
END

Each of the bond lengths will be listed, with its associated e.s.d.

There will be an optimum I/sigma(I) cutoff and weighting scheme to minimize the e.s.d's:

Cutoff too high => not enough data.
Cutoff too low	=> too much noise.

You can use the reflection filter dialogue box to experiment with different filters, and use the optimise weights dialogue to experiment with weighting.