Guide to presenting complex results based on powder refinements

[jamesrhester]

Results based on refinements of models against powder data are inherently more complex than typical single crystal experiments. Data may be collected at multiple instruments and/or at multiple temperatures or pressures and/or involve samples with multiple components. Such results cannot be contained within a single CIF data block. A standard way of presenting such results should be defined by the IUCr/COMCIFS to simplify the work of software programmers.

A suggested approach has been developed. Please comment below.

[rowlesmr]

I think I broadly agree. I need to work through all the nuances to get them straight in my head.

One thing that that makes me a little uncomfortable (choosing the multipattern/multiphase example) is the lack of unique _pd_phase.ids.

Also, in the last example, looking at the key datanames for PD_PHASE_MASS and REFLN, rather than devolving these into the own blocks, couldn't they be condensed into the already existing structure blocks? Especially as there are 6 unique structure blocks for the 3x2 structures present in the data.

data_cr2cuo4_7k

_structure.id                cr2cuo4_7K
_structure.diffrn_id         7K
_structure.space_group_id    fddd
_structure.phase_id          cr2cuo4 #why do you need this when the next line is there?
_pd_phase.id       cr2cuo4

# blah blah

_pd_diffractogram.id       0H_00
_pd_phase_mass.percent     0.9888(4)

# The following assumes that _refln.diffractogram_id
# has been defined so that per-diffractogram results
# can be provided.

loop_
   _refln.index_h
   _refln.index_k
   _refln.index_l
   _refln.F_squared_meas
   _refln.F_squared_calc
   _refln.phase_calc
   _refln.d_spacing
  1    1    1    2923.7101  1848.8118    0.0    4.75465
  0    2    2    50887.9824 44176.2312 180.0   3.01930
  2    2    0    41129.5142 38146.8191 180.0   2.86244
  2    0    2    40719.7976 38182.2594 180.0   2.86141
# ...

# preferred orientation information also goes here

[merkys]

I have implemented several pieces of software for the Crystallography Open Database for handling and validation of powder diffraction data according to Toby et al. scheme. I have often encountered some difficulties which it would be good to offset with the new proposal:

1. Finding out which data block contains what. Per Toby et al., there are phase, diffractogram, link and publication data blocks. When phase:diffractogram ratio is 1:1, everything sits in the same data block. I used to rely on CIF data items to understand what goes where. Moreover, there were explicit crosslinks between phases and diffractograms which as well made it easier to understand which data block contains which data. Then the link data block contains links to all phases and diffractograms. The remaining data block usually was the publication one, usually sitting the first among data blocks in a CIF file.

2. Detecting relations. Explicit crosslinks are really useful. I would suggest retaining them. But again, among published PD CIFs missing or broken links were often found. What is more, PD CIFs lack explicit relation between publication data block and the rest ones. Thus upon concatenation it becomes difficult to understand which publication data block should be connected to the others.

3. Understanding data inheritance. Sometimes it was difficult for me to understand the scope of data items in publication and link data blocks. In short, does it mean the same if I copy all data items from publication and link data blocks to the other data blocks? Thus I think it would be nice to have explicit rules here.

[rowlesmr]

Re point 2, I think the introduction of _audit_dataset.id should go a long way in grouping together relevant blocks.

It could be that the "publication" block contains many id values, and there are then many experiments belonging to that paper.

.

For the rest, I think that the multiblock keys should do most of the heavy lifting? Will need to sit down and do a mapping of Brian's paper to the proposed methodology.

[jamesrhester]

Also, in the last example, looking at the key datanames for PD_PHASE_MASS and REFLN, rather than devolving these into the own blocks, couldn't they be condensed into the already existing structure blocks? Especially as there are 6 unique structure blocks for the 3x2 structures present in the data.

I see your point. As a structure = a particular phase at particular diffraction conditions, there is only one diffractogram a combination of structure and phase could be related to and therefore only one refln list and mass value. So I think I agree we could move the refln loop and pd_phase_mass to the structure/phase blocks by always including the _pd_diffractogram.id that a particular combination of structure and phase relate to. In fact all categories in row 2 of Table 2 go back there.

I don't think it's accidental that we've rediscovered the contents of a single crystal data block. A single crystal data block is, after all, just a single-phase measurement and a single temperature/set of diffraction conditions.

Looking at the Toby et. al. paper, we are moving the reflection loop (section 3.4.5) back into the chemical species blocks, which is no more of a divergence from TvDL than we already had (ie. in a separate block not envisaged by TvDL).

[rowlesmr]

Belay that last, I actually think PD_PHASE_MASS and REFLN could be given in the diffractogram block. Just looking at what I do with my TOPAS macros and pdCIFplotter...

Having said that, putting the RELFN loop in the phase block will help in linking the phase to the diffractogram in the eyes of the reader. The PD_PHASE_MASS loop in the diffractogram links the diffractogram to the phase, so it works to link both phase to diffractogram and diffractogram to phase in the eyes of the consumer.

I know that it all ends up in the same database setup...

[rowlesmr]

Some more thinking is happening. Thinking specifically about lab data and multiple wavelengths, the REFLN loop should have _refln.wavelength_id, which is a link to _diffrn_radiation_wavelength.id. This means you would need to also link the wavelength to the phase, if the refln table is in the structure block.

The next question is. what happens with a single structure from two diffractograms? One Mo and one Cu? (or syncho + neutron, or any other combination). You need to include two wavelength loops in the one structure block, and DIFFRN_RADIATION is a Set category. Sure, you can game it by synthesising a single wavelength loop that contains all the wavelegnths, but that is cheating

A diffractogram is collected with a single wavelength loop (anything from monochchromatic, to tube, to laue), and the _diffrn_radiation_wavelength.id values are just sitting there begging to be used.

[jamesrhester]

This means you would need to also link the wavelength to the phase, if the refln table is in the structure block.

I'm not sure I understand what is meant here. The wavelengths will be tabulated in the diffractogram block, why is phase relevant? But what is relevant is the need for REFLN to refer to the DIFFRN_RADIATION to which the wavelength id relates. So we need a new non-key dataname definition in the multi-block dictionary, _refln.radiation_id.

The next question is. what happens with a single structure from two diffractograms? One Mo and one Cu? (or syncho + neutron, or any other combination). You need to include two wavelength loops in the one structure block, and DIFFRN_RADIATION is a Set category.

In current dictionaries, DIFFRN_RADIATION_WAVELENGTH is a child of DIFFRN_RADIATION. According to our draft, these categories should be included with the diffractogram collected with that wavelength. We do not directly undertake to link a structure with one or more diffractograms. Instead a structure is associated with a phase, and a phase is associated with one or more diffractograms by having the phase and diffractogram appear together in one of the P X D datablocks (second line of table 2), which we are proposing to combine with the structure blocks.

[jamesrhester]

The next question is. what happens with a single structure from two diffractograms? One Mo and one Cu? (or syncho + neutron, or any other combination). You need to include two wavelength loops in the one structure block, and DIFFRN_RADIATION is a Set category.

I've thought some more about this. Given this scenario, we can't assume (as I thought we could) that given a phase and structure we can determine a unique diffractogram. Therefore we can't put REFLN into the Structure/Phase block and so we're back to the current draft, with the REFLN loop in a separate block.