Semantics of the _chemical_formula.sum and related data items

[vaitkus]

The problem

The _chemical_formula.sum data items is intended to report the summary formula. However, for some time now I have been wondering if it is:

1. Formula of the compound we expect to crystallise.
2. Formula of what was actually modelled.

That is, quite often some hydrogen atoms or highly disordered solvent moieties cannot be resolved and are therefore not represented by atomic coordinates in the model. Should they still be included in the formula?

I always assumed that option 1 was the correct one, and I think that most structures published in IUCr journals follow(ed) this rule. If I understand correctly, checkCIF also assumes option 1?

However, structures coming from other journals, especially those dealing with large organometallic complexes, often tend to ignore the smaller solvent molecules or even the unmodelled parts of the main molecules and simply calculate the summary formula from the atomic coordinates.

@nautolycus, @jamesrhester maybe you know which interpretation is the proper one?

Potential contradictions in the definitions

I am also asking, since in the DDLm version of the dictionary the definitions of other data items and their dREL code seem to now also lean more towards interpretation 2.

My line of thinking:

A. The _chemical_formula.weight definition states that:

    Mass corresponding to the formulae _chemical_formula.structural,
    *_IUPAC, *_moiety or *_sum and, together with the Z value and cell
    parameters yield the density given as _exptl_crystal.density_diffrn.

B. Then if we look at the definition of _exptl_crystal.density_diffrn:

Crystal density calculated from crystal unit cell and atomic content.

Looks good so far in support of option (1). However, if we look at the dREL code of this item:

_exptl_crystal.density_diffrn = 1.6605 * _cell.atomic_mass / _cell.volume

It is calculated from the _cell.atomic_mass and not directly from the chemical formulae.

C. The _cell.atomic_mass item seems to be calculated from the ATOM_TYPE loop.
Definition:

    Atomic mass of the contents of the unit cell. This is calculated
    from the atom sites present in the ATOM_TYPE list, rather than
    the ATOM_SITE lists of atoms in the refined model.

dREL:

    mass = 0.

    Loop t as atom_type  {

                   mass += t.number_in_cell * t.atomic_mass
    }
      _cell.atomic_mass = mass

D. Here, again, I am unsure whether the _atom_type.number_in_cell
should hold the number of atoms we managed to model, or the number
of atoms we expected to see in the cell. Judging from the dREL, it is
the former, since this number can be calculated from the ATOM_SITE
loop (is this compatible with the definition of _cell.atomic_mass?):

    With t as atom_type

    cnt =  0.

    Loop a  as  atom_site  {

       if ( a.type_symbol == t.symbol ) {

          cnt +=  a.occupancy * a.site_symmetry_multiplicity
    }  }
    _atom_type.number_in_cell =  cnt

From all this, it would follow that _exptl_crystal.density_diffrn is always true only if all the different chemical formulae were also calculated from the atomic coordinates. While this might make sense for the summary formula, the structural and moiety formulae deal with interatomic connectivity and are thus more likely to reflect the expected composition.

[jamesrhester]

The definition of _chemical_formula.weight as matching _exptl_crystal.diffrn_density is aspirational: if everything has been done correctly, the density calculated from the formula weight will match the density from the structural model: this is something that CheckCIF verifies. You'll note in the verification code that up to a one percent deviation is not considered worthy of an Alert, so it is not expected that these values will exactly match up.

I believe the ultimate intent of the original definitions, and before them the requirements for authors in Acta C (back in the 1980s) was to provide a cross-check that the there hadn't been a mistake in construction of the chemical formulae during preparation of the manuscript, or that nothing important had been left out of the structural model. If e.g. solvent is missing, then that may provoke an Alert, and the author would have to justify why missing the solvent doesn't matter.

The definition could be reworded to make it clear that the density calculated from the chemical formula weight, Z and cell parameters "is expected to" match the density based on the diffraction experiment.

We could potentially provide dREL in _chemical_formula.weight to parse the formula and then calculate the formula weight. Parsing strings in dREL is pretty painful, so nobody's done it.

[vaitkus]

The definition of _chemical_formula.weight as matching _exptl_crystal.diffrn_density is aspirational: if everything has been done correctly, the density calculated from the formula weight will match the density from the structural model: this is something that CheckCIF verifies. You'll note in the verification code that up to a one percent deviation is not considered worthy of an Alert, so it is not expected that these values will exactly match up.

Ok, thank you for the clarification. So if I understand correctly, interpretation 1 of the _chemical_formula.sum is the correct one. If so, I think it would be useful to have this explicitly specified in the definition (the same probably applies to *.structural, *.IUPAC and *.moiety formulae as well). My naive formulation of compound we expect to crystallise is clearly not the correct one, so suggestions for a more appropriate terminology would be welcome.

I believe the ultimate intent of the original definitions, and before them the requirements for authors in Acta C (back in the 1980s) was to provide a cross-check that the there hadn't been a mistake in construction of the chemical formulae during preparation of the manuscript, or that nothing important had been left out of the structural model. If e.g. solvent is missing, then that may provoke an Alert, and the author would have to justify why missing the solvent doesn't matter.

We use this information in a similar way in the COD, e.g. to locate unmarked or incorrectly marked disorder, to detect potential voids, etc. That is why I want to be really clear about the semantics of the formulae fields.

The definition could be reworded to make it clear that the density calculated from the chemical formula weight, Z and cell parameters "is expected to" match the density based on the diffraction experiment.

Rephrasing this would also be useful.

We could potentially provide dREL in _chemical_formula.weight to parse the formula and then calculate the formula weight. Parsing strings in dREL is pretty painful, so nobody's done it.

This would be nice to have, but this task can definitely be added to the TODO list. The important thing to me is that the semantics are clear.

And what about the _atom_type.number_in_cell field? Should it be representative of the actual coordinates or of the expected composition, i.e. should we be able to reconstruct the summary formula from Z and the ATOM_TYPE loop regardless of what is actually in the coordinate loop?

[jamesrhester]

And what about the _atom_type.number_in_cell field? Should it be representative of the actual coordinates or of the expected composition, i.e. should we be able to reconstruct the summary formula from Z and the ATOM_TYPE loop regardless of what is actually in the coordinate loop?

Given the dREL, the current dictionary considers _atom_type.number_in_cell to be derived from the atom site list, not the chemical formula. We can make that more explicit in the definition.