Importing very large CIFs

[lnaglecocco]

Hi all,

I am attempting to use Pymatgen for the analysis of a supercell containing ~10,000 atoms. I am getting the following warning message:

UserWarning: Issues encountered while parsing CIF: Some fractional co-ordinates rounded to ideal values to avoid issues with finite precision.
  warnings.warn("Issues encountered while parsing CIF: %s" % "\n".join(self.warnings))

I understand the rationale for including this feature, but in my specific circumstance this will harm my analysis. I think that the from_file() method should have an optional argument to switch this off, i.e. force it to accept the atomic positions in use. Is there any way to do this? Could someone add this feature?

Liam

[janosh]

@lnaglecocco Just out of curiosity, could you explain how this affects your analysis?

The relevant code is in CifParser._sanitize_data. If you want to add a flag to returns data early before rounding finite precision fractional coordinates, I'd be happy to take a PR for that.

pymatgen/pymatgen/io/cif.py

Lines 543 to 563 in 666f90f

	# check for finite precision frac coordinates (e.g. 0.6667 instead of 0.6666666...7)
	# this can sometimes cause serious issues when applying symmetry operations
	important_fracs = (1 / 3.0, 2 / 3.0)
	fracs_to_change = {}
	for label in ("_atom_site_fract_x", "_atom_site_fract_y", "_atom_site_fract_z"):
	if label in data.data:
	for idx, frac in enumerate(data.data[label]):
	try:
	frac = str2float(frac)
	except Exception:
	# coordinate might not be defined e.g. '?'
	continue
	for comparison_frac in important_fracs:
	if abs(1 - frac / comparison_frac) < 1e-4:
	fracs_to_change[(label, idx)] = str(comparison_frac)
	if fracs_to_change:
	self.warnings.append(
	"Some fractional coordinates rounded to ideal values to avoid issues with finite precision."
	)
	for (label, idx), val in fracs_to_change.items():
	data.data[label][idx] = val

[lnaglecocco]

Hi @janosh , thank you for getting back to me so quickly.

I'm analysing the output of a big box refinement of neutron Pair Distribution Function data, running calculations for various parameters (bond lengths, bond length distortion index, Van Vleck mode calculations) using code I have written on top of pymatgen. My supercell is ~50 x ~50 x ~50 Angstroms with approx 10,000 atoms. If the positions of atomic sites are getting rounded during importing from a CIF to a pymatgen Structure object, then it seems to me that any subsequent calculations are going to be less precise. It's entirely possible I've misunderstood the error and perhaps this is a non-issue. If I've misunderstood, please let me know

The solution you describe sounds like it would help.

[shyuep]

I just pushed an option to allow this to be bypassed.

[lnaglecocco]

Hi Dr Ong,

Thank you for doing this. I can see you've made a change to CifParser with commit 8fa4dcd but in terms of the Structure.from_file() method I'm not sure how to fix your frac_tolerance parameter to a certain value. Would this require an additional commit to pymatgen.core.structure, to add an argument to this .from_file() method?

Liam

[shyuep]

I just released a new version. Once you update to 2023.1.20, you should be able to use Structure.from_file(frac_tolerance=0)

[lnaglecocco]

Thank you!

[lnaglecocco]

Hi @shyuep ,
I've updated pymatgen and it is working as you said it would. Thanks again for your help, and for this incredibly speedy fix!
Liam