spglib-users — Mailing list for the spglib users

Re: [Spglib-users] A comparison of FINDSYM and Spglib symmetry finders

[Evgeny B. <ja...@ya...>]

Dear Dr. Togo san,

I have figured out the reasons of my mismatches. They were non-tuned
accuracies and skipping of magnetic moments (in my Spglib wrapper test
class, in particular), exactly as i thought. I noted that you have a
function spg_get_symmetry_with_collinear_spin in Spglib C API, which
is inaccessible from python extension. Now I removed magmoms
consideration (comparing only the "pure structural space groups") and
made 1e-04 accuracy for both programs. And so the new results do not
contain any group numbers mismatching at all. There are still some
cases where group settings do not match though. Most of them belong to
orthorhombic space groups (the most frequent case in my tests was Pnma
/ Pbnm, the cyclic transposition of basic translation vectors a, b, c
to c, a, b with Wyckoff position change c(z, x, 1/4) to d(z, x, y), an
example can be found in spglib tests folder).

Regards,
Evgeny Blokhin

Re: [Spglib-users] A comparison of FINDSYM and Spglib symmetry finders

[Atsushi T. <atz...@gm...>]

Dear Evgeny,

Thank you for your testing. This is important information to improve spglib.

In some case, spglib gives higher symmetry and some case lower
symmetry than FINDSYM, and often they are systematic as follows.

148 167
123 221
14 62
3 13
11 4
12 10

I also do my test for spglib using FINDSYM as the reference, and
usually FINDSYM is more correct than spglib. So I would appreciate if
you could send the non-matching structures directly to me if you don't
mind, to improve spglib. The symmetry operations can be compared with
VASP etc.

To tolerate relative large disturbance is important for the practical
use in computational material science, e.g., for crystal structure
exploring. The implementation to include symmetry tolerance is very
tough (than I hoped!) and the code is not clean. So I want to do my
best to improve it, too. So I appreciate to hear any problem on it.

Best regards,

Togo


On Tue, Aug 14, 2012 at 12:25 PM, Evgeny Blokhin <ja...@ya...> wrote:
> Dear Dr. Togo san and spglib users,
>
> here the results of a comparison of two symmetry finders are reported:
> FINDSYM/ISOTROPY by Stokes and Hatch and Spglib. They were obtained on
> a subset of available sample structures of mine. I hope, this would
> make an interest for you. The main result is that only 6% of the
> testing subset exposed some issues, and only half of these issues are
> complete group non-matching, the others are less important group
> settings differences. One of the reasons of non-matching is structure
> magnetism, which was probably treated differently by the codes.
> Additionally, the results sometimes may vary depending on accuracy
> settings. The complete log as well as a testing script are attached.
> For non-matching cases the first values correspond to FINDSYM, the
> second to Spglib.
>
> Regards,
> Evgeny Blokhin
>
> ------------------------------------------------------------------------------
> Live Security Virtual Conference
> Exclusive live event will cover all the ways today's security and
> threat landscape has changed and how IT managers can respond. Discussions
> will include endpoint security, mobile security and the latest in malware
> threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
> _______________________________________________
> Spglib-users mailing list
> Spg...@li...
> https://lists.sourceforge.net/lists/listinfo/spglib-users
>



-- 
Atsushi Togo
http://atztogo.users.sourceforge.net/
atz...@gm...

[Spglib-users] A comparison of FINDSYM and Spglib symmetry finders

[Evgeny B. <ja...@ya...>]

Dear Dr. Togo san and spglib users,

here the results of a comparison of two symmetry finders are reported:
FINDSYM/ISOTROPY by Stokes and Hatch and Spglib. They were obtained on
a subset of available sample structures of mine. I hope, this would
make an interest for you. The main result is that only 6% of the
testing subset exposed some issues, and only half of these issues are
complete group non-matching, the others are less important group
settings differences. One of the reasons of non-matching is structure
magnetism, which was probably treated differently by the codes.
Additionally, the results sometimes may vary depending on accuracy
settings. The complete log as well as a testing script are attached.
For non-matching cases the first values correspond to FINDSYM, the
second to Spglib.

Regards,
Evgeny Blokhin