Question on a different result between ovito and asap implementation

[jjhjm17]

Dear Dr, Peter Larsen,

I appreciate for developing the polyhedral template matching algorithm, which works well at finite temperature.

Using the algorithm, the local environment of a solid structure was analyzed. Interestingly, ovito and asap implemtation gave different results, and the difference of frequency was around 3 % of total atoms. (Anyway 3 % is a small value.)

Could you please confirm which value is better?

The sturcture is bcc titanium at temperature where bcc titanium is unstable, but hcp titanium is stable. Note that at the temperature where bcc titanium is stable, the proportion of atoms marked as hcp increases, probably because temperature is high.

Do you think that these atoms marked as hcp structures are significant?

The same rmse_max of 0.2 was used. Is there another parameter that should be adjusted?

1. asap implementation using python

asap3.version
'3.12.8'
Type frequency (max 1): (total 128 atoms)
bcc 0.867
hcp 0.102
fcc 0.016
none 0.016

2. ovito implementation using ovito gui version 3.7.5 (both free and pro version):
   Type frequency (max 1):
   bcc 0.898
   hcp 0.086
   fcc 0.016
   none 0.000

Attached is structure file.
bcc_Ti.zip

Because I do not know how to use this repository directly, asap and ovito are used.

The code for asap follows and is attached:
get_local_structure_asap.zip

#!/usr/bin/env python3
# polyhedral template matching. Is the local structure bcc or hcp or others?

import os
import numpy as np
from asap3.analysis import PTM
from ase.visualize import view
from ase.io import read

# Ref.: https://wiki.fysik.dtu.dk/asap/Local%20crystalline%20order

# for polyhedral template matching method
rmsd_max = 0.2  # Ang

visualize = False
# visualize = True

# VERBOSE = True
VERBOSE = False


def analyze_polyhedral_template_matching(atoms):
    ptmdata = PTM(atoms, rmsd_max=rmsd_max)
    num_atoms = len(atoms)
    # get unique structures and counts
    structures, counts = np.unique(ptmdata['structure'], return_counts=True)
    frequency = counts / num_atoms
    # Flip so that most frequent structure comes first.
    structures = np.flip(structures)
    frequency = np.flip(frequency)

    # print('Type 0 = none; 1 = FCC; 2 = HCP; 3 = BCC; 4 = Icosahedral; 5 = SC.')
    structures = structures.tolist()
    for index, value in enumerate(structures):
        if value == 0:
            structures[index] = 'none'
        elif value == 1:
            structures[index] = 'fcc'
        elif value == 2:
            structures[index] = 'hcp'
        elif value == 3:
           structures[index] = 'bcc'
        elif value == 4:
            structures[index] = 'icosahedral'
        elif value == 5:
            structures[index] = 'sc'

    print('  Type   frequency (max 1):')
    for index, value in enumerate(structures):
        print(f'  {structures[index]}   {frequency[index]:.3f}')

    if visualize:
        atoms.set_tags(ptmdata['structure'])
        view(atoms)

        # menu -> view, colors -> by tag
        # 3: bcc


print('The result of polyhedral template matching follows. \n')


atoms = read('bcc_Ti.cfg')
analyze_polyhedral_template_matching(atoms)

[pmla]

Hi @jjhjm17.
Thank you for raising this issue. ASAP uses an older version of PTM. The version in OVITO has had a lot of updates and bugfixes so I would use that one.

I will try to update the PTM version in ASAP at some point so that there are no discrepancies between them.

[schiotz]

I will try to update the PTM version in ASAP at some point so that there are no discrepancies between them.

Hi @pmla ,

I give it a try soon(ish). I am sure you have made a lot of improvements in the meantime. There are some hacks involved in merging the code with Asap, I think I have a recipe somewhere....
/Jakob

[jjhjm17]

Dear @pmla and @schiotz,

thank you very much for the reply, that ovito version is the newer version. Because the small difference is expected and is coming from bug fixes, the difference is acceptable for the purpose of analysis, at least for me. For analysis, both values agree closely within 5 % (a small value) and both versions are basically supported by publication, so for me it is sufficient.

Thank you very much!