VASP/Phonopy Phonon Workflow #137

JaGeo · 2022-05-23T16:30:09Z

Summary

This is a work in progress implementation of a Phonon Workflow with VASP and Phonopy. It is based on the work and experiences of @ab5424 in his Master's thesis (for atomate).

I'm still a long way from finishing the workflow, but I believe I've identified the most critical classes that need to be developed and have an idea regarding data transfer.

Of course, feedback is welcome now, but I believe it requires some more changes from my side before someone of you invests too much time.

JaGeo · 2022-05-23T16:33:19Z

As I know that @gpetretto and @davidwaroquiers are also working on transferring workflows from abiflows to atomate2, there might be already some ideas about a Document for PhononData. If so, I would be happy to hear about it.

utf · 2022-05-23T16:57:35Z

Thanks very much for this Janine!

I'll aim to take a proper look later this week.

Would you mind installing the pre-commit? This will automatically raise (and most likely fix) the linting errors. To install you can run the following in the root directory:

pip install pre-commit
pre-commit install

For this first time, you can run the pre-commit on all files:

pre-commit run --all

Afterwards it will get run automatically whenever you commit something.

codecov · 2022-05-23T16:59:51Z

Codecov Report

Merging #137 (faf9515) into main (e7e4133) will increase coverage by 1.48%.
The diff coverage is 92.16%.

@@            Coverage Diff             @@
##             main     #137      +/-   ##
==========================================
+ Coverage   73.51%   74.99%   +1.48%     
==========================================
  Files          49       52       +3     
  Lines        4349     4675     +326     
  Branches      686      736      +50     
==========================================
+ Hits         3197     3506     +309     
- Misses        980      983       +3     
- Partials      172      186      +14

Impacted Files	Coverage Δ
src/atomate2/vasp/sets/base.py	`67.66% <83.33%> (+0.08%)`	⬆️
src/atomate2/vasp/jobs/phonons.py	`84.61% <84.61%> (ø)`
src/atomate2/vasp/flows/phonons.py	`94.44% <94.44%> (ø)`
src/atomate2/vasp/schemas/phonons.py	`95.68% <95.68%> (ø)`
src/atomate2/common/jobs.py	`100.00% <100.00%> (+100.00%)`	⬆️

JaGeo · 2022-05-23T20:39:09Z

Thanks. Will do. As I said, it really still needs some days of work and testing but it probably helps to get some suggestions with regard to general design now.

gpetretto · 2022-05-24T08:39:37Z

Hi Janine, it is very good to see that you started working on this.

I have one general comment. I think that this would be a good example where the workflow does not need to be specific to a DFT code. Roughly the basic idea is that it would be fine to pass Makers from a different code to static_energy_maker and the others. Correct me if I am wrong, but it seems that the portion of the code strictly related to vasp is already relatively small, so it might be feasible to isolate it and abstract what is needed from a calculation with a specific code.

I don't know if @utf has any thought about this. Anyway, since the implementation of the abinit workflows is not ready yet, I would propose to proceed with the implementation for vasp and we will try to generalize it at a later stage when more interfaces with other codes are available. Maybe already trying to keep this into account and trying to separate the part that are specific to vasp from those that are not. Does this seem reasonable?

Concerning the output, we do not have anything specific for the phonons yet, sorry. However, based on how I usually work with phonons, for phonopy I think it would be important to have access to the Phonopy instance to regenerate the BS and DOS with other options or any other property(i.e. to load it more or less like with phonopy.load). Since all the parsed data is written in yaml format by phonopy (using PhonopyYaml) it should not be complicated to just store it in the database as well.

gpetretto · 2022-05-24T08:56:32Z

src/atomate2/vasp/flows/phonons.py

+        jobs.append(vasp_displacement_calcs)
+
+        # Computation of BORN charges
+        if self.born_maker is None:


I have two comments about this point:

Maybe you already planned to update this later, but for how it is written now the BECs are always calculated. However there are good reasons to skip this step (e.g. metals). It would be good that None means that BECs are not calculated.

with the born_maker: BaseVaspMaker = field(default_factory=StaticSetGenerator) it will enter this if only if born_maker is explicitly set to None. Also the default of the of born_maker will not include lepsilon. An option would be to define the born_maker:
born_maker: BaseVaspMaker = field(default_factory=lambda: StaticSetGenerator(lepsilon=True))

Thank you for pointing this out. I will add this option by using born_maker: BaseVaspMaker |None = field(default_factory=lambda: StaticSetGenerator(lepsilon=True)) similar to what is done for the bulk_relax_maker. I was actually thinking about doing this but somehow did not completely implement this option ...

JaGeo · 2022-05-30T13:37:56Z

As mentioned, I have added one of the suggestions by Guido and the pre-commit flags. I am happy to hear more. Otherwise, I hope that I can refine the workflow in the coming weeks.
In parts of the workflow, I am using directories to access the files to create the output data. I guess it would be much cleaner if I would transfer forces and born data as outputs. This would be something, where I would appreciate an opinion.

JaGeo · 2022-05-31T20:55:11Z

Switched to transfer via outputs for now. I think this should simplify extending the workflow to other codes as well.

utf

Hi @JaGeo, apologies its taken some time to look at this in detail.

Overall it looks great! This is exactly along the lines of what I envisioned. I definitely think using the outputs directly (rather than files) is a better approach. I don't think the amount of data transfer will be too bad.

I have two main comments:

We need to be careful about relying on objects inside of Flow Maker make functions. Any function that actually requires the objects should be wrapped in a job.
I think it would be useful to refactor the generate_frequencies_eigenvectors so that most of the work is done inside the pydantic model.

I've gone into a lot more detail in the code review.

If anything isn't clear, I'd be happy to discuss on here/email/zoom. Thanks again for this. I'm very excited to have a phonon workflow in atomate2.

utf · 2022-06-07T18:46:28Z

src/atomate2/vasp/flows/phonons.py

+    generate_phonon_displacements_kwargs: dict = field(default_factory=dict)
+    run_phonon_displacements_kwargs: dict = field(default_factory=dict)
+    born_maker: BaseVaspMaker | None = field(
+        default_factory=lambda: StaticSetGenerator(lepsilon=True)


This should be a VaspMaker rather than an input set generator. There is already a core job Maker that does what you need:

from atomate2.vasp.jobs.core import DielectricMaker

And then:

Suggested change

default_factory=lambda: StaticSetGenerator(lepsilon=True)

default_factory=DielectricMaker

utf · 2022-06-07T18:48:23Z

src/atomate2/vasp/flows/phonons.py

+        default_factory=lambda: DoubleRelaxMaker.from_relax_maker(TightRelaxMaker())
+    )
+    static_energy_maker: BaseVaspMaker | None = field(
+        default_factory=StaticSetGenerator


This should be a VaspMaker rather than an input set generator. E.g.:

from atomate2.vasp.jobs.core import StaticMaker

And then:

Suggested change

default_factory=StaticSetGenerator

default_factory=StaticMaker

utf · 2022-06-07T18:56:31Z

src/atomate2/vasp/flows/phonons.py

+        sga = SpacegroupAnalyzer(structure, symprec=self.symprec)
+        structure = sga.get_primitive_standard_structure()


The make function should be designed so that it accepts OutputReferences as well as the actual structure objects. This allows you to use the Maker in the middle of a flow. E.g.

initial_relax_job = MyCustomRelaxMaker().make(structure) phonon_flow = PhononMaker().make(initial_relax_job.output.structure)

The way the workflow is currently written, this will fail as SpacegroupAnalyzer will get passed an OutputReference object.

The way around this is doing everything that actually needs the real objects inside a job. E.g., you should make a new job that just converts the structure to the primitive cell and returns that. Something like:

@job def structure_to_primitive(structure, symprec): sga = SpacegroupAnalyzer(structure, symprec=symprec) return sga.get_primitive_standard_structure()

And then inside the make function:

Suggested change

sga = SpacegroupAnalyzer(structure, symprec=self.symprec)

structure = sga.get_primitive_standard_structure()

prim_job = structure_to_primitive(structure, self.symprec)

structure = prim_job.output

utf · 2022-06-07T19:00:03Z

src/atomate2/vasp/flows/phonons.py

+    symprec: float = SETTINGS.SYMPREC
+    displacement: float = 0.01
+    min_length: float = 20.0
+    conventional: bool = False


I'm a little bit iffy about converting the structure to the primitive cell without any option to disable this. Could you turn option into something more sophisticated. E.g., something like "symmetrize" where they options could be "primitive", "conventional", None. Where None disables any structure modification.

Agreed but then we should probably also make the computation of the band structure optional?

Potentially? Although it might be possible to use the AUTO option for PRIMITIVE_AXIS to automatically align the band structure whatever input structure you use?

Yes, I know this option. Need to check.

I would like to be able to use the kpath options from pymatgen. I am not sure phonopy uses the same convention for the primitive cell (actually think it does not).

I have thought about it further:
what we could do is allowing to switch to another kpath scheme than seekpath when the primitive cell in the corresponding convention is chosen for the phonon run. We could use seekpath as a default though.

utf · 2022-06-07T19:01:09Z

src/atomate2/vasp/flows/phonons.py

+
+        # Computation of BORN charges
+        if self.born_maker is not None:
+            born_job = StaticMaker(input_set_generator=self.born_maker).make(


If self.born_maker is a Maker then you don't need to initialise the maker like this.

Thanks! Understood!

utf · 2022-06-07T20:02:01Z

src/atomate2/vasp/jobs/phonons.py

+    return displacements
+
+
+@job(output_schema=PhononBSDOSDoc)


I think phonon band structures and density of states should be stored in the data store (for large objects). You can do this automatically using:

Suggested change

@job(output_schema=PhononBSDOSDoc)

from pymatgen.phonon.bandstructure import PhononBandStructure

from pymatgen.phonon.dos import PhononDos

@job(output_schema=PhononBSDOSDoc, data=[PhononDos, PhononBandStructure])

utf · 2022-06-07T20:06:11Z

src/atomate2/vasp/jobs/phonons.py

+    #     print(born)
+    if born is not None:
+        if not conventional:
+            primitive = [[1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, 1.0]]


I think phonopy now has tools to determine the primitive matrix automatically?

src/atomate2/vasp/jobs/phonons.py

utf · 2022-06-07T20:30:20Z

src/atomate2/vasp/jobs/phonons.py

+def get_phonon_object(
+    displacement, min_length, structure, sym_reduce, symprec, conventional
+):
+    transformation = CubicSupercellTransformation(min_length=min_length)


Rather than having this here (where CubicSupercellTransformation will get called multiple times at different parts of the workflow), could it be refactored so that the supercell matrix is calculated once at the beginning of the flow and then passed directly to the various other jobs throughout.

Yes, agreed. I will indeed change this.

JaGeo · 2022-06-08T08:10:30Z

Thank you very much for the detailed review. I will try to improve the code according to these suggestions. I hope I can manage to find time in the next 1-2 weeks :).

JaGeo · 2022-06-09T14:02:09Z

I will let you know once I have a next version. Still a lot of testing and debugging necessary and also work on the document.

utf · 2022-06-09T14:07:12Z

The changes you've made already look great! Happy to help diagnose any bugs should they come up.

JaGeo · 2022-06-10T11:11:59Z

src/atomate2/vasp/jobs/phonons.py

+    # I don't see it
+    # I think I am missing something in the code
+    phonon.save("phonopy.yaml")
+    phonon = load("phonopy.yaml")


@utf : I currently have a problem here. I can only get correct bandstructures with NAC correction when I save the phonon object and then reload. There should be another way but I currently don't find the bug. If you have any idea, I am happy to hear. I guess something goes wrong in the construction of the dynamical matrix.

I am attaching minimal example where it does not happen. I can currently not reproduce this bug outside of the workflow.
minimal_phonopy.zip

I'm afraid I can't see what is going wrong!

Can you confirm that you're using the same phonopy version outside of the workflow?

I will confirm this and check again but thanks already for having a look!

I have upgraded to the latest phonopy version. It is still happening. I think I will continue with this file creation part for now and try to finalize the document part first. And I tested the other script with the phonopy versions from this year. They all don't have any problems at Gamma.

This is the script that I am using for the workflow. I might need to try it in a completely new environment to make sure this is also not the reason for this problem.
start_workflow.zip
(There might be some differences in the q-points in the files. I know and I am on this but the problem is at Gamma and not affected)

I wonder if this could be related to the same issue I had to fix in sumo. Basically, the default symprec for generating the distortions is 1e-5 but I was applying a different symprec during the analysis. If I used the same symprec of 1e-5 when generating the band structure it fixed the issue.

It seems like the workflow is using the same symprec throughout but it is quite high. I wonder if it will fix things if you use 1e-5?

I will try! Thanks!

You were correct. It was the symprec... I have now introduced a second symprec setting.

I will try to further clean up the code in the next days and check the kpath setting.

JaGeo · 2022-06-17T23:13:33Z

JaGeo · 2022-06-18T10:33:27Z

@utf in case you have some time in the next weeks, I would be happy about feedback on the document part.

I think we need to store force constants, born charges etc. to easily develop workflows for Grüneisen parameters/quasi-harmonic approximation.

utf · 2022-06-23T17:51:50Z

src/atomate2/vasp/schemas/phonons.py

+    )
+    # could be optional and implemented at a later stage?
+    npoints_band: int = Field("number of points for band structure computation")
+    kpath_scheme: str = Field("indicates the kpath scheme")


I wonder if these calculation settings could all be grouped into a subfield. Something e.g., just "calculation_settings" or "phonopy_settings". As if more get added later this could become quite messy.

Have to think about this field again. Will try to include it in one of the next pull-requests. Overall, the workflow is quite complex and I have to go through it again several times with different test data.

utf · 2022-06-23T17:52:40Z

src/atomate2/vasp/schemas/phonons.py

+    )
+    # have to add all folders for the computations!
+    # TODO: should we add folder for born and optimization as well?
+    uuids: List[str] = Field(None, description="The uuids of the displacement jobs.")


Maybe you also want the UUID of the bulk structure to obtain the total energy?

src/atomate2/vasp/jobs/phonons.py

src/atomate2/vasp/flows/phonons.py

utf · 2022-06-23T18:03:20Z

I think we need to store force constants, born charges etc. to easily develop workflows for Grüneisen parameters/quasi-harmonic approximation.

I think its a good idea to store the force constants too. Maybe there could be a flag to disable storing them if people are want to be very space conscious or e.g., if you are only doing convergence calculations.

JaGeo · 2022-09-02T15:40:37Z

Sorry... Added it to the wrong workflow ....

JaGeo · 2022-09-02T15:49:35Z

Had to add seekpath to the optional requirements [phonons] as well. Everytime I install phonopy, I forget to install seekpath ...

JaGeo · 2022-09-02T16:02:50Z

Tests have passed. I would reply to any further comments later today or next week.

utf

Hi @JaGeo, a few more observations.

I'm really happy how this PR turned out. Thanks for all your edits. Personally, I think once you've made these changes we can merge this and then do the more substantial testing afterwards.

utf · 2022-09-05T08:37:41Z

.github/workflows/testing.yml

@@ -26,7 +26,6 @@ jobs:
          pip install .[strict]
          pip install .[tests]
          pip install .[dev]
-


Can you add this back in please.

utf · 2022-09-05T08:40:56Z

.github/workflows/testing.yml

@@ -57,6 +56,7 @@ jobs:
            pip install .[strict]
            pip install .[tests]
            pip install .[docs]
+            pip install .[phonons]


I had a thought. Could you also add phonopy and seekpath to the "strict" dependency set and specify the most recent version with ==. You can then remove this line.

This will mean that the tests are deterministic. E.g., if a new phonopy version is released that breaks the API, the tests won't start failing randomly.

pyproject.toml

utf · 2022-09-05T08:43:30Z

src/atomate2/vasp/flows/phonons.py

+        if set to True, supercell algorithm will first try to find a supercell
+        with 3 90 degree angles
+    get_supercell_kwargs: additional arguments
+        to determine supercell


This the type for this parameter is wrong, it should be dict. Also can you list the function which these parameters will be passed to. That way the user can look this function up to see what options are supported.

utf · 2022-09-05T08:44:14Z

src/atomate2/vasp/flows/phonons.py

+         you can only use seekpath with any kind of cell
+         Otherwise, please use the standard primitive structure
+         Available schemes are:
+         "seekpath", "hinuma", "setyawan_curtarolo", "latimer_munro".
+         "seekpath" and "hinuma" are the same definition but
+         seekpath can be used with any kind of unit cell as
+         it relies on phonopy to handle the relationship
+         to the primitive cell and not pymatgen


These lines are indented by 1 extra space.

src/atomate2/vasp/jobs/phonons.py

utf · 2022-09-05T08:55:06Z

src/atomate2/vasp/jobs/phonons.py

+        if True, symmetry will be used to generate displacements
+    symprec: float
+        precision to determine symmetry
+    use_symmetrized_structure: str»|None


Suggested change

use_symmetrized_structure: str»|None

use_symmetrized_structure: str or None

src/atomate2/vasp/jobs/phonons.py

utf · 2022-09-05T09:34:39Z

src/atomate2/vasp/jobs/phonons.py

+    born: Matrix3D
+        Born charges
+    kwargs:
+        additional arguments that are passed


Say where they are to PhononBSDOSDoc.from_forces_born.

utf · 2022-09-05T09:35:02Z

src/atomate2/vasp/jobs/phonons.py

+    displacements
+    structure: original structure for meta data
+    supercell_matrix: supercell matrix for meta data


Missing types and descriptions.

JaGeo · 2022-09-05T10:02:38Z

Thanks. Will go through all of the suggested changes today!
I am currently trying to setup everything on another computer so that I can hopefully test more compounds faster. I still sometimes have weird issues with custodian and VASP for extremely simple systems that I need to resolve ...

Co-authored-by: Alex Ganose <utf@users.noreply.github.com>

JaGeo · 2022-09-05T11:08:03Z

I hope I added all suggestions. I made one more change on the tests. I realized on the weekend that I had several useless statements in there ...

We could rerun the tests again. I hope that I changed the installation configurations correctly. I will be really happy if you don't need to restart the automatic tests by hand every time I make a mistake ...

JaGeo · 2022-09-05T11:17:07Z

Yeah, of course, more typos. I usually use the read-out-loud function everywhere to find such things... Now, I think it should be fine ;).

utf · 2022-09-05T11:26:56Z

I will be really happy if you don't need to restart the automatic tests by hand every time I make a mistake ...

I think once this is merged GitHub won't require me to do it manually any more, since you will officially be a contributors. 🤞

One final comment. Are you happy if I merge once that is resolved?

utf · 2022-09-05T11:25:12Z

src/atomate2/common/jobs.py

-        A structure.
-    symprec : float
-        The symmetry precision.
+        structure: Structure object


I don't think these should be indented an extra level.

And, tests have to be restarted, of course. Sorry.

JaGeo · 2022-09-05T11:31:08Z

Yes, I would be happy if this is merged. I will then run several test structures, use it in my research and start a new pull-request to fix potential bugs in case they show up.

utf · 2022-09-06T09:00:03Z

Thanks very much @JaGeo. This is an exemplar workflow!

JaGeo · 2022-09-06T09:03:35Z

Thank you so much

JaGeo added 4 commits May 16, 2022 23:37

first rough implemenation phonon workflow

4c898a5

Further additions to free energy part

28f8b43

add more flexibility

56b0643

further fixes along the workflow

c707bd3

JaGeo mentioned this pull request May 23, 2022

Feature suggestion: harmonic phonon workflow? #135

Closed

gpetretto reviewed May 24, 2022

View reviewed changes

make born optional and precommit

63f338d

switch to transfer via output

ce74c41

utf reviewed Jun 7, 2022

View reviewed changes

pass precommit

b113c70

JaGeo added 3 commits June 9, 2022 16:56

fixes

762c86c

more bug fixing and typo removal

4aac6c1

Fix born charge bug

02e3a8d

JaGeo commented Jun 10, 2022

View reviewed changes

new PhononBSDoc and inclusion of thermal matrices

39aa228

utf reviewed Jun 23, 2022

View reviewed changes

Fix parameters

26fe95b

Add seekpath

dbdad2c

utf reviewed Sep 5, 2022

View reviewed changes

JaGeo and others added 10 commits September 5, 2022 12:02

Update src/atomate2/vasp/flows/phonons.py

70b05fe

Co-authored-by: Alex Ganose <utf@users.noreply.github.com>

Remove useless lines from test

0d67c62

Merge branch 'main' of github.com:JaGeo/atomate2

94b341c

Update installation and test config

1e01005

Fix first documentation issues

52ad56c

Move functions to common

10aa4a0

Move functions to common

724a471

extend documentation

4a44736

Add further metadata

7fc36d4

add further data

50793f9

JaGeo added 3 commits September 5, 2022 13:10

One last tiny change on the documentation

131a1ae

add blank line

514eb16

Add a t to hat

9a4e98c

utf reviewed Sep 5, 2022

View reviewed changes

Fix identation issue

faf9515

utf merged commit 4dbeccb into materialsproject:main Sep 6, 2022

JaGeo changed the title ~~[WIP] VASP/Phonopy Phonon Workflow~~ VASP/Phonopy Phonon Workflow May 23, 2023

JaGeo mentioned this pull request Oct 16, 2023

Add Alex Bonkowski to contributors list #573

Merged

	default_factory=lambda: StaticSetGenerator(lepsilon=True)
	default_factory=DielectricMaker

	default_factory=StaticSetGenerator
	default_factory=StaticMaker

		sga = SpacegroupAnalyzer(structure, symprec=self.symprec)
		structure = sga.get_primitive_standard_structure()

-@job(output_schema=PhononBSDOSDoc)
+from pymatgen.phonon.bandstructure import PhononBandStructure
+from pymatgen.phonon.dos import PhononDos
+@job(output_schema=PhononBSDOSDoc, data=[PhononDos, PhononBandStructure])

	use_symmetrized_structure: str»\|None
	use_symmetrized_structure: str or None

VASP/Phonopy Phonon Workflow #137

VASP/Phonopy Phonon Workflow #137

Conversation

JaGeo commented May 23, 2022

Summary

JaGeo commented May 23, 2022

utf commented May 23, 2022

codecov bot commented May 23, 2022 • edited Loading

Codecov Report

JaGeo commented May 23, 2022

gpetretto commented May 24, 2022

Choose a reason for hiding this comment

Choose a reason for hiding this comment

JaGeo commented May 30, 2022

JaGeo commented May 31, 2022

utf left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

utf Jun 8, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

JaGeo Jun 8, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

JaGeo commented Jun 8, 2022

JaGeo commented Jun 9, 2022

utf commented Jun 9, 2022

JaGeo Jun 10, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

JaGeo Jun 10, 2022 • edited Loading

Choose a reason for hiding this comment

JaGeo Jun 10, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

JaGeo commented Jun 17, 2022 • edited Loading

JaGeo commented Jun 18, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

utf commented Jun 23, 2022

JaGeo commented Sep 2, 2022

JaGeo commented Sep 2, 2022

JaGeo commented Sep 2, 2022

utf left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

JaGeo commented Sep 5, 2022

JaGeo commented Sep 5, 2022

JaGeo commented Sep 5, 2022

utf commented Sep 5, 2022

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

JaGeo commented Sep 5, 2022

utf commented Sep 6, 2022

JaGeo commented Sep 6, 2022

codecov bot commented May 23, 2022 •

edited

Loading

utf Jun 8, 2022 •

edited

Loading

JaGeo Jun 8, 2022 •

edited

Loading

JaGeo Jun 10, 2022 •

edited

Loading

JaGeo Jun 10, 2022 •

edited

Loading

JaGeo Jun 10, 2022 •

edited

Loading

JaGeo commented Jun 17, 2022 •

edited

Loading

JaGeo commented Jun 18, 2022 •

edited

Loading