libAtoms / QUIP

libAtoms/QUIP molecular dynamics framework: https://libatoms.github.io
346 stars 124 forks source link

GAP training for Silicon, following the work from Phys. Rev. X 8 041048, 2018. #270

Open vishankkumar opened 3 years ago

vishankkumar commented 3 years ago

Hi,

I was trying to run the tutorial (make_crack_1.py), however due to some change in library structure there are a lot of import errors. Can you update the tutorial code for the correct libraries?

Thanks

jameskermode commented 3 years ago

I’m aware of this issue, but unfortunately it’s a lot of work to update the tutorial so it won’t be possible in the near future.

However, you can find newer versions of the scripts in the matscipy code:

https://github.com/libAtoms/matscipy/tree/master/scripts/fracture_mechanics

see make_crack_thin_strip.py and run_crack_thin_strip.py

vishankkumar commented 3 years ago

Aah okay, I understand. No worries, thank you for redirecting me to the script. I will have a look, however my primary interest in quippy is to learn how to create GAP + SOAP based Interatomic Potentials that can be used in lammps or any other code. I would like to start with a simple silicon system: generating the structure and training GAP on it like it was in Phys. Rev. X 8 041048, 2018.

jameskermode commented 3 years ago

In that case the tutorials at https://libatoms.github.io/GAP/ will be more relevant for you

gabor1 commented 3 years ago

OK, so for that, they best way to start is to take the Silicon PRX training data set and the command line that creates it, and start playing with it. the whole database is huge, so I recommend cutting it down a lot, and also cutting down on the number of sparse (representative, basis) points that the training uses, so you can get fast turn-around times. then as you test your potentials and get comfortable making them and adjusting parameters, you can increase your data set size and the number of basis functions in the fit. Post here if you encounter any trouble or have any questions.

-- Gábor

Gábor Csányi Professor of Molecular Modelling Engineering Laboratory, University of Cambridge Pembroke College Cambridge

Pembroke College supports CARA. A Lifeline to Academics at Risk. http://www.cara.ngo/

On 13 Jan 2021, at 12:44, Vishank Kumar notifications@github.com wrote:

Aah okay, I understand. No worries, thank you for redirecting me to the script. I will have a look, however my primary interest in quippy is to learn how to create GAP + SOAP based Interatomic Potentials that can be used in lammps or any other code. I would like to start with a simple silicon system: generating the structure and training GAP on it like it was in Phys. Rev. X 8 041048, 2018.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.

vishankkumar commented 3 years ago

Thank you @jameskermode and @gabor1 for the quick suggestion, I will use the Silicon PRX data and will post here if I face any problem.

gabor1 commented 3 years ago

the command line is recorded in the potential xml file, grep for "command"

-- Gábor

Gábor Csányi Professor of Molecular Modelling Engineering Laboratory, University of Cambridge Pembroke College Cambridge

Pembroke College supports CARA. A Lifeline to Academics at Risk. http://www.cara.ngo/

On 13 Jan 2021, at 12:58, Vishank Kumar notifications@github.com wrote:

Thank you @jameskermode and @gabor1 for the quick suggestion, I will use the Silicon PRX data and will post here if I face any problem.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

vishankkumar commented 3 years ago

@gabor1 , I downloaded the dataset and found the potential xml file in models/GAP/gp_iter6_sparse9k.xml and on doing grep command, I found this command: <command_line><![CDATA[ at_file=si_iter6_database.xyz gap={soap l_max=12 n_max=10 atom_sigma=0.5 zeta=4 cutoff=5.0 cutoff_transition_width=1.0 central_weight=1.0 n_sparse=9000 delta=3.0 f0=0.0 covariance_type=dot_product sparse_method=cur_points} default_sigma={0.001 0.1 0.05 0.0} config_type_sigma={liq:0.003:0.15:0.2:0.0:amorph:0.01:0.2:0.4:0.0:sp:0.01:0.2:0.4:0.0} energy_parameter_name=dft_energy force_parameter_name=dft_force virial_parameter_name=dft_virial config_type_parameter_name=config_type sparse_jitter=1.0e-8 e0_offset=2.0 core_param_file=glue.xml core_ip_args={IP Glue} gp_file=gp_iter6_sparse9k.xml]]></command_line>

I can see that it gives the parameters of soap descriptor and using the si_iter6_database.xyz database. However, I do not know where this database is located and this CDATA is a command that I can run on command line to generate a subset of this data or I am mistaken? Sorry for the basic question, I do not know about the structure of this database, I can see a lot of relevant parameters but do not know the exact form. I will explore the GAP webpage to look for more details in the meantime. Thank you in advance for your advice.