mir-group / flare

An open-source Python package for creating fast and accurate interatomic potentials.
https://mir-group.github.io/flare
MIT License
295 stars 71 forks source link

Trajectory extraction #375

Open KrishnanjanP opened 1 year ago

KrishnanjanP commented 1 year ago

Hello,

I am newly using flair. So the question might sound naive, but how the trajectory is extracted from the output file after the completion of a otf run ? I tried the script for the otf_parser to do it, but it shows the following issue: ImportError: cannot import name 'otf_parser' from 'flare'. Any clue will be helpful.

Thanks, Krish

cjowen1 commented 1 year ago

Hi Krish,

If you are using the ACE/SGP version, there should be a *md.xyz file provided that contains the trajectory.

Best, Cameron

KrishnanjanP commented 1 year ago

Hello Cameron, Thanks for the reply. I am using the SGP version, but there are no md.xyz or thermo.txt being written. The three file that are being written along with the checkpoint files are _dft.xyz, -hyps.dat, .out, and an empty .xyz file . I have using 'yaml' file as input as described in one of the tutorials.

below are the version details - python 3.9.18 (main, Sep 11 2023, 13:41:44) [GCC 11.2.0] numpy 1.22.4 ase 3.22.1 flare 1.4.1

Looking forward to your suggestion. Thanks.

best, Krish

cjowen1 commented 1 year ago

Has your OTF passed the first DFT calculation? Can you please provide more information, specifically if there are any errors being written?

KrishnanjanP commented 1 year ago

It seems that the run is finishing without any error but I noticed that it is calling DFT only once at the beginning. During the rest of the otf the pressure is remaining high. I am running it with NVTBerendsen. I am attaching the MD step vs temperature plot, and the yaml file as text. Thanks again for our time.

T test.txt

cjowen1 commented 1 year ago

Please try reducing both the std_tolerance_factor and update_threshold so that DFT is called more than once. We also recommend all users to use LAMMPS for MD during active learning due to more robust integrators and increased flexibility wrt ensemble choice.

YuuuXie commented 1 year ago

@KrishnanjanP

As @cjowen1 mentioned, we recommend users to use LAMMPS MD for active learning, then you can get the entire trajectory saved in the _md.xyz, and the DFT frames saved in the _dft.xyz. You can follow the tutorial which you can also find in the read me page

If you keep using the ASE MD engine, then you could take a look at our unit tests as examples: https://github.com/mir-group/flare/blob/master/tests/test_parse_otf.py, where you can use OtfAnalysis to parse information from the .out file

KrishnanjanP commented 1 year ago

Hello, Thanks for the suggestions. I am currently infact trying with lammps following the tutorial. I will comeback and close this issue once I succesfully run it. Thanks again for you time.

Best, Krish