Weekly Progress of DP-GEN 20220829

[HuangJiameng]

What's new

• We plan to migrate the README to DP-GEN's document soon and are checking and updating the contents of the manual. Your suggestions are welcome! At the same time, considering that many users are confused about the configuration of parameter files while using DP-GEN, some examples have also been added to example, and you can refer to them to prepare parameter files. You can refer to the documentation of dpdispatcher for the parameters of machine.json.
  Several new volunteers to help us check the documentation together this week. They are researchers from various provinces and cities.
  You can follow DP-GEN's document (devel) to see our newly updated documentation. Contribution is welcome!

• add an error remind in dpgen_fp #903

• fix: equi_stress should have the same unit as stress #910

• add DPRc (model_devi_engine amber and fp_style amber/diff) arguments #911
  Close #772. Close #776.

• Patch to #867 and #881 about mass_map #893
  You can skip "type_map" or set it to "auto". Then DP-GEN will automatically generate relative atomic masses based on "type_map". If you want to use isotopes, you can specify the relative atomic mass yourself. Note that auto is made up of lowercase letters.

Issues

• The "user_forward_files" command for dpgen Init_surf is not working？ Please check, thanks. #889
  init_surf doesn't support forward_files yet.It will be fixed in pr Support forward files for init_surf #909
• [BUG] _autotest_make_interstitial_"AttributeError: element"_ #502
  Might be an element-type-dependent problem of pymatgen.
• [BUG] boto3 raise an exception if not configure aws region. #674
  Fixed in move some imports into runtime functions #894.
• [Feature Request] Reduce MD loop time #600
  One can use para_deg in machine.json to run multiple MD tasks on a GPU workstation.
• [Parameters] Why are there two "type_map" items in param.json? #541
  The first type_map is the one that DP-GEN actually uses. The second one is in default_training_param according to the template of DeePMD-kit's training parameters. It will be OK if you skip the second type_map in default_training_param. Whether you set it or not, the second "type_map" will be assigned the value of the first one.

Discussion

• On the issue of energy not converging #920
  If you find that there is a big difference between the total energy of the system and 02.fp/data.000/set.000/energy.npy, while this situation does not appear when comparing the force, here are some suggestions for you.

1. Is your test data in the training set? If not, this may be the case.
2. If the test data exists in the training set, please check whether the units of force and energy are written correctly.
3. Some software will turn the orientation of coordinates. Please check it.
4. You can also scan the potential energy surface with the trained model to see the difference between your machine learning potential and quantification.

• Minimum number of structures to label #908
  The value of the fp_task_min from the param.json file only works in the make_train step. In fp step, fp_task_min is '-1' in default. You can set it in model_devi_jobs part of the param.json file. See https://github.com/deepmodeling/dpgen/blob/master/dpgen/generator/run.py#L2465-L2471

• dpgen runs and ends normally, but the 02.fp file is empty. #915
  When using plumed in the model_devi process, "atomic relative 1.0" needs to be added in lammps input file. @hu will make a pull request to fix it. Thx to him!

• Fail to run dpgen after installation #922
  If you find that after installation, the command "dpgen -h" shows the following error when module Pymatgen has already been installed.

ModuleNotFoundError: No module named 'pymatgen.analysis.defects'

Please check the version of Pymatgen. Pymatgen version v2022.7.19 is the last version with pymatgen.analysis.defects module. An earlier version of Pymatgen is needed.

Dependency

Dpdispatcher#229

fix TOTP bugs and implement TOTP authentication

[AnguseZhang]

Cheers.