How to Make Your Own Potential?
===============================

This section explains how users can construct their own potentials, and what
to pay attention to when configuring the workflow.

1. Set up the workflow in a new directory:

   .. code-block:: bash

      make setup-workflow dir=my_potential
      cd my_potential

2. Inspect the ``bd.yaml`` files in each task folder in ``workflow/``.
   Most default parameters should be acceptable, but you may want to change
   some of them, for example the number of MD time steps in ``md_select``.

   Most probably, you also want to change the ab initio model. For testing
   purposes, an EAM potential is used, which is also the default option. For
   using VASP as the ab initio model, change the alias of ``sp_calc`` in
   ``motoko.yaml`` to:

   .. code-block:: yaml

      aliases:
        sp_calc: vasp

   The INCAR file must be located in the ``potentials`` directory; an example
   INCAR file is provided there. If you use a different INCAR file, change the
   filename in the ``bd.yaml`` of the ``vasp`` task.

   VASP is run through `ASE <https://ase-lib.org>`_, which requires the VASP
   executable and the path of the pseudopotential directories to be set as
   environment variables in ``launch.sh`` in ``workflow/vasp/``:

   .. code-block:: bash

      export ASE_VASP_COMMAND=...
      export VASP_PP_PATH=...

   More information on setting up VASP with ASE can be found in the
   `ASE VASP calculator documentation <https://ase-lib.org/ase/calculators/vasp.html>`_.

3. Start the daemon processes that execute the tasks:

   .. code-block:: bash

      make -C workflow create launcher

4. Start the workflow from the workflow directory:

   .. code-block:: bash

      cd workflow
      motoko orchestrator start --help
      motoko --verbose orchestrator start your_arguments

   AutoPot is designed so that creating different potentials in the same
   database should not be a problem if a unique ``run_name`` is added. To
   create more potentials, append a unique ``run_name`` to ``your_arguments``:

   .. code-block:: text

      --run_name my_potential874

   The workflow has finished when the terminal shows:

   .. code-block:: bash

      terminate: workflow successful

5. Find the final potential and training set.

   The final potential and training set, ``trained_pot.mtp`` and ``ts.cfg``,
   are stored in the last ``train`` task. If ``X`` is the number of the last
   ``train`` task, they can be found in ``train/BD-train-runs/run-X``.

   Alternatively, access them in Python:

   .. code-block:: python

      from motoko.workflow import Workflow

      wf = Workflow("motoko.yaml")

      last_run = wf.train.connect().runs[wf.train.connect().runs_counter - 1]

      mtp_fname = os.path.join(
          last_run.run_path, last_run.train_args["trained_pot_name"]
      )

6. Delete the database and kill all daemons:

   .. code-block:: bash

      make clean