User Guide#

Settings that depend on each other#

Which settings apply can depend on others. The step’s dialogs show only the settings that apply with the current choices, and the same rules are used when a flowchart is built or edited without the editor (seamm-flowchart or SEAMM’s MCP server): a setting that would have no effect is refused, with the reason, and a value that contradicts another is refused too. See “Flowcharts without the editor” in SEAMM’s user guide.

A model chemistry that none of the installed programs offers is refused when the flowchart is built, not only when it runs. The basis set is a free choice and is not part of that check.

Machine-learned forcefields#

A machine-learned forcefield, xnn:MLFF@<model>, has two more settings, shown only for it:

  • Checkpoint: a checkpoint file or a committee’s directory, used instead of the model chosen above – e.g. one trained earlier in the same job. A path, relative to the job’s directory unless absolute, or a variable such as $checkpoint; it must be on the machine where the job runs. The model chosen above then only selects the program that runs it, so a new checkpoint or committee, not yet in any model directory, can be given with any of that program’s models.

    A committee is a directory of members m0, m1, …, each holding a best.pt, with at least two members; a committee.json in it may list the members to use, e.g. {"members": ["m0", "m1", "m3"]}. It runs as the committee-mean engine of the Active Learning step, which must be installed with SEAMM. See Committees in the xnn step’s user guide.

    The program names the model chemistry:

    • a checkpoint with a name of its own is called by its file stem, e.g. xnn:MLFF@water_mace for water_mace.pt;

    • a generic best.pt or last.pt, as training writes, is called after its directory, with the start of a hash of the weights, e.g. xnn:MLFF@x013_m1-3f9a2c;

    • a committee is called by its directory’s name or, when that is just committee, after the nearest directory above it that says more, skipping a job’s step directories, with the start of a hash of its members, e.g. xnn:MLFF@Job_012130-committee-5c20d7 for Job_012130/2/1/committee.

    So the name is that of the weights: retraining in the same place gives a new one.

  • Engine options: options for the model’s engine in this run, e.g. --eeq-reuse or --dispersion "{regime: large}" for xnn.

Forcefields#

Forcefields are model chemistries too, offered by the Forcefield plug-in: FF@oplsaa for a classical forcefield and ReaxFF@CHO_cho_2008 for a ReaxFF parameter set. They have no program, since the step that uses them – LAMMPS, or Packmol to type the molecules it packs – evaluates them itself, so the dialog shows no Program for them. A forcefield from one of your own files (personal: or local:) is sent with a job submitted to a JobServer.

Indices and tables#