Skip to content

Writing Jobs

Overview

Individual compute tasks are called jobs. You can think of these as distinct calculations you would submit to the job scheduler (e.g. Slurm). Refer to the Workflow Syntax page for more information on the syntax of jobs.

Examples

For an example job using an ASE optimizer, refer to quacc.recipes.tblite.core.relax_job.

For a representative job involving a code that uses an internal geometry optimizer, refer to quacc.recipes.vasp.core.relax_job.

Requirements and Recommendations

In quacc, jobs are defined as Python functions that are decorated with @job.

The following are typical guidelines for jobs:

  • The function inputs and outputs must be serializable with pickle.

  • The function generally should take an Atoms object as the input, along with any other keyword arguments needed to set calculator parameters and define the logic flow.

  • The returned object should typically be a dictionary of results, as defined in quacc.schemas. If a dictionary is returned, the keys cannot have "." in them.

  • Calculations should be executed using a runner from quacc.runners.

  • When given the choice between using an ASE optimizer and the electronic structure package's built-in optimizer, you should typically use the latter.

Schemas

A schema is a dictionary containing tabulated input and output properties from a calculation.

All available schemas are stored in quacc.schemas, and type-hints associated with the outputs are stored in quacc.types.

All schemas create quacc-compatible dictionaries from a given calculation and also will automatically store results in the user's database, if one is specified in the global settings.