Ctrl+K

Calculators in Workflow

Calculators in Workflow#

In principle, any ASE calculator can be parallelized using Workflow. The parallelization happens at Atoms object level. That is, if we wanted to get single-point energies & forces on 16 Atoms structures and using 16 cores, all 16 Atoms objects would get processed at the same time, each on a single core.

Parallelize with generic.calculate()#

In ASE, we iterate over all Atoms objects, initialize a calculator, set it to an Atoms object and call it to evaluate energies and forces sequentially. In Workflow, with generic.calculate we define a way to initialize a calculator, define where from and to read and write configurations (ConfigSet and OutputSpec) and set a directive for how many cores to parallelize over.

The calculator has to be defined as a tuple of (Calculator, [args], **kwargs), for example

dftb_calc = (
    quippy.potentials.Potential, 
    ["TB DFTB"], 
    {"param_filename": :"tightbind.parms.DFTB.mio-0-1.xm"}
    )

Further see autoparallelization page and examples page.

File-based calculators#

ASE’s calculators that write & read files to & from disk must to be modified if they were to be parallelized via Workflow’s generic calculator. Specifically, each instance of calculator must execute the calculation in a separate folder so processes running in parallel don’t attempt to read and write to the same files. Workflow handles the files, as well as creation and clean-up of temporary directories.

Currently, [ORCA], VASP, QuantumEspresso and FHI-Aims are compatible with the generic calculator; CASTEP parallelization is accessed via calculators.dft.evaluate_dft().

Special calculators#

Finally, there is a (currently broken) non-conventional “Basin Hopping” calculator implementation for ORCA. BasinHoppingORCA() runs multiple single point evaluations perturbing the initial guess of the wavefunction each time. It returns the results corresponding to the global minimum and lowest-energy solution.

Contents