Generating Dimer Structures

For the generation of machine-learned interatomic potentials dimer curves represent a source of information frequently included in a training set. In Workflow the generation of corresponding structures can be performed with the wfl.generate.atoms_and_dimers.prepare routine.

The example below illustrates its application to a system containing hydrogen and oxygen atoms. At first, we define an OutputSpec that will be used to handle the output, i.e. the structural data we are going to generate. For the generation itself the prepare() function is executed where the atomic_numbers arguments specifies which combinations of species to consider (here all combinations between H and O, i.e. H-H, H-O and O-O). The bond_lengths argument allows us to specify a characteristic length that is used to sample a range of dimer distances. While isolated atoms are another source of information frequently added to a training set, in this example we are interested in dimers only and, thus, set do_isolated_atoms=False.

from wfl.configset import OutputSpec
from wfl.generate import atoms_and_dimers


outputspec = OutputSpec(files='dimers.xyz')

outputs = atoms_and_dimers.prepare(
    outputs=outputspec,
    atomic_numbers=[1, 8],
    bond_lengths={1: 0.74, 8:1.21},
    do_isolated_atoms=False,
    )

With Workflow the generation of dimer structures can be as simple as shown in the example. However, additional arguments can be passed to the function for more tailored outputs—for instance by adjusting the range of dimer distances to be sampled.