Using Python tools

Setting up PyCall.jl

PyCall.jl is used to call python funcition define the classes that can use the trained models are calculator.

Guides for setting it up can be found at its home page. On Linux, this is as simple as (from scratch):

conda create -n pycall_env python ase phonopy <other packages2> <other package2> ...
conda activate pycall
export PYTHON=`which python`
julia -e 'using Pkg; Pkg.add("PyCall"); Pkg.build("PyCall")'
unset PYTHON

The key is to build the PyCall.jl package with a PYTHON environmental variable that points to the interpreter of the virtual environment to be used for Julia.

You can verify if the environment is up and running by importing ase in the Julia REPL:

julia> using PyCall

julia> pyimport("ase")

Phonon calculations

Phonon calculations requires a fully relaxed structures, so one has to relax the structure first:

julia> builder = Builder("link.toml")  # Load the train model

julia> @show builder.state.iteration  # Check if the latest iteration has been detected

julia> res = read_res("<path to SHLEX file>")  # Read in the structure file

julia> calc = NNCalc(res, builder.cf, load_ensemble(builder))   # Construct a NNCalc object

julia> EDDPotentials.optimise!(calc |> VariableCellCalc)   # Optimise with variable cell shape

Finally, we run the finite displacement calculations:

julia> using EDDPotentialsTools

julia> EDDPotentialsTools.run_phonon(calc;outdir="phonon")

This writes the YAML files and FORCE_SETS and phonopy_params.yaml files into a new folder called phonon. One can use phonopy for further processing from this point.