Saving and Loading Potentials

General Principles

Loading a saved potentials is only guaranteed if the Julia environment remains the same. A new project should therefore always work with a specified Project.toml and Manifest.toml. See out Pkg intro for a brief introduction and references to further details.

If the manifest changes, but the ACEpotentials version remains the same or a backward compatible update (cf semver) then in principle a saved potential should remain loadable. We cannot guarntee this but would consider it a bug if this is not the case.

Normally, we save the entire Julia environment together with a fitted potential. This way it should always be possible to reconstruct the environment and hence the potential. More details follow.

Saving JSON-specified potentials

If using the runfit.jl script, then an output folder is specified, where all information including the full model specification and model parameters are stored as a JSON file (together with other meta-information).

Loading a JSON-specified potential

Suppose the result of runfit.jl (or an analogous approach) is saved to path/result.json. If the original or a compatible Julia environment is activated, then

model, meta = ACEpotentials.load_model("path/result.json")

will return a model::ACEPotential structure that should be equivalent to the original fitted potential. The meta::Dict dictionary contains all the remaining information saved in result.json.

Recovering the Julia environment

At the moment, this process is not implemented, but the result.json file can loaded into a dictionary which can then be investigated to manually reconstruct the environment and then load the potential as described in the previous section.