Overview

ACE1pack.jl has two purposes: (1) to import and re-export ACE1.jl, IPFitting.jl, JuLIP.jl with guaranteed version compatibility; and (2) to have several convenience wrappers for setting up the least-squares problem (ACE1.jl & JuLIP.jl) and solving it (IPFitting.jl).

A short summary of packages behind ACE1pack:

  • ACE1.jl specifies the parameterisation of interatomic potentials in terms of the (linear) atomic cluster expansion; it provides functions to generate invariant basis sets, and to evaluate the resulting interatomic potentials.
  • IPFitting.jl supplied the functionality for parameter estimation. It focuses purely on linear models and linear observations, but provides various tools to deal with the typical data to which interatomic potentials are fitted (total energies, forces, virials, etc) and the reading and transforming of training data. A broad range of solvers are available through this package.
  • JuLIP.jl is a simple molecular simulation code in pure Julia, focusing primarily on an infrastructure to develop interatomic potentials. It provides various generic functions on top of which all our packages on this page build.

General structure

The main convenience functions are:

  • make_ace_db() - make the design matrix ("ACE database") for the least-squares problem.
  • fit_ace_db() - fit the given ACE database
  • fit_ace() - make_ace_db() and fit_ace_db() in one go.

See Fitting ACE for more information.

All of these functions take nested dictionaries that specify various parameters in making ACE. For convenience, there are a number of *params functions exported by ACE1pack that return these dictionaries with complete set of parameters specified. These are:

  • fit_params() - highest-level parameters' dictionary, compatible with all of make_ace_db(), fit_ace_db() and fit_ace(), see [fit.md];
  • basis_params() - parameters for constructing various bases for the design matrix of the ACE least squares database, see Fitting ACE;
  • degree_params() - for specifying the degree of ACE basis, see Constructing Basis;
  • transform_params() - parameters to specify the transform for a given basis, see Constructing Basis;
  • data_params() - for reading geometries and to-be-fitted property values, see Handling Data;
  • regularizer_params() - to set up an extra regularizer, see Regularizers section in Solers;
  • solver_params() - to set up solver for the least-squares problem, see Solvers.

In addition, there are some utility functions:

  • save_fit() - save given potential to file, see Fitting ACE;
  • fill_defaults() - recursively fills in default values for any of the optional parameters that were left unspecified, see Helper Functions;
  • parse_ace_basis_keys() - for parsing "(element1, number)" -> ("element1", number)-type entries that were read in from .json or .yaml files, see Helper Functions;
  • db_params() - a subset of parameters returned by fit_params(), compatible with make_ace_db() only, see Fitting ACE;
  • load_dict() - reads in parameters from .yaml or .json format, see Helper Functions.

TODO

  • functions with different calls (fitacedb) have docs look correctly?