Configurable code for solving atomic structures.
The diffpy.srfit package provides the framework for building a global optimizer on the fly from components such as function calculators (that calculate different data spectra), regression algorithms and structure models. The software is capable of co-refinement using multiple information sources or models. It provides a uniform interface for various regression algorithms. The target function being optimized can be specified by the user according to the data available.
Within the diffpy.srfit framework, any parameter used in describing the structure of a material can be passed as a refinable variable to the global optimizer. Once parameters are declared as variables they can easily be turned "on" or "off", i.e. fixed or allowed to vary. Additionally, variables may be constrained to obey mathematical relationships with other parameters or variables used in the structural model. Restraints can be applied to variables, which adds a penalty to the refinement process commensurate with the deviation from the known value or range. The cost function can also be customized by the user. If the refinement contains multiple models, each model can have its own cost function which will be properly weighted and combined to obtain the total cost function. Additionally, diffpy.srfit is designed to be extensible, allowing the user to integrate external calculators to perform co-refinements with other techniques.
For more information about the diffpy.srfit library, see the users manual at http://diffpy.github.io/diffpy.srfit.
The diffpy.srfit package requires Python 3.5 or later or 2.7 and the following software:
setuptools- software distribution tools for PythonNumPy- numerical mathematics and fast array operations for PythonSciPy- scientific libraries for Pythonmatplotlib- python plotting library
Recommended software:
Optimizations involving crystal structures or molecules require
diffpy.structure- crystal structure container and parsers, https://github.com/diffpy/diffpy.structurepyobjcryst- Crystal and Molecule storage, rigid units, bond length and bond angle restraints, https://github.com/diffpy/pyobjcryst
Optimizations involving pair distribution functions PDF or bond valence sums require
diffpy.srreal- python library for PDF calculation, https://github.com/diffpy/diffpy.srreal
Optimizations involving small angle scattering or shape characteristic functions from the diffpy.srfit.sas module require
sas- module for calculation of P(R) in small-angle scattering from the SasView project, http://www.sasview.org
We recommend to use Anaconda Python as it allows to install all software dependencies together with diffpy.srfit. For other Python distributions it is necessary to install the required software separately. As an example, on Ubuntu Linux some of the required software can be installed using
sudo apt-get install \ python3-setuptools python3-numpy python3-scipy python3-matplotlib
For other required packages see their respective web pages for installation instructions.
The preferred method is to use Anaconda Python and install from the "diffpy" channel of Anaconda packages
conda config --add channels diffpy conda install diffpy.srfit
diffpy.srfit is also included in the "diffpy-cmi" collection of packages for structure analysis
conda install diffpy-cmi
Another option is to use easy_install to download and install the
latest release from Python Package Index
easy_install diffpy.srfit
If you prefer to install from sources, make sure all required software packages are in place and then run
python setup.py install
You may need to use sudo with system Python so the process is
allowed to put files to the system directories. If administrator (root)
access is not available, consult the output from
python setup.py install --help for options to install to a
user-writable locations. The installation integrity can be verified by
changing to the HOME directory and running
python -m diffpy.srfit.tests.run
diffpy.srfit is an open-source software developed as a part of the DiffPy-CMI complex modeling initiative at the Brookhaven National Laboratory. The diffpy.srfit sources are hosted at https://github.com/diffpy/diffpy.srfit.
Feel free to fork the project and contribute. To install diffpy.srfit in a development mode, with its sources being directly used by Python rather than copied to a package directory, use
python setup.py develop --user
The source code in observable.py was derived from the 1.0 version of the Caltech "Pyre" project.
For more information on diffpy.srfit please visit the project web-page
or email Prof. Simon Billinge at sb2896@columbia.edu.