April 2022

Note about libxc/libnxc unpolarized functional

In a libxc or libnxc correlation functional, the current dftsw implementation always uses the functional in spin polarized form in calculations.

In the case of spin unpolarized systems with GGA, XC_POLARIZED Ec[$\rho_\uparrow$, $\rho_\downarrow$, $\sigma_{\uparrow\uparrow}$, $\sigma_{\uparrow\downarrow}$, $\sigma_{\downarrow\downarrow}$] (where $\rho_\downarrow = \rho_\uparrow$) is equivalent of XC_UNPOLARIZED Ec[$\rho_{tot}$, $\sigma_{tot}$].

Neuralxc installation

In order to install neuralxc, a python virtual environment must be used. This ensures compatibility with the application and the versions of the python binary and other dependent applications (scikit-learn,pytorch,ase, etc). The virtual environment must be installed with python version 3.6 from the /usr/bin folder.
The command to create the virtual environment is:
virtualenv -p=/usr/bin/python3.6 virtual_env_name
where virtual_env_name is the name given to the virtual environment (a directory is created with that name).

Installing libnxc

This covers how to install Libnxc, a library that contains Machine Learned functionals to be used in DFT codes developed by Marivi Fernandez and Sebastian Dick, available in github https://github.com/semodi/libnxc

In order to use the library, the libtorch library from Pytorch is needed, it can be obtained here https://download.pytorch.org/libtorch/cpu/libtorch-cxx11-abi-shared-with...