This library implements accurate and efficient radial Schrödinger and Dirac finite element solvers. The formulation admits general potentials and meshes: uniform, exponential, or other. Additionally, a squared Hamiltonian approach has been used for the Dirac equation, which eliminates spurious states.
Detailed description of methods, convergence studies and implementation details may be found in the following article:
Čertík, Ondřej, et al. High-Order Finite Element Method for Atomic Structure Calculations. Computer Physics Communications, Volume 297, 2024, ISSN 0010-4655. https://www.sciencedirect.com/science/article/pii/S001046552300396X, http://arxiv.org/abs/2307.05856.
With the provided meshes, the solvers (both Schrödinger and Dirac) can converge to at least 1e-8 Ha accuracy (with double precision of approximately 16 significant digits) for all eigenvalues and total DFT energies for all atoms up to uranium (Z=92).
The converged nonrelativistic and relativistic results agree with dftatom to 1e-8 Ha accuracy, and with the NIST benchmarks to the stated accuracy of those benchmarks (2e-6 Ha in eigenvalues and 1e-6 Ha in total energies).
http://physics.nist.gov/PhysRefData/DFTdata/Tables/ptable.html
The accuracy is on par with dftatom and uses significantly less computationally expensive routines.
This program is packaged with fpm
, the Fortran package manager.
Where the parameters to conv
are:
We can use an anaconda
helper like micromamba
(installation instructions are here.
We can now set up the tools needed. We support both fpm
and meson
as build systems.
Tests can be run by:
Individual test binaries can also be executed, e.g.:
The API documentation is generated by doxygen with the doxyYoda theme.
To build a local copy and serve it consider:
This program is MIT licensed, see the LICENSE file for details.