If :math:`{N}` is the number of LAPW basis functions, one would naively expect the number of product functions to be roughly :math:`{N^2}`. In the case of the interstitial plane waves, this is not so, since, with a cutoff :math:`{g_\mathrm{max}}`, the maximum momentum of the product would be :math:`{2g_\mathrm{max}}`, leading to :math:`{8N}` as the number of product plane waves. Fortunately, it turns out that the basis size can be much smaller in practice. Therefore, we introduce a reciprocal cutoff radius :math:`{G_\mathrm{max}}` for the interstitial plane waves and find that, instead of :math:`{G_\mathrm{max}=2g_\mathrm{max}}`, good convergence is achieved already with :math:`{G_\mathrm{max}=0.75g_\mathrm{max}}`, the default value. The parameter :math:`{G_\mathrm{max}}` can be set to a different value with the keyword ``GCUT`` in the section ``MBASIS`` of the input file.
(*) The set of MT products selected by ``SELECT`` can still be highly linearly dependent. Therefore, in a subsequent optimization step one diagonalizes the MT overlap matrix and retains only those eigenfunctions whose eigenvalues exceed a predefined tolerance value. This tolerance is 0.0001 by default and can be changed with the keyword ``TOL`` in the input file.
The mixed product basis can still be quite large. In the calculation of the screened interaction, each matrix element, when represented in the basis of Coulomb eigenfunctions, is multiplied by :math:`{\sqrt{v_\mu v_\nu}}` with the Coulomb eigenvalues :math:`{\{v_\mu\}}`. This gives an opportunity for reducing the basis-set size further by introducing a Coulomb cutoff :math:`{v_\mathrm{min}}`. The reduced basis set is then used for the polarization function, the dielectric function, and the screened interaction. The parameter :math:`{v_\mathrm{min}}` can be specified after the keyword ``OPTIMIZE MB`` in three ways: first, as a "pseudo" reciprocal cutoff radius :math:`{\sqrt{4\pi/v_\mathrm{min}}}` (which derives from the plane-wave Coulomb eigenvalues :math:`{v_\mathbf{G}=4\pi/G^2}), second, directly as the parameter :math:`{v_\mathrm{min}}` by using a negative real number, and, finally, as the number of basis functions that should be retained when given as an integer. The so-defined basis functions are mathematically close to plane waves. For testing purposes, one can also enforce the usage of plane waves (or rather projections onto plane waves) with the keyword ``OPTIMIZE PW``, in which case the Coulomb matrix is known analytically. No optimization of the basis is applied, if ``OPTIMIZE`` is omitted.
The mixed product basis can still be quite large. In the calculation of the screened interaction, each matrix element, when represented in the basis of Coulomb eigenfunctions, is multiplied by :math:`{\sqrt{v_\mu v_\nu}}` with the Coulomb eigenvalues :math:`{\{v_\mu\}}`. This gives an opportunity for reducing the basis-set size further by introducing a Coulomb cutoff :math:`{v_\mathrm{min}}`. The reduced basis set is then used for the polarization function, the dielectric function, and the screened interaction. The parameter :math:`{v_\mathrm{min}}` can be specified after the keyword ``OPTIMIZE MB`` in three ways: first, as a "pseudo" reciprocal cutoff radius :math:`{\sqrt{4\pi/v_\mathrm{min}}}` (which derives from the plane-wave Coulomb eigenvalues :math:`{v_\mathbf{G}=4\pi/G^2}`), second, directly as the parameter :math:`{v_\mathrm{min}}` by using a negative real number, and, finally, as the number of basis functions that should be retained when given as an integer. The so-defined basis functions are mathematically close to plane waves. For testing purposes, one can also enforce the usage of plane waves (or rather projections onto plane waves) with the keyword ``OPTIMIZE PW``, in which case the Coulomb matrix is known analytically. No optimization of the basis is applied, if ``OPTIMIZE`` is omitted.
In summary, there are a number of parameters that influence the accuracy of the basis set. Whenever a new physical system is investigated, it is recommendable to converge the basis set for that system. The parameters to consider in this respect are ``GCUT``, ``LCUT``, ``SELECT``, and ``OPTIMIZE``.
...
...
@@ -124,7 +131,7 @@ FFT (WFPROD)
--------------
When the interaction potential is represented in the mixed product basis, the coupling to the single-particle states involve projections of the form
The calculation of these projections can be quite expensive. Therefore, there are a number of keywords that can be used for acceleration. Most of them are, by now, somewhat obsolete. An important keyword, though, is ``FFT`` in the section ``WFPROD`` of the input file. When used, the interstitial terms are evaluated using Fast Fourier Transformations (FFTs), i.e., by transforming into real space (where the convolutions turn into products), instead of by explicit convolutions in reciprocal space. For small systems the latter is faster, but for large systems it is recommendable to use FFTs because of a better scaling with system size. A run with FFTs can be made to yield results identical to the explicit summation. This requires an FFT reciprocal cutoff radius of :math:`{2G_\mathrm{max}+g_\mathrm{max}}`, which can be achieved by setting ``FFT EXACT``, but such a calculation is quite costly. It is, therefore, advisable to use smaller cutoff radii, thereby sacrificing a bit of accuracy but speeding up the computations a lot. If given without an argument, Spex will use 2/3 of the above ''exact'' cutoff. One can also specify a cutoff by a real-valued argument explicitly, good compromises between accuracy and speed are values between 6 and 8 Bohr'^-1^'.
The calculation of these projections can be quite expensive. Therefore, there are a number of keywords that can be used for acceleration. Most of them are, by now, somewhat obsolete. An important keyword, though, is ``FFT`` in the section ``WFPROD`` of the input file. When used, the interstitial terms are evaluated using Fast Fourier Transformations (FFTs), i.e., by transforming into real space (where the convolutions turn into products), instead of by explicit convolutions in reciprocal space. For small systems the latter is faster, but for large systems it is recommendable to use FFTs because of a better scaling with system size. A run with FFTs can be made to yield results identical to the explicit summation. This requires an FFT reciprocal cutoff radius of :math:`{2G_\mathrm{max}+g_\mathrm{max}}`, which can be achieved by setting ``FFT EXACT``, but such a calculation is quite costly. It is, therefore, advisable to use smaller cutoff radii, thereby sacrificing a bit of accuracy but speeding up the computations a lot. If given without an argument, Spex will use 2/3 of the above *exact* cutoff. One can also specify a cutoff by a real-valued argument explicitly, good compromises between accuracy and speed are values between 6 and 8 Bohr'^-1^'.
@@ -41,13 +41,14 @@ Each Spex run needs a job definition, which defines what Spex should do, e.g., `
Details of these jobs are explained in subsequent sections. The job definition must not be omitted but may be empty: ``JOB``, in which case Spex will just read the wavefunctions and energies, perform some checks and some elemental calculations (e.g., Wannier interpolation), and stop. In principle, Spex supports multiple jobs such as ``JOB GW 1:(1-5) DIELEC 1:{0:1,0.01}``. This feature is, however, seldom used and is not guaranteed to work correctly in all versions.
(*) The LAPW method relies on a partitioning of space into MT spheres and the interstitial region. The basis functions are defined differently in the two regions, interstitial plane waves in the latter and numerical functions in the spheres with radial parts :math:`{u(r)}, {\dot{u}(r)=\partial u(r)/\partial\epsilon}`, :math:`{u^\mathrm{LO}(r)}` and spherical harmonics :math:`{Y_{lm}(\hat{\mathbf{r}})}` The plane waves and the angular part of the MT functions can be converged straightforwardly with the reciprocal cutoff radius :math:`{g_\mathrm{max}}` and the maximal l quantum number :math:`{l_\mathrm{max}}`, respectively, whereas the radial part of the MT functions is not converged as easily. The standard LAPW basis is restricted to the functions :math:`{u}` and :math:`{\dot{u}}`. Local orbitals :math:`{u^\mathrm{LO}}` can be used to extend the basis set, to enable the description of semicore and high-lying conduction states. The accuracy of the radial MT basis can be analyzed with the keyword MTACCUR e1 e2 which gives the MT representation error [Phys. Rev. B 83, 081101] in the energy range between e1 and e2. (If unspecified, e1 and e2 are chosen automatically.) The results are written to the output files spex.mt.t where t is the atom type index, or spex.mt.s.t with the spin index s(=1 or 2) for spin-polarized calculations. The files contain sets of data for all l quantum numbers, which can be plotted separately with gnuplot (e.g., plot "spex.mt.1" i 3 for :math:`{l=3}`
(*) The LAPW method relies on a partitioning of space into MT spheres and the interstitial region. The basis functions are defined differently in the two regions, interstitial plane waves in the latter and numerical functions in the spheres with radial parts :math:`{u(r)}, {\dot{u}(r)=\partial u(r)/\partial\epsilon}`, :math:`{u^\mathrm{LO}(r)}` and spherical harmonics :math:`{Y_{lm}(\hat{\mathbf{r}})}` The plane waves and the angular part of the MT functions can be converged straightforwardly with the reciprocal cutoff radius :math:`{g_\mathrm{max}}` and the maximal l quantum number :math:`{l_\mathrm{max}}`, respectively, whereas the radial part of the MT functions is not converged as easily. The standard LAPW basis is restricted to the functions :math:`{u}` and :math:`{\dot{u}}`. Local orbitals :math:`{u^\mathrm{LO}}` can be used to extend the basis set, to enable the description of semicore and high-lying conduction states. The accuracy of the radial MT basis can be analyzed with the keyword MTACCUR e1 e2 which gives the MT representation error [Phys. Rev. B 83, 081101] in the energy range between e1 and e2. (If unspecified, e1 and e2 are chosen automatically.) The results are written to the output files spex.mt.t where t is the atom type index, or spex.mt.s.t with the spin index s(=1 or 2) for spin-polarized calculations. The files contain sets of data for all l quantum numbers, which can be plotted separately with gnuplot (e.g., plot "spex.mt.1" i 3 for :math:`{l=3}`)
| Examples | ``MTACCUR -1 2`` | Calculate MT representation error between -1 and 2 Hartree |
...
...
@@ -116,19 +118,19 @@ BANDINFO
--------
(*) In some cases, it may be necessary to replace the energy eigenvalues, provided by the mean-field (DFT) code, by energies (e.g., GW quasiparticle energies) obtained in a previous Spex calculation, for example, to determine the GW Fermi energy or to perform energy-only self-consistent calculations. This can be achieved with the keyword ``ENERGY file``, where ``file`` contains the new energies in eV. The format of ``file`` corresponds to the output of the ``spex.extr`` utility: ``spex.extr g spex.out > file``. It must be made sure that ``file`` contains energy values for the whole irreducible Brillouin zone. Band energies not contained in ``file`` will be adjusted so that the energies are in accending order (provided that there is at least one energy value for the particular k point).
(*) This keyword modifies the exchange splitting of a collinear magnetic system, i.e., it shifts spin-up and spin-down energies relative to to each other so as to increase or decrease the exchange splitting. With ``DELTAEX x``, the spin-up (spin-down) energies are lowered (elevated) by x/2. The parameter x can be used to enforce the Goldstone condition in spin-wave calculations [Phys. Rev. B 94, 064433 (2016)]
(*) If specified, Spex only reads the LAPW basis set from the input data, provided by the mean-field (DFT) code, but performs the diagonalization of the Hamiltonian at the k points itself. This calculation effectively replaces the second run of the DFT code. In this sense, the name of the keyword is a bit misleading, as the calculation is non-iterative. The keyword ``ITERATE`` is mostly intended for testing and debugging. It is not available for executables compiled with ``-DLOAD`` (configured with ``--enable-load``).
@@ -13,7 +13,7 @@ It needs input from a converged DFT calculation, which can be generated by Fleur
If you use SPEX for your research, please cite the following work:
.. highlights:: Christoph Friedrich, Stefan Blügel, Arno Schindlmayr, "Efficient implementation of the GW approximation within the all-electron FLAPW method", Phys. Rev. B 81, 125102 (2010).
.. highlights:: Christoph Friedrich, Stefan Blügel, Arno Schindlmayr, "Efficient implementation of the GW approximation within the all-electron FLAPW method", *Phys. Rev. B 81, 125102 (2010)*.
