This document lists and provides the description of the name (keywords) of the input variables for runs based on the Projector Augmented Waves methodology, to be used in the main input file of the abinit code.
The new user is advised to read first the new user's guide, before reading the present file. It will be easier to discover the present file with the help of the tutorial.
When the user is sufficiently familiarized with ABINIT, the reading of the ~abinit/doc/users/tuning file might be useful. For response-function calculations using abinit, please read the response function help file
Relevant only when
usepaw=1.
The box cut-off ratio is the ratio between the wavefunction plane wave
sphere radius, and the radius of the sphere that can be inserted in the FFT box,
in reciprocal space.
If the density was generated only from wavefunctions,
this ratio should at least two in order for the density to be exact. If one
uses a smaller ratio, one will gain speed, at the expense of accuracy. In case
of pure ground state calculation (e.g. for the determination of geometries),
this is sensible. However, the wavefunctions that are obtained CANNOT be used
for starting response function calculation.
However, some augmentation charge is always added in PAW, and even with the box cut-off
ratio larger than two, the density is never exact. Sometimes, this ratio must be
much larger than two for the computation to be converged at the
required level of accuracy.
Go to the top
| Complete list of input variables
Relevant only when
usepaw=1,
usepawu=1,
usedmatpu/=0 for Ground-State calculations.
Gives the value of an initial density matrix used in LDA+U and kept
fixed during the first abs(usedmatpu) SCF iterations.
Only components corresponding to lpawu angular momentum are requested.
Restriction: In order to use dmatpawu, lpawu must be identical for all atom types (or -1).
The occupation matrix is in the basis of real spherical harmonics Slm (note that this differs from the choice made when
prtdosm=1, that is in the basis of complex spherical harmonics).
Their are ordered by increasing m, and are defined e.g. in
the article
"Evaluation of the rotation matrices in the basis of real spherical harmonics",
by Miguel A. Blancoa, M. Floreza, M. Bermejo,
Journal of Molecular Structure (Theochem) 419, 19 (1997), that can be downloaded from
the author Web site.
For the case l=2 (d states), the five columns corresponds respectively to (the normalisation factor has been dropped)
Relevant only when
usepaw=1 and
usepawu=1
This option governs the way occupations of localized atomic levels are computed:
Relevant only when
usepaw=1 and
usepawu=1 and for Ground-State calculations.
This option can be used to diagonalize the occupation matrix Nocc_{m,m'}.
Relevant values are:
Relevant only when
usepaw=1 and
usepawu=1 or
usedmft=1 .
This gives the ratio of Slater Integrals F4 and F2.
It is used in DFT+U or DFT+DMFT for the calculation of the orbital
dependent screened coulomb interaction.
Go to the top
| Complete list of input variables
Relevant only when
usepaw=1 and
(usepawu=1 or
usedmft=1) and
lpawu=3) .
This gives the ratio of Slater Integrals F6 and F2.
It is used with f4of2_sla=3) .
in DFT+U or DFT+DMFT for the calculation of the orbital
dependent screened coulomb interaction.
Go to the top
| Complete list of input variables
Concern all summations in the reciprocal space and is allowed in PAW and norm-conserving.
Relevant only when
usepaw=1, and usepawu=1.
Gives the value of the screened exchange interaction between correlated
electrons corresponding to lpawu
for each species.
In the case where lpawu =-1,
the value is not used.
Go to the top
| Complete list of input variables
Activated if
useexexch is equal to 1.
Give for each species the value of the angular momentum (only values 2 or 3 are allowed)
on which to apply the exact exchange correction.
Go to the top
| Complete list of input variables
Activated if usepawu
is equal to 1 or 2.
Give for each species the value of the angular momentum (only
values 2 or 3 are allowed) on which to apply the LDA+U correction.
Maximum number of wavevectors used to sample the local part of the potential, in PAW.
Actually referred to as mqgrid_vl internally. Should change name to the latter ...
See also mqgrid
Go to the top
| Complete list of input variables
Relevant only when
usepaw=1.
This variable has the same meaning as ngfft (gives the size of fast Fourier
transform (fft) grid in three dimensions) but concerns the "double
grid" only used for PAW calculations.
Go to the top
| Complete list of input variables
Relevant only when
usepaw=1.
When pawcpxocc=2, PAW augmentation occupancies are treated as COMPLEX; else they are considered are REAL.
This is needed when time-reversal symmetry is broken (typically when spin-orbit coupling is activated).
Note for ground-state calculations (optdriver=0):
The imaginary part of PAW augmentation occupancies is only used for the computation of the total energy by "direct scheme";
this only necessary when SCF mixing on potential is chosen (iscf<10).
When SCF mixing on density is chosen (iscf>=10), the "direct" decomposition
of energy is only printed out without being used. It is thus possible to use pawcpxocc=1 is the latter case.
In order to save CPU time, when molecular dynamics is selected (ionmov>=6) and
SCF mixing done on density (iscf>=10), pawcpxocc=2 is (by default) set to 1.
When pawcpxocc=1, "direct" decomposition of total energy cannot be printed out.
Go to the top
| Complete list of input variables
Only relevant if optdriver=3 or 4, that is, screening or sigma calculations, and usepaw=1.
When pawcross=1, the overlap between the plane-wave part of one band and the on-site part of an other
is taken into account in the computation of the oscillator strengths. Hence, the completeness of the on-site basis is no longer assumed.
Go to the top
| Complete list of input variables
Needed only when
usepaw=1.
Define the energy cut-off for the fine FFT grid
(the "double grid", that allows to transfer data from the normal, coarse, FFT grid to the
spherical grid around each atom).
pawecutdg must be larger or equal to
ecut. If it is equal to it, then no fine
grid is used. The results are not very accurate, but the computations
proceed quite fast.
For typical PAW computations, where ecut
is on the order of 15 Ha, pawecutdg should be on the order of 40 Ha.
Choosing a larger value should not increase the accuracy, but does not slow down the
computation either, only the memory. The choice made for this variable DOES have a bearing
on the numerical accuracy of the results, and, as such, should be the object
of a convergence study. The convergence test might be made on the total energy
or derived quantities, like forces, but also on the two values of the
"Compensation charge inside spheres", a quantity written in the log file.
Go to the top
| Complete list of input variables
Relevant only when
usepaw=1
for Ground-State calculations and non self-consistent calculations.
This option can be used to plot band structure. For each atom (specified by
natsph and iatsph), each angular momentum, and each
spin polarisation, the band structure is written in files (such as e.g. FATBANDS_at0001_Ni_is2_l2_m-1). Each file
contains the eigenvalue, and the contribution of angular momentum L, and projection of angular momentum m,
(for the corresponding wavefunction) to the PAW density inside the PAW sphere
as a function of the index of the k-point.
The output can be readily plotted with the software xmgrace (e.g xmgrace FATBANDS_at0001_Ni_is2_l2_m-1).
Relevant values are:
Relevant only when
usepaw=1.
The expansion of the densities in angular momenta is
performed up to l=pawlcutd.
Note that, for a given system, the maximum value of pawlcutd is 2*l_max,
where l_max is the maximum l of the PAW partial waves basis.
The choice made for this variable DOES have a bearing
on the numerical accuracy of the results, and, as such, should be the object
of a convergence study. The convergence test might be made on the total energy
or derived quantities, like forces, but also on the two values of the
"Compensation charge inside spheres", a quantity written in the log file.
Go to the top
| Complete list of input variables
Relevant only when
usepaw=1.
The choice made for this variable determine how the spherical part of the
density is mixed during electronic iterations.
Only parts of (rhoij quantities) associated with
l angular momenta up to l=pawlmix are mixed.
Other parts of augmentation occupancies are not included in the mixing process.
This option is useful to save CPU time but DOES have a bearing on the numerical accuracy of the results.
Go to the top
| Complete list of input variables
Relevant only when
usepaw=1.
The choice made for this variable determines the grid on which the density (or potential) is mixed during the SCF cycle.
- If pawmixdg=1 the density/potential is mixed in REAL space using the fine FFT grid
(defined by pawecutdg or ngfftdg).
- If pawmixdg=0 the density/potential is mixed in RECIPROCAL space using the coarse FFT grid
(defined by ecut or ngfft).
Only components of the coarse grid are mixed using the scheme defined by iscf;
other components are only precondionned by diemix and simply mixed.
This option is useful to save memory and does not affect numerical accuracy of converged results.
If pawmixdg=1, density and corresponding residual are stored for previous iterations and are
REAL arrays of size nfftdg.
If pawmixdg=0, density and corresponding residual are stored
for previous iterations and are COMPLEX arrays of size nfft.
The memory saving is particulary efficient when using the Pulay mixing
(iscf=7 or 17).
Go to the top
| Complete list of input variables
Needed only when
usepaw=1.
Relevant only when a GGA exchange-correlation functional is used.
When this flag is activated, the gradients of compensation charge density (n_hat) are exactly
computed (i.e. analytically); when it is desactivated, they are computed with a numerical scheme in
reciprocal space (which can produce inaccurate results if the compensation charge density
is highly localized).
As analytical treatment of compensation charge density gradients is CPU time demanding,
it is possible to bypass it with pawnhatxc=0; but the numerical accuracy can be affected
by this choice. It is recommended to test the validity of this approximation before use.
Go to the top
| Complete list of input variables
Needed only when
usepaw=1.
Number of phi angles (longitude) used to discretize the
data on the atomic spheres. This discretization is completely
defined by pawnphi
and pawntheta.
Go to the top
| Complete list of input variables
Relevant only when
usepaw=1.
Number of theta angles (latitude) used to discretize the
data on the atomic spheres. This discretization is completely
defined by pawntheta
and pawnphi.
Go to the top
| Complete list of input variables
Relevant only when
usepaw=1.
Concerns the computation of the contributions to the density from the spheres (named rho_1 -
rho_tild_1).
If set to 0, all lm-moments of the sphere contributions to the density are computed at each
electronic iteration.
If set to 1, only non-zero lm-moments of the sphere contributions to the density are computed
at each electronic iteration (they are all computed at the first iteration then only
those found to be non-zero will be computed ; thus the first iteration is more cpu intensive)
Go to the top
| Complete list of input variables
When PAW is activated, the self-consistent requires the mixing of both the total potential (or density)
and the "spherical part" (in fact the augmentation occupancies rho_ij).
The same mixing scheme is applied to the potential (density) and the spherical part.
It is optimized in order to minimize a residual.
If pawoptmix=0 the residual is the potential (or density) residual.
If pawoptmix=1 the residual is a sum of the potential (or density) residual
and the "spherical part" residual.
Go to the top
| Complete list of input variables
When PAW is activated, a localized atomic basis is added to describe wave functions.
Spheres around atoms are defined and they are IN PRINCIPLE not allowed to overlap. However,
a small overlap can be allowed without compromising the accurary of results. Be aware that
too high overlaps can lead to unphysical results.
With the pawovlp variable, the user can control the (voluminal) overlap percentage
allowed without stopping the execution.
pawovlp is the value (in percentage: 0...100%) obtained by dividing
the volume of the overlap of two spheres by the volume of the smallest sphere.
The following values are permitted for pawovlp:
Go to the top
| Complete list of input variables
This input variable controls the computation and/or printing of contributions to the PAW partial DOS in _DOS file(s):
Control print volume and debugging output for PAW in log file or standard output.
If set to 0, the print volume is at its minimum.
pawprtvol can have values from -3 to 3:
- pawprtvol=-1 or 1: matrices rho_ij (atomic occupancies) and D_ij (psp strength)
are printed at each SCF cycle with details about their contributions.
- pawprtvol=-2 or 2: like -1 or 1 plus additional printing: moments of "on-site" densities,
details about local exact exchange.
- pawprtvol=-3 or 3: like -2 or 2 plus additional printing: details about
PAW+U, rotation matrices of sphercal harmonics.
When pawprtvol>=0, up to 12 components of rho_ij and D_ij matrices for the 1st and last atom are printed.
When pawprtvol<0, all components of rho_ij and D_ij matrices for all atoms are printed.
Go to the top
| Complete list of input variables
This input variable controls the output of the full PAW wave functions including the on-site
contribution inside each PAW sphere needed to reconstruct the correct nodal shape in the augmentation region.
pawprtwf=1 causes the generation of a file _AE_WFK that contains the full
wavefunctions in real space on the fine FFT grid defined by
pawecutdg or ngfftdg).
Limitations: At present (v6.0), pawprtwf=1 is not compatible neither with the k-point parallelism
nor with the parallelism over G-vectors. Therefore the output of the _AE_WFK has to be done in sequential.
Moreover, in order to use this feature, one has to be enable the support for ETSF-IO at configure-time
as the _AW_WFK file is written using the NETCDF file format following the ETSF-IO specification for
wavefunctions in real space. If the code is run in entirely in serial, additional output is made of various
contributions to the all-electron avefunction. By default the full available set of bands and k-points are
ouput, but a single band and k-point index can be requested by using the variables
pawprt_b and pawprt_k.
Go to the top
| Complete list of input variables
When PAW is activated, the spin-orbit coupling can be added without the use
of specific PAW datasets (pseudopotentials).
If pawspnorb=1, spin-orbit will be added.
Note that only the all-electron "on-site" contribution to the Hamiltonian is taken into account;
this a very good approximation but requires the
following conditions to be fullfilled:
When PAW is activated, the computation of compensation charge density (so called
"hat" density) requires the computation of g_l(r).Y_lm(r) factors (and cartesian derivatives) at each point
of real space contained in PAW spheres. The number of atoms, of (l,m) quantum numbers
and the sharpness of the real FFT grid can lead to a very big {g_l.Y_lm} datastructure.
One can save memory by putting pawstgylm=0; but, in that case, g_l(r).Y_lm(r) factors
a re-computed each time they are needed and CPU time increases.
Relevant only when
usepaw=1 and optdriver=0 (Ground-State calculation).
When a sophisticated preconditioning scheme is selected for the SCF cycle of a Ground-State calculation
(iprcel>0), the computation of the susceptibility matrix
is required several times during the cycle. This computation is computer time consuming, especially
-- within PAW -- because of the inclusion of additional terms due to the compensation charge density.
As only a crude valuation of the susceptibilty matrix is needed (to evaluate a preconditioning matrix),
the compensation charge contribution can be neglected to save CPU time (select pawsushat=0).
This approximation could be unfavourable in some cases; in the latter, we advice to put pawsushat=1.
Relevant only when
usepaw=1.
When PAW is activated, the computation of cprj arrays is memory and time consuming.
When pawusecp=0, then the cprj are never kept in memory, they are recomputed when needed (this is CPU-time consuming).
When pawusecp=1, then the cprj are computed once and then kept in memory.
Change the value of the keyword only if you are an experienced user (developper).
Remember: cprj = (WF_n .dot. p_i) (WF_n=wave function, p_i=non-local projector).
For the time being, only activated for RF calculations.
Go to the top
| Complete list of input variables
Relevant only when usepaw=1.
Relevant only when usepaw=1.
Relevant only when usepaw=1.
Relevant only when usepaw=1.
Relevant only when usepaw=1.
Relevant only when
usepaw=1, and prtefg = 3 or greater.
Relevant only when
usepaw=1, and prtefg = 1 or greater.
Relevant only when
usepaw=1, and pawspnorb = 1 (or greater).
Scaling of the spin-orbit interaction. The default values gives the first-principles value, while
other values are used for the analysis of the effect of the spin-orbit interaction,
but are not expected to correspond to any physical situation.
Go to the top
| Complete list of input variables
Relevant only when
usepaw=1, and usepawu=1.
Gives the value of the
screened coulomb interaction between correlated electrons corresponding
to lpawu for each species.
In the case where
lpawu
=-1, the value is not used.
In the case of a GW calculation, the U interaction defined by upawu will be REMOVED from the self energy.
In particular, for G0 W0 calculations (perturbative calculations), the energy eigenvalues obtained
after an underlying DFT+U calculation will be
E_GW = E_DFT+U + < phi | Self-energy - U | phi>
Actually, in order to perform a GW @ DFT+U calculation, one should define the same value of U in the self-energy calculation,
than the one defined in the DFT calculation.
The easiest is actually to define the value of U for the whole set of calculations (for the different datasets),
including the screening, even if the U value does not play explicitly a role in the computation of the latter (well, the input
wavefunctions will be different anyhow).
It is possible to perform calculations of the type GW+U' @ DFT+U , so keeping a U interaction (usually smaller than the initial U)
in the GW calculation, by defining a smaller U than the one used in the DFT calculation. This value will be subtracted in the GW correction calculation,
as outlined above.
Explicitly, in order to do a calculation of a material with a DFT U value of 7.5 eV, followed by a GW calculation where there is a residual
U value of 2 eV, one has to define :
uldau1 7.5 eV ! This is for the DFT calculation ... optdriver4 4 uldau4 5.5 eV ! This is for the screening calculation
Relevant only when
usepaw=1, and usepawu=1.
When usedmatpu/=0, an initial density matrix (given by dmatpawu
keyword) is used and kept fixed during the first ABS(usedmatpu) SCF steps.
This starting value of the density matrix can be useful to find the correct ground state.
Within LDA+U formalism, finding the minimal energy of the system is tricky; thus it is advised to test several values
of the initial density matrix.
Note also that the density matrix has to respect some symmetry rules determined by the space group.
If the symmetry is not respected in the input, the matrix is however automatically symmetrised.
The sign of usedmatpu has influence only when ionmov/=0 (dynamics or relaxation):
- When usedmatpu>0, the density matrix is kept constant only at first ionic step
- When usedmatpu<0, the density matrix is kept constant at each ionic step
Go to the top
| Complete list of input variables
When useexexch=1, the hybrid functional PBE0 is used
in PAW, inside PAW spheres only, and only for correlated orbitals given by
lexexch. To change the ratio of exact exchange, see also exchmix.
Go to the top
| Complete list of input variables
Must be non-zero if a DFT+U calculation is done, or if a GW calculation following a DFT+U calculation is done (important !).
Relevant only when usepaw=1.
This flag determines how the exchange-correlation terms are computed for the pseudo-density.
When usexcnhat=0, exchange-correlation potential does not include the compensation charge density, i.e.
Vxc=Vxc(tild_Ncore + tild_Nvalence).
When usexcnhat=1, exchange-correlation potential includes the compensation charge density, i.e.
Vxc=Vxc(tild_Ncore + tild_Nvalence + hat_N).
When usexcnhat=-1,the value of usexcnhat is determined from the reading of the PAW dataset file
(pseudopotential file). When PAW datasets with different treatment of Vxc are used in the same
run, the code stops.
Go to the top
| Complete list of input variables