This document lists and provides the description of the name (keywords) of the "basic" input variables 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/Infos/tuning file might be useful. For response-function calculations using abinit, the complementary file ~ABINIT/Infos/respfn_help is needed.
Gives the length scales by which dimensionless primitive translations (in rprim) are to be multiplied. By default, given in Bohr atomic units (1 Bohr=0.5291772108 Angstroms), although Angstrom can be specified, if preferred, since acell has the 'LENGTH' characteristics. See further description of acell related to the rprim input variable, the scalecart input variable, and the associated internal rprimd input variable.
Note that acell is NOT the length of the conventional orthogonal basis vectors, but the scaling factors of the primitive vectors. Use scalecart to scale the cartesian coordinates.
Variable under development
Gives the angles between directions of primitive vectors of the unit cell (in degrees), as an alternative to the input array rprim . Will be used to set up rprim, that, together with the array acell, will be used to define the primitive vectors.
R1=( a , 0,c) R2=(-a/2, sqrt(3)/2*a,c) R3=(-a/2,-sqrt(3)/2*a,c)where a2+c2=1.0d0
Used for kinetic energy cutoff
which controls number
of planewaves at given k point by:
(1/2)[(2 Pi)*(k+Gmax)]2=ecut for Gmax.
All planewaves inside this "basis sphere" centered
at k are included in the basis (except if dilatmx
is defined).
Can be specified in Ha (the default), Ry, eV or Kelvin, since
ecut has the
'ENERGY' characteristics.
(1 Ha=27.2113845 eV)
This is the single parameter which can have an enormous
effect on the quality of a calculation; basically the larger
ecut is, the better converged the calculation is. For fixed
geometry, the total energy MUST always decrease as ecut is
raised because of the variational nature of the problem.
Usually one runs at least several calculations at various ecut to investigate the convergence needed for reliable results.
For k-points whose coordinates are build from 0 or 1/2, the implementation of time-reversal symmetry that links coefficients of the wavefunctions in reciprocal space has been realized. See the input variable istwfk. If activated (which corresponds to the Default mode), this input variable istwfk will allow to divide the number of plane wave (npw) treated explicitly by a factor of two. Still, the final result should be identical with the 'full' set of plane waves.
See the input variable ecutsm, for the smoothing of the kinetic energy, needed to optimize unit cell parameters.
Controls the self-consistency.
Positive values =>
this is the usual choice for doing the usual ground state (GS)
calculations or for structural relaxation, where
the potential has to be determined self-consistently.
The choice between different algorithms for SCF is possible :
Such algorithms for treating the "SCF iteration history" should be coupled with accompanying algorithms for the SCF "preconditioning". See the input variable iprcel. The default value iprcel=0 is often a good choice, but for inhomogeneous systems, you might gain a lot with iprcel=45.
(Warning : if iscf>10, at present (v4.6), the energy printed at each SCF cycle is not variational - this should not affect the other properties, and at convergence, all values are OK)
- In the norm-conserving case,
the default option is iscf=7, which is a compromise between speed and reliability.
The value iscf= 2 is safer but slower.
- In the PAW case, default option is iscf=17.
In PAW you have the possibility to mix density/potential on the fine or coarse FFT grid (see pawmixdg).
- Note that a Pulay mixing (iscf=7 or 17) with npulayit
=1 (resp. 2) is equivalent to an Anderson mixing with iscf=3 or 13 (resp. 4 or 14).
- Also note that:
* when mixing is done on potential (iscf<10), total energy is computed by "direct" decomposition.
* when mixing is done on density (iscf>=10), total energy is computed by "double counting" decomposition.
"Direct" and "double counting" decomposition of energy are equal when SCF cycle is converged. Note that,
when using GGA XC functionals, these decompositions of energy can be slighty different due
to imprecise computation of density gradients on FFT grid (difference decreases as size of FFT grid increases -
see ecut for NC pseudopotentials, pawecutdg for PAW).
Other (negative) options:
Controls the choice of exchange and correlation (xc). The list of XC functionals is given
below. Positive values are for ABINIT native library of XC functionals, while negative values are for calling
the much wider set of functionals from the ETSF LibXC library (by M. Marques), also available at the
ETSF library Web page
Note that the choice made here should agree with the choice
made in generating the original pseudopotential, except
for ixc=0 (usually only used for debugging).
A warning is issued if this is not the case.
However, the choices ixc=1, 2, 3 and 7 are fits to the same data, from
Ceperley-Alder, and are rather similar, at least for spin-unpolarized systems.
The choice between the non-spin-polarized and spin-polarized case
is governed by the value of nsppol (see below).
Native ABINIT XC functionals
NOTE : in the implementation of the spin-dependence of these
functionals, and in order to avoid divergences in their
derivatives, the interpolating function between spin-unpolarized
and fully-spin-polarized function has been slightly modified,
by including a zeta rescaled by 1.d0-1.d-6. This should affect
total energy at the level of 1.d-6Ha, and should
have an even smaller effect on differences of energies, or derivatives.
The value ixc=10 is used internally : gives the difference between ixc=7 and
ixc=9, for use with an accurate RPA correlation energy.
ETSF Lib XC functionals
Note that you must compile ABINIT with the LibXC plug-in in order to be able to access these functionals.
The LibXC functionals are accessed by negative values of ixc.
The LibXC contains functional forms for either exchange-only functionals, correlation-only functionals,
or combined exchange and correlation functionals. Each of them is to be specified by a three-digit number.
In case of a combined exchange and correlation functional, only one such three-digit number has to be specified as value of ixc,
with a minus sign (to indicate that it comes from the LibXC).
In the case of separate exchange functional (let us represent its identifier by XXX) and
correlation functional (let us represent its identified by CCC),
a six-digit number will have to be specified for ixc, by concatenation, be it XXXCCC or CCCXXX.
As an example, ixc=-020 gives access to the Teter93 LDA, while
ixc=-101130 gives access to the PBE GGA.
In version 0.9 of LibXC (December 2008), there are 16 three-dimensional (S)LDA functionals (1 for X, 14 for C, 1 for combined XC),
and there are 41 three-dimensional GGA (23 for X, 8 for C, 10 for combined XC).
Note that for a meta-GGA, the kinetic energy density is needed. This means having usekden=1 .
(S)LDA functionals (do not forget to add a minus sign, as discussed above)
GGA functionals (do not forget to add a minus sign, as discussed above)
MetaGGA functionals (do not forget to add a minus sign, as discussed above)
Gives the dataset index of each of the datasets. This index will be used :
Contains the k points in terms
of reciprocal space primitive translations (NOT in
cartesian coordinates!).
Needed ONLY
if kptopt=0, otherwise
deduced from other input variables.
It contains dimensionless numbers in terms of which
the cartesian coordinates would be:
where
Note: one of the algorithms used to set up the sphere
of G vectors for the basis needs components of k-points
in the range [-1,1], so the
remapping is easily done by adding or subtracting 1 from
each component until it is in the range [-1,1]. That is,
given the k point normalization kptnrm described below,
each component must lie in [-kptnrm,kptnrm].
Note: a global shift can be provided by qptn
Not read if kptopt/=0 .
Establishes a normalizing denominator
for each k point.
Needed only
if kptopt<=0, otherwise
deduced from other input variables.
The k point coordinates as fractions
of reciprocal lattice translations are therefore
kpt(mu,ikpt)/kptnrm. kptnrm defaults to 1 and can
be ignored by the user. It is introduced to avoid
the need for many digits in representing numbers such as 1/3.
It cannot be smaller than 1.0d0
Controls the set up of the k-points list. The aim will be to initialize, by straight reading or by a preprocessing approach based on other input variables, the following input variables, giving the k points, their number, and their weight: kpt, kptnrm, nkpt, and, for iscf/=-2, wtk.
Often, the k points will form a lattice in reciprocal space. In this case, one will also aim at initializing input variables that give the reciprocal of this k-point lattice, as well as its shift with respect to the origin: ngkpt or kptrlatt, as well as on nshiftk and shiftk.
A global additional shift can be provided by qptn
Gives the total number of atoms in the unit cell.
Default is 1 but you will obviously want to input this
value explicitly.
Note that natom refers to all atoms in the unit cell, not
only to the irreducible set of atoms in the unit cell (using symmetry operations,
this set allows to recover all atoms). If you want
to specify only the irreducible set of atoms, use the
symmetriser, see the input variable natrd.
Gives number of bands, occupied plus
possibly unoccupied, for which wavefunctions are being computed
along with eigenvalues.
Note : if the parameter
occopt (see below) is not set to 2,
nband is a scalar integer, but
if the parameter occopt is set to 2,
then nband must be an array nband(nkpt*
nsppol) giving the
number of bands explicitly for each k point. This
option is provided in order to allow the number of
bands treated to vary from k point to k point.
For the values of occopt not equal to 0 or 2, nband
can be omitted. The number of bands will be set up
thanks to the use of the variable fband. The present Default
will not be used.
If nspinor is 2, nband must be even for each k point.
In the case of a GW calculation (optdriver=3 or 4), nband gives the number of bands to be treated to generate the screening (susceptibility and dielectric matrix), as well as the self-energy. However, to generate the _KSS file (see kssform) the relevant number of bands is given by nbandkss.
Gives the maximum number of occupied bands with which Fock exact exchange is being computed for the wavefunctions.
Gives the number of data sets to be
treated.
If 0, means that the multi-data set treatment is not used,
so that the root filenames will not be appended with _DSx,
where 'x' is the dataset index defined
by the input variable jdtset,
and also that input names with a dataset index are not allowed.
Otherwise, ndtset=0 is equivalent to ndtset=1.
Used when kptopt>=0,
if kptrlatt
has not been defined (kptrlatt
and ngkpt are exclusive of each other).
Its three positive components
give the number of k points of Monkhorst-Pack grids
(defined with respect to primitive axis in reciprocal space)
in each of the three dimensions.
ngkpt will be used to generate the
corresponding kptrlatt
input variable.
The use of nshiftk
and shiftk, allows to generate
shifted grids, or Monkhorst-Pack grids defined
with respect to conventional unit cells.
When nshiftk=1, kptrlatt is initialized as a diagonal (3x3) matrix, whose diagonal elements are the three values ngkpt(1:3). When nshiftk is greater than 1, ABINIT will try to generate kptrlatt on the basis of the primitive vectors of the k-lattice: the number of shifts might be reduced, in which case kptrlatt will not be diagonal anymore.
Monkhorst-Pack grids are usually the most efficient when their defining integer numbers are even. For a measure of the efficiency, see the input variable kptrlen.
If non-zero, nkpt gives the number of k points in the k point array kpt. These points are used either to sample the Brillouin zone, or to build a band structure along specified lines.
If nkpt is zero, the code deduces from other input variables (see the list in the description of kptopt) the number of k points, which is possible only when kptopt/=0. If kptopt/=0 and the input value of nkpt/=0, then ABINIT will check that the number of k points generated from the other input variables is exactly the same than nkpt.
If kptopt is positive,
nkpt must be coherent with the values
of kptrlatt,
nshiftk
and shiftk.
For ground state calculations, one should select the
k point in the irreducible Brillouin Zone (obtained
by taking into account point symmetries and the time-reversal
symmetry).
For response function calculations, one should
select k points in the full Brillouin zone, if the wavevector
of the perturbation does not vanish, or in a half of
the Brillouin Zone if q=0. The code will automatically decrease
the number of k points to the minimal set needed for
each particular perturbation.
If kptopt is negative, nkpt will be the sum of the number of points on the different lines of the band structure. For example, if kptopt=-3, one will have three segments; supposing ndivk is 10 12 17, the total number of k points of the circuit will be 10+12+17+1(for the final point)=40.
nkpthf gives the number of k points used to sample the full Brillouin zone for the Fock exact exchange contribution.
Go to the top
| Complete list of input variables
This parameter gives the number of shifted grids to be used concurrently to generate the full grid of k points. It can be used with primitive grids defined either from ngkpt or kptrlatt. The maximum allowed value of nshiftk is 8. The values of the shifts are given by shiftk.
Give the number of INDEPENDENT spin polarisations. Can take the values 1 or 2.
If nsppol=1, one has an unpolarized calculation (nspinor=1, nspden=1) or an antiferromagnetic system (nspinor=1, nspden=2), or a calculation in which spin up and spin down cannot be disantengled (nspinor=2), that is, either non-collinear magnetism or presence of spin-orbit coupling, for which one needs spinor wavefunctions.
If nsppol=2, one has a spin-polarized (collinear) calculation with separate and different wavefunctions for up and down spin electrons for each band and k point. Compatible only with nspinor=1, nspden=2.
In the present status of development, with nsppol=1, all values of ixc are allowed, while with nsppol=2, some values of ixc might not be allowed (e.g. 2, 3, 4, 5, 6, 20, 21, 22 are not allowed).
See also the input variable nspden for the components of the density matrix with respect to the spin-polarization.
Gives the maximum number of cycles (or "iterations") in a SCF or non-SCF run.
Full convergence from random numbers is usually achieved in
12-20 SCF iterations. Each can take from minutes to hours.
In certain difficult cases, usually related to a small or
zero bandgap or magnetism, convergence performance may be much worse.
When the convergence tolerance tolwfr on the wavefunctions
is satisfied, iterations will stop, so for well converged
calculations you should set nstep to a value larger than
you think will be needed for full convergence, e.g.
if using 20 steps usually converges the system, set nstep to 30.
For non-self-consistent runs (iscf < 0) nstep governs
the number of cycles of convergence for the wavefunctions for a fixed density
and Hamiltonian.
NOTE that a choice of nstep=0 is permitted; this will
either read wavefunctions from disk (with irdwfk=1
or irdwfq=1,
or non-zero getwfk
or getwfq in the case
of multi-dataset) and
compute the density, the total energy and stop, or else
(with all of the above vanishing) will initialize
randomly the wavefunctions and
compute the resulting density and total energy.
This is provided for testing purposes.
Also NOTE that nstep=0
with irdwfk=1 will exactly give the same result as
the previous run only if the latter is done with iscf<10
(potential mixing).
One can output the density by using prtden.
The forces and stress tensor are computed with nstep=0.
Gives number of space group symmetries
to be applied in this problem. Symmetries will be input in
array "symrel" and (nonsymmorphic) translations vectors
will be input
in array "tnons". If there is no symmetry in the problem
then set nsym to 1, because the identity is still a symmetry.
In case of a RF calculation, the code is able to use
the symmetries of the system to decrease the number of
perturbations to be calculated, and to decrease of the
number of special k points to be used for the sampling of
the Brillouin zone.
After the response to the perturbations have been calculated,
the symmetries are used to generate as many as
possible elements of the 2DTE from those already
computed.
SYMMETRY FINDER mode (Default mode).
If nsym is 0, all the atomic coordinates must be
explicitely given (one cannot use the geometry builder
and the symmetrizer): the code will then find automatically
the symmetry operations that leave the lattice and each
atomic sublattice invariant. It also checks whether the
cell is primitive (see chkprim).
Note that the tolerance on symmetric atomic positions and
lattice is rather stringent :
for a symmetry operation to be admitted,
the lattice and atomic positions must map on themselves
within 1.0e-8 .
The user is allowed to set up systems with non-primitive unit cells (i.e. conventional FCC or BCC cells, or supercells without any distortion). In this case, pure translations will be identified as symmetries of the system by the symmetry finder. Then, the combined "pure translation + usual rotation and inversion" symmetry operations can be very numerous. For example, a conventional FCC cell has 192 symmetry operations, instead of the 48 ones of the primitive cell. A maximum limit of 384 symmetry operations is hard-coded. This corresponds to the maximum number of symmetry operations of a 2x2x2 undistorted supercell. Going beyond that number will make the code stop very rapidly. If you want nevertheless, for testing purposes, to treat a larger number of symmetries, change the initialization of "msym" in the abinit.F90 main routine, then recompile the code.
For GW calculation, the user might want to select only the symmetry operations whose non-symmorphic translation vector tnons is zero. This can be done with the help of the input variable symmorphi
Gives the number of types of atoms. E.g. for
a homopolar system (e.g. pure Si) ntypat is 1, while for BaTiO3,
ntypat is 3.
Except when alchemical mixing of pseudopotentials is used, the number
of types of atoms will be equal to the number of pseudopotentials
npsp to be provided by the user.
Thus, the code will try to read the same number of pseudopotential files,
whose names should have been given in the "files" file.
The first pseudopotential will be assigned the type number 1, and so
on ...
Controls how input parameters nband, occ, and wtk are handled.
Give, in columnwise entry,
the three dimensionless primitive translations in real space, to be rescaled by
acell and scalecart.
If the Default is used, that is, rprim is the unity matrix,
the three dimensionless primitive vectors are three
unit vectors in cartesian coordinates. Each will be (possibly) multiplied
by the corresponding acell value, then (possibly)
stretched along the cartesian coordinates by the corresponding scalecart value,
to give the dimensional primitive vectors, called rprimd.
In the general case, the dimensional cartesian
coordinates of the crystal primitive translations R1p, R2p and R3p, see
rprimd, are
Alternatively to rprim, directions of dimensionless primitive vectors can be specified by using the input variable angdeg. This is especially useful for hexagonal lattices (with 120 or 60 degrees angles). Indeed, in order for symmetries to be recognized, rprim must be symmetric up to tolsym (10 digits by default), inducing a specification such as
rprim 0.86602540378 0.5 0.0 -0.86602540378 0.5 0.0 0.0 0.0 1.0that can be avoided thanks to angdeg:
angdeg 90 90 120
rprim "a" 0 0 0 "a" 0 "a/2" "a/2" "c/2" acell 3*1 scalecart 3*1 ! ( These are default values)The following is a valid, alternative way to define the same primitive vectors :
rprim 1 0 0 0 1 0 1/2 1/2 1/2 scalecart "a" "a" "c" acell 3*1 ! ( These are default values)Indeed, the cell has been stretched along the cartesian coordinates, by "a", "a" and "c" factors.
At variance, the following is WRONG :
rprim 1 0 0 0 1 0 1/2 1/2 1/2 acell "a" "a" "c" ! THIS IS WRONG scalecart 3*1 ! ( These are default values)Indeed, the latter would correspond to :
rprim "a" 0 0 0 "a" 0 "c/2" "c/2" "c/2" acell 3*1 scalecart 3*1 ! ( These are default values)Namely, the third vector has been rescaled by "c". It is not at all in the center of the tetragonal cell whose basis vectors are defined by the scaling factor "a".
This internal variable gives the dimensional real space primitive vectors, computed from acell, scalecart, and rprim.
Gives the scaling factors of cartesian coordinates by which
dimensionless primitive translations (in "rprim") are
to be multiplied.
rprim input variable,
the acell input variable,
and the associated internal rprimd internal variable.
Especially useful for body-centered and face-centered tetragonal lattices, as well as
body-centered and face-centered orthorhombic lattices, see rprimd.
Note that this input variable is NOT INTERNAL : its content will be immediately applied to rprim, at parsing time,
and then scalecart will be set to the default values. So, it will not be echoed.
It is used only when kptopt>=0,
and must be defined if nshiftk is larger than 1.
shiftk(1:3,1:nshiftk) defines
nshiftk shifts
of the homogeneous grid of k points
based on ngkpt or
kptrlatt.
The shifts induced by shiftk corresponds
to the reduced coordinates in the coordinate system
defining the k-point lattice. For example,
if the k point lattice is defined using ngkpt,
the point whose reciprocal space reduced coordinates are
( shiftk(1,ii)/ngkpt(1)
shiftk(2,ii)/ngkpt(2)
shiftk(3,ii)/ngkpt(3) )
belongs to the shifted grid number ii.
The user might rely on ABINIT to suggest suitable and efficient combinations of kptrlatt and shiftk. The procedure to be followed is described with the input variables kptrlen. In what follows, we suggest some interesting values of the shifts, to be used with even values of ngkpt. This list is much less exhaustive than the above-mentioned automatic procedure.
1) When the primitive vectors of the lattice do NOT form a FCC or a BCC lattice, the usual (shifted) Monkhorst-Pack grids are formed by using nshiftk=1 and shiftk 0.5 0.5 0.5 . This is often the preferred k point sampling. For a non-shifted Monkhorst-Pack grid, use nshiftk=1 and shiftk 0.0 0.0 0.0 , but there is little reason to do that.
2) When the primitive vectors of the lattice form a FCC lattice, with rprim
0.0 0.5 0.5 0.5 0.0 0.5 0.5 0.5 0.0the (very efficient) usual Monkhorst-Pack sampling will be generated by using nshiftk= 4 and shiftk
0.5 0.5 0.5 0.5 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.5
3) When the primitive vectors of the lattice form a BCC lattice, with rprim
-0.5 0.5 0.5 0.5 -0.5 0.5 0.5 0.5 -0.5the usual Monkhorst-Pack sampling will be generated by using nshiftk= 2 and shiftk
0.25 0.25 0.25 -0.25 -0.25 -0.25However, the simple sampling nshiftk=1 and shiftk 0.5 0.5 0.5 is excellent.
4) For hexagonal lattices with hexagonal axes, e.g. rprim
1.0 0.0 0.0 -0.5 sqrt(3)/2 0.0 0.0 0.0 1.0one can use nshiftk= 1 and shiftk 0.0 0.0 0.5
In rhombohedral axes, e.g. using angdeg 3*60., this corresponds to shiftk 0.5 0.5 0.5, to keep the shift along the symmetry axis.
Gives "nsym" 3x3 matrices
expressing space group symmetries in terms of their action
on the direct (or real) space primitive translations.
It turns out that these can always be expressed as integers.
Always give the identity matrix even if no other symmetries
hold, e.g.
symrel 1 0 0 0 1 0 0 0 1
Also note that for this array as for all others the array
elements are filled in a columnwise order as is usual for
Fortran.
The relation between the above symmetry matrices symrel,
expressed in the basis of primitive translations, and
the same symmetry matrices expressed in cartesian coordinates,
is as follows. Denote the matrix whose columns are the
primitive translations as R, and denote the cartesian
symmetry matrix as S. Then
symrel = R(inverse) * S * R
where matrix multiplication is implied.
When the symmetry finder is used (see nsym), symrel
will be computed automatically.
Gives the (nonsymmorphic) translation
vectors associated with the symmetries expressed
in "symrel".
These may all be 0, or may be fractional (nonprimitive)
translations expressed relative to the real space
primitive translations (so, using the "reduced" system
of coordinates, see "xred").
If all elements of the space
group leave 0 0 0 invariant, then these are all 0.
When the symmetry finder is used (see nsym), tnons
is computed automatically.
Sets a tolerance for absolute differences
of total energy that, reached TWICE successively,
will cause one SCF cycle to stop (and ions to be moved).
Can be specified in Ha (the default), Ry, eV or Kelvin, since
toldfe has the
'ENERGY' characteristics.
(1 Ha=27.2113845 eV)
If set to zero, this stopping condition is ignored.
Effective only when SCF cycles are done (iscf>0).
Because of machine precision, it is not worth to try
to obtain differences in energy that are smaller
than about 1.0d-12 of the total energy.
To get accurate stresses may be quite demanding.
When the geometry is optimized (relaxation of atomic positions or primitive vectors), the use of
toldfe is to be avoided. The use of toldff or
tolrff is by far preferable, in order to have a handle on the
geometry characteristics. When all forces vanish by symmetry (e.g. optimization of the lattice parameters
of a high-symmetry crystal), then place toldfe to 1.0d-12, or use (better) tolvrs.
Since toldfe, toldff,
tolrff,
tolvrs and tolwfr
are aimed at the same goal (causing the SCF cycle to stop),
they are seen as a unique input variable at reading. Hence, it is forbidden that two of these input variables
have non-zero values for the same dataset, or generically (for all datasets).
However, a non-zero value for one such variable for one dataset will have precedence on the non-zero value for another
input variable defined generically.
Sets a tolerance for differences of forces
(in hartree/Bohr) that, reached TWICE successively,
will cause one SCF cycle to stop (and ions to be moved).
If set to zero, this stopping condition is ignored.
Effective only when SCF cycles are done (iscf>0).
This tolerance applies to any particular cartesian
component of any atom, INCLUDING fixed ones.
This is to be used when trying to equilibrate a
structure to its lowest energy configuration (ionmov=2),
or in case of molecular dynamics (ionmov=1)
A value ten times smaller
than tolmxf is suggested (for example 5.0d-6 hartree/Bohr).
This stopping criterion is not allowed for RF calculations.
Since toldfe, toldff,
tolrff,
tolvrs and tolwfr
are aimed at the same goal (causing the SCF cycle to stop),
they are seen as a unique input variable at reading. Hence, it is forbidden that two of these input variables
have non-zero values for the same dataset, or generically (for all datasets).
However, a non-zero value for one such variable for one dataset will have precedence on the non-zero value for another
input variable defined generically.
Sets a tolerance for the ratio of differences of forces
(in hartree/Bohr) to maximum force, that, reached TWICE successively,
will cause one SCF cycle to stop (and ions to be moved) : diffor < tolrff * maxfor.
If set to zero, this stopping condition is ignored.
Effective only when SCF cycles are done (iscf>0).
This tolerance applies to any particular cartesian
component of any atom, INCLUDING fixed ones.
This is to be used when trying to equilibrate a
structure to its lowest energy configuration (ionmov=2),
or in case of molecular dynamics (ionmov=1)
A value of 0.02 is suggested.
This stopping criterion is not allowed for RF calculations.
Since toldfe, toldff,
tolrff,
tolvrs and tolwfr
are aimed at the same goal (causing the SCF cycle to stop),
they are seen as a unique input variable at reading. Hence, it is forbidden that two of these input variables
have non-zero values for the same dataset, or generically (for all datasets).
However, a non-zero value for one such variable for one dataset will have precedence on the non-zero value for another
input variable defined generically.
Sets a tolerance for potential
residual that, when reached, will cause one SCF cycle
to stop (and ions to be moved).
If set to zero, this stopping condition is ignored.
Effective only when SCF cycles are done (iscf>0).
To get accurate stresses may be quite demanding.
Additional explanation : the residual of the potential is the difference between the
input potential and the output potential, when the latter is obtained from the density
determined from the eigenfunctions of the input potential. When the self-consistency
loop is achieved, both input and output potentials must be equal, and the residual
of the potential must be zero. The tolerance on the
potential residual is imposed by first subtracting the mean of the residual of the potential
(or the trace of the potential matrix, if the system is spin-polarized),
then summing the square of this function over all FFT grid points. The result should be
lower than tolvrs.
Since toldfe, toldff,
tolrff,
tolvrs and tolwfr
are aimed at the same goal (causing the SCF cycle to stop),
they are seen as a unique input variable at reading. Hence, it is forbidden that two of these input variables
have non-zero values for the same dataset, or generically (for all datasets).
However, a non-zero value for one such variable for one dataset will have precedence on the non-zero value for another
input variable defined generically.
The signification of this tolerance depends on
the basis set. In plane waves, it gives a convergence tolerance for the
largest squared "residual" (defined below) for any
given band. The squared residual is:
< nk|(H-E)2|nk >, E = < nk|H|nk >
Note that tolwfr is often used in the test cases, but this is tolwfr purely for historical reasons : except when iscf<0, other critera should be used.
In the wavelet case (see usewvl = 1), this criterion is the favoured one. It is based on the norm 2 of the gradient of the wavefunctions. Typical values range from 5*10-4 to 5*10-5.
Array giving an integer label to every atom in the unit
cell to denote its type.
The different types of atoms
are constructed from the pseudopotential files.
There are at most ntypat types of atoms.
As an example, for BaTiO3, where the pseudopotential for Ba is number 1,
the one of Ti is number 2, and the one of O is number 3, the actual
value of the typat array might be :
typat 1 2 3 3 3
Used to define the set of indices in the multi-data set
mode, when a double loop is needed (see later).
The values of udtset(1) must be between 1 and 999,
the values of udtset(2) must be between 1 and 9, and their
product must be equal to ndtset.
The values of jdtset are obtained by
looping on the two indices defined by udtset(1) and udtset(2) as follows :
do i1=1,intarr(1) do i2=1,intarr(2) idtset=idtset+1 dtsets(idtset)%jdtset=i1*10+i2 end do end doSo, udtset(2) sets the largest value for the unity digit, that varies between 1 and udtset(2).
Used to define if the calculation is done on a
wavelet basis set or not.
The values of usewvl must be 0 or 1. Putting usewvl
to 1, makes icoulomb
mandatory to 1. The number of band (nband) must be set manually to
the strict number need for an isolator system (i.e.
number of electron over two). The cut-off is not relevant in the
wavelet case, use wvl_hgrid
instead.
In wavelet case, the system must be isolated systems (molecules or
clusters). All geometry optimization are available (see ionmov, especially the geometry
optimisation and the molecular dynamics.
The spin computation is not currently possible with wavelets and
metalic systems may be slow to converge.
Gives the k point weights.
The
k point weights will have their sum (re)normalized to 1
(unless occopt=2 and kptopt=0;
see description of occopt)
within the program and therefore may be input with any
arbitrary normalization. This feature helps avoid the
need for many digits in representing fractional weights
such as 1/3.
wtk is ignored if iscf is not positive,
except if iscf=-3.
It gives the step size in real space for the grid resolution in the wavelet basis set. This value is highly responsible for the memory occupation in the wavelet computation. The value is a length in atomic units.
Gives the cartesian coordinates
of atoms within unit cell, in angstrom. This information is
redundant with that supplied by array xred or xcart.
If xred and xangst are ABSENT from the input file and
xcart is
provided, then the values of xred will be computed from
the provided xcart (i.e. the user may use xangst instead
of xred or xcart to provide starting coordinates).
One and only one of xred, xcart
and xangst must be provided.
The conversion factor between Bohr and Angstrom
is 1 Bohr=0.5291772108 Angstrom, see the NIST site.
Atomic positions evolve if ionmov/=0 .
In constrast with xred and
xcart, xangst is not internal.
Gives the cartesian coordinates
of atoms within unit cell. This information is
redundant with that supplied by array xred or xangst.
By default, xcart is given in Bohr atomic units
(1 Bohr=0.5291772108 Angstroms), although Angstrom can be specified,
if preferred, since xcart has the
'LENGTH' characteristics.
If xred and xangst are
ABSENT from the input file and xcart is
provided, then the values of xred will be computed from
the provided xcart (i.e. the user may use xcart instead
of xred or xangst to provide starting coordinates).
One and only one of xred, xcart
and xangst must be provided.
Atomic positions evolve if ionmov/=0 .
Gives the atomic locations within
unit cell in coordinates relative to real space primitive
translations (NOT in cartesian coordinates). Thus these
are fractional numbers typically between 0 and 1 and
are dimensionless. The cartesian coordinates of atoms (in Bohr)
are given by:
where (t1,t2,t3) are the "reduced coordinates" given in
columns of "xred", (rprimd1,rprimd2,rprimd3) are the columns of
primitive vectors array "rprimd" in Bohr.
If you prefer to work only with cartesian coordinates, you
may work entirely with "xcart" or "xangst" and ignore xred, in
which case xred must be absent from the input file.
One and only one of xred, xcart
and xangst must be provided.
Atomic positions evolve if ionmov/=0 .
Gives nuclear charge for each
type of pseudopotential, in order.
If znucl does not agree with nuclear charge,
as given in pseudopotential files, the program writes
an error message and stops.
N.B. : In the pseudopotential files, znucl is called "zatom".
For a "dummy" atom, with znucl=0 , as used in the case of calculations with only a jellium surface, ABINIT sets arbitrarily the covalent radius to one.