Search:
ALLOW • BASIS • CONFINE • CONFINEMENT • CONVERGENCE • CPVECTOR • DEBUG • DEPENDENCY • DIIS • DIRIS • EIGTHRESHOLD • FERMI • FILELIMIT • INTEGRATION • IOVECTOR • KGRP0 • KGRPX • KSPACE • NUELSTAT • NVELSTAT • PRINT • OCCUPATIONS • POPTHRESHOLD • POTENTIALNOISE • SCF • SCREENING • WORKSPACE
debugging feature to let the program continue even when intermediate results seem to be wrong or very inaccurate. Currently there are only few places in the program where this key is used. Argument to the key is what should be allowed (e.g. BadIntegration).
Block key containing options and thresholds related to the basis set.
Check the dependency of the valence basis on the frozen core orbitals, by analysis of the core-valence overlap matrix. Since this takes some cpu time, it is by default off.
Applies only in fragments calculations. By default the start-up density is taken as sum-of-fragments. This key specifies that the density is constructed from the fragment orbitals after these have been mutually orthonormalized (Pauli principle). It's only related to SCF convergence considerations. Depending on the system it may or may not improve the required number of SCF cycles.
By default it is assumed that the core states display some small but non-negligible dispersion. Using this key counteracts this assumption. The core Bloch functions are then computed only in the G-point and assumed to be identical for all k-points in the Brillouin Zone.
This options switches on transformation to the orthogonal basis of the Bloch sums of elementary one-center basis functions. Such behaviour was default in the versions prior to 2005.01. Starting from ADF2005.01, if this option is not specified, BAND uses a general eigenvalue solver that takes the non-trivial overlap matrix into account instead of diagonalizing the Fock matrix during SCF.
The real argument (default zero) sets a threshold for setting elements to zero in transformation matrices (typically, for transforming the bloch basis to an orthonormal one, and similar transformations) whenever the absolute value is below the threshold.
CONFINE (block-type): global version (not recommended!)
Instead of this global confine option, we recommend the smooth confinement of each atom type, making use of a Fermi-Dirac function. This preferred option is documented below and should lead to numerically more stable results.
A confinement approach has been implemented in BAND to squeeze the radial atomic functions a bit faster to zero. This can be achieved using three different methods: multiplication by a confining function (method exponential), by replacing the atomic function (method polynomial) or by scaling the atomic function (method scaling).
CONFINEMENT
{ Method { exponential | polynomial | scaling } }
{ Rmatch rmin }
{ Rcut rmax }
{ Order norder }
End
The confinement is applied between rmin and rmax, though due to renormalization, the entire function will change (also in the inner region). Before rmin the function will not be adapted apart from the renormalization, after rmax the function becomes zero. Obviously, confinement of functions may have some impact on the numerical outcomes. The value norder is used for the scaling technique. Currently only works for non-relativistic calculations.
CONFINEMENT (block-type): subkey of atomtype
The global confinement option leads to discontinuous higher order derivatives in the basis function, leading to numerical instability. It is better to use smooth confinement. Smooth confinement means that the radial part of a function is multiplied with a Fermi-Dirac (FD) function. We have implemented smooth confinement at the atom type level. In a slab calculation this allows one to use different settings for surface atoms than for those in inner layers. You can specify the range and the decay speed of the FD function like
Confinement
Radius 7
Delta 0.7
SubEnd
If the decay 'Delta' is not specified it defaults to 0.1*'Radius'. Relativistic effects are treated correctly. If the confinement option is used in combination with the TAILS option, the calculation may run significantly faster. The confinement option can also be useful without the tails option in cases where near linear dependency reduces the numerical reliability of the results (cf. dependency keyword). This typically occurs for highly coordinated systems with large basis sets containing diffuse functions. For such cases, the confinement option (possibly in combination with the dependency keyword) reduces the numerical problems.
All options and parameters related to the convergence behavior of the SCF are defined in this block key. Also the finite temperature distribution is part of this.
Number of points by which the finite temperature Fermi-Dirac distribution of electronic occupations is approximated. Advise: don't use.
If smoothing of occupations over nearly degenerate orbitals is applied (see the key DEGENERATE)), then, if this key is set in the input file, the program will limit the smoothing energy range to 1e-4 a.u. as soon as the SCF has converged 'halfway', i.e. when the SCF error has decreased to the square root of its convergence criterion.
This key prevents any internal automatic setting of the key DEGENERATE, see that key's description.
Defines the distribution of occupations around the Fermi level. By default (T=10), the effect is that of zero temperature. In fact this key is a preliminary to a future full implementation of finite temperature effects; currently it has no sensible application. See, however, the ALLBANDS key.
The code is vectorized and this key can be used to set the vector length. Default depends on the machine and should be set at the installation of the program.
To quickly implement a new key. DEBUG takes as argument a key name. The programmer can check for this in the same way as PRINT or ALLOW keys.
DEBUG LNOSPN
Applies only in a spin-unrestricted calculation. The program assumes that energy bands are constituted of the results at the discrete k-points that correspond in energy-ordering: the first band is made up of all lowest eigenvalues across the BZ, the second band of the second lowest values, et cetera. This procedure is, by default, carried out independently for both spins. Using the key 'mixes' spin-alpha and spin-beta orbitals and allows spin-mixed bands, so to speak. This affects the calculation of occupation numbers in case of partially filled bands. It is primarily a testing and debugging tool.
Criterion for dependency of the basis and fit set.
DEPENDENCY {basis tolbas} {core tolcor} {fit tolfit} {corevalence tolovl}
basis
Smallest eigenvalue of the overlap matrix of normalized bloch functions. Default 1e-5. See also the discussion in 'Recommendations & Problems' about basis set dependency.
core
The program verifies that the frozen core approximation is reasonable, by checking the smallest value of the overlap matrix of the core (bloch) orbitals against this criterion. Default: 0.98
fit
Criterion for dependency of the total set of fit functions. The value monitored is the smallest eigenvalue of the overlap matrix of normalized bloch sums of symmetrized fit functions. Default=1e-6.
corevalence
Criterion for dependency of the core functions on the valence basis. The maximum overlap between any two normalized functions in the two respective function spaces should not exceed 1.0-coreval(...). Default=1e-5.
The diis procedure to obtain the SCF solution in the crystal depends on several parameters. Default values can be overruled with this key-block. Each option must be specified, it at all, on a separate record in the data block:
DIIS
option1 value1
option2 value2
...
END
Recognized options are:
condition: the condition number of the DIIS matrix, the largest eigenvalue divided by the smallest, must not exceed this value; default 1e6
clarge: when the largest coefficient in the DIIS expansion exceeds this value, damping is applied; default 20
chuge: when the largest DIIS coefficient exceeds this value, the oldest DIIS vector is removed and the procedure re-applied. Default 50
dimix: mixing parameter when damping is used, rather than the DIIS procedure. By default off (dimix=1.0): result is taken as dimix*value + (1-dimix)*previous
nvctrx: maximum number of DIIS expansion vectors. Default 20
ncycledamp: number of initial iterations where damping is applied, before any DIIS is considered. Default 5
potential: (no argument): presence of this string means that the DIIS method will be applied to the potential, rather than to the valence density (default). It excludes the competing option deformationdensity.
deformationdensity: (no argument): presence of this string means that the DIIS method will be applied to the deformation density, rather than to the valence density (default). This option excludes the potential option, see previous.
print: turns on a print switch to report details of the used DIIS procedure. Default off.
special: (no argument) will let the program try to optimize the mixing parameter (dimix) and adjust it when difficulties occur. It is not certain that this may not make things worse!
comstr: its argument is only a string to be printed as label to output, if any, of the DIIS parameter.
Completely similar to the DIIS key, except that this one applies to the DIIS procedure used in the DIRAC subprogram, for numerical single atom calculations.
Components smaller (absolute value) than this parameter (default 1e-2) are not printed in the output of the DOS section, where the breakdown of crystal orbitals in the primitive basis is output.
This key sets technical parameter used in the search for the Fermi energy, which is carried out at each cycle of the SCF procedure. All applicable options must be specified, if at all, in separate records in the data block.
FERMI
option1 value1
option2 value2
(etc)
END
Recognized options:
maxtry: maximum number of attempts to locate the Fermi energy accurately. Default 50. The procedure is iterative in nature, narrowing the energy band in which the Fermi energy must lie, between an upper and a lower bound. If the procedure has not sufficiently converged within maxtry iterations, the program takes a reasonable value and constructs the charge density by interpolation between the functions corresponding to the last used upper and lower bounds for the Fermi energy
delta: converge criterion: upper and lower bounds for the Fermi energy and the corresponding integrated charge volumes must be equal within delta. Default 1e-4
eps: after convergence of the Fermi energy search procedure, a final estimate is defined by interpolation and the corresponding integrated charge volume is tested. It should be exact, to machine precision. Tested is that it deviates not more than eps. Default 1e-10
maximum amount of data (in bytes) on one logical file. Default depends on the machine and should be set at the installation of the program. Advise: Choose as large as possible, to keep the number of files limited. The automatic algorithm that cuts the files to pieces does not work flawlessly. Using a high file limit effectively disables the use of this mechanism. Be aware however of the maximum integer value on your machine!
parameter-specifications for the generation of numerical integration points and weights. Most data records must be of the form 'parameter value'. The most important parameter is accint, which is defaulted to the value of key Accuracy. Unless one is very familiar with the details of the numerical integration package, we strongly recommend not to use the INTEGRATION-key, and to specify only Accuracy. More information can be found in the literature. A key that is specific to BAND (cannot be used in ADF) is the key Pirpt3. This invokes a slightly different generator for the numerical integration grid. Like the standard method, it is based on an atomic cellular partitioning of space, but it differs in the treatment of the truncated pyramids, by more strictly monitoring test functions. It generates more points (which means: increased CPU times and disk storage), and in some cases it yields more accurate results. General advise is hard to give.
I/O actions to and from files is segmented in blocks of iovector. Default depends on the machine and should be set at the installation of the program.
treatment of the k-points (integration over the Brillouin Zone) in
the preparation phase (construction of bloch basis functions, computation of
overlap matrix, and so on, in each k-point)
is in blocks of kgrp0 points at the same time. Note: last character of
this key is a zero (not the letter "o").
During the preparation phase increasing kgrp0
may reduce the CPU time. This depends also on available workspace). However,
larger kgrp0 values result in more files being open at the same time, and more data being stored on disc during this stage of
the program. By default the program tries to optimize kgrp0 only
with respect to the expected effect on cpu-time.
If the key is used, the actual value of kgrp0 may
differ slightly from the input-value: from the input-value the program computes
first the number of blocks of k-points;
then kgrp0 is re-computed by distributing the total
number of k-points equally over the blocks. To restrict the
size of the blocks via input it is most convenient to use the
key kgrpx, rather than kgrp0.
is an absolute upper bound on kgrp0 as computed by the program.
Various settings of the k-space integration can be supplied here. The simple form of this key is the accuracy of the integration and can be specified within the block key as Kinteg (see also the first list).
invokes the hybrid quadratic (rather than fully ) quadratic integration method over the BZ. It is meaningful only for 2D Brillouin Zones that would otherwise use a fully quadratic procedure (odd-valued k-space integration parameter ). In all other cases the key is ignored.
See key Kspace in the first list. This defines the accuracy of the integration of the reciprocal space.
secondary parameter for integration over the Brillouin Zone: some aspects in the quadratic method may be carried out in fact by using a fine-grid linear-tetrahedron method (: hybrid approach). kmesh defines this linear-method mesh. Default=2 (invariably found to be adequate).
Integration over the BZ is carried out with the analytic quadratic tetrahedron method. If the input KSPACE key is set to an even number, this approach is not used and the linear tetrahedron method is used (usually far inferior). To invoke the linear method also for odd values of KSPACE, insert the LinearKspace key (no argument) in the input file.
The BZ sampling grid is generated in a summation over simplices that build the irreducible wedge of the BZ, considering only the geometric symmetry of the BZ itself. Any additional k-points required, when the atomic positions in the unit cell imply symmetry lowering, are added by symmetry operations. In pre-98 releases of BAND the BZ grid was generated in the symmetry unique part of the BZ considering the real space symmetry (which could, therefore, be a lower symmetry) directly. The difference between the two methods is in particular relevant when doing comparison calculations where the systems differ only in the atomic positions implying different space group symmetries. One would then want to use the same BZ grid. This is the case since release 98, but was not (by default) so in earlier versions. The original approach can be selected by using this key..
Suppresses integration in k-space in one or more directions. May be used for instance if the 3D Brillouin Zone is very extended in one or two dimensions and of 'normal' size in the other dimension(s). We plan to remove this key in the future. If it is given in input, with an integer value, integration will be suppressed in the indicated number of dimensions.
Electrostatic interaction integrals between spherical atomic
densities are computed by numerical integration over an elliptic grid. Nuelstat
is the outward (parabolic) coordinate number of integration points.
Default: 50
Electrostatic interaction integrals between spherical atomic
densities are computed by numerical integration over an elliptic grid. Nvelstat
is the angular (elliptic) coordinate number of integration points.
Default: 80
One or more strings (separated by blanks or comma's) from a pre-defined set may be typed after the key. The following names replace the IRPNT[IPRSE] keys of the previous versions of BAND
PrepNone, PrepMore, PrepDetail replace IPRNTP 0, 2, 5
IntNone, IntMore, IntDetail replace IPRNTI 0, 2, 5
FrmNone, FrmMore, FrmDetail replace IPRNTR 0, 2, 5
SCFNone, SCFMore, SCFDetail replace IPRNTS 0, 2, 5
EigNone, EigMore, EigDetail replace IPRNTE 0, 2, 5
Allows to input specific occupations numbers. Applies only for calculations that use only one k-point (i.e. pseudo-molecule calculations).
OCCUPATIONS
irrepno occupations_alpha {// occupations_beta}
...
End
the irrepno must be 1,
unless symmetry is used (an unsupported option, currently).
occupations_beta, and the separating double slash (//) must not be used in a spin-restricted
calculation.
occupations_alpha/beta is a sequence
of values assigned to the states ('bands') in energy ordering.
This allows you, for instance, to specify an empty state below occupied ones.
Threshold for printing Mulliken population terms. Default 1e-2
The initial potential for the SCF procedure is constructed from a sum-of-atoms density. Added to this is some small noise in the numerical values of the potential in the points of the integration grid. The purpose of the noise is to help the program break the initial symmetry, if that would lower the energy, by effectively inducing small differences between (initially) degenerate orbitals. The noise in the potential is randomly generated between zero and an upper limit, which is 1e-4 a.u. by default. The key, which must have a numerical argument, adjusts this upper limit. This can be used therefore to suppress the noise by choosing zero, or to increase it by specifying some large number.
Contains the same data as the ADF-MOL key with the same name (except for DIIS procedure parameters).
The program knows two alternative ways to evaluate the charge density iteratively in the SCF procedure: from the P-matrix, and directly from the squared occupied eigenstates. By default the program actually uses both at least one time and tries to take the most efficient. Eigenstates turns off this comparison and lets the program stick to one method (from the eigenstates).
Evaluate the charge density from the P-matrix. See also the subkey Eigenstates.
Minimum rate of convergence for the SCF procedure. If progress is too slow the program will take measures (such as smearing out occupations around the Fermi level, see subkey degenerate of key convergence) or, if everything seems to fail, it will stop. Default=0.99
To disturb degeneracy of alpha and beta spin MOs the value of this key is added to the beta spin potential at the startup. Default is 5E-2.
Parameters that influence the screening and tails of basis functions. Recognized options are
Criterion for negligibility of tails in the construction of Bloch sums. Default depends on Accuracy.
One of the parameters that define the screening of Coulomb-potentials in lattice sums. Depends by default on Accuracy, rmadel, and rcelx. One should consult the literature for more information.
Max. distance of lattice site from which tails of atomic functions will be taken into account for the bloch sums. Default depends on Accuracy.
One of the parameters that define screening of the Coulomb potentials in lattice summations. Depends by default on Accuracy, dmadel, rcelx. One should consult the literature for more information.
Real space lattice sums of slowly (or non-) convergent terms, such as the Coulomb potential, are computed by a screening technique. In previous releases, the screening was applied to all (long-range) Coulomb expressions. Starting from BAND98 screening is only applied in the periodicity directions. This key restores the original situation: screening in all directions.
SKIP
followed by any number of strings (separated by blanks or commas) tells the program to skip certain parts. Should only be used by those who know what they're doing. Recognized are certain pre-defined strings. Useful argument may be eigenvalues (to suppress printing the eigenvalues at the (first and last) SCF cycles).
Options for the workspace manager. Do not use, but use settings file.
smallest block of memory to allocate to store integers in (in Megabytes).
smallest block of memory to allocate to store logicals in (in Megabytes)
smallest block of memory to allocate to store reals in (in Megabytes).
smallest block of memory to allocate to store strings in (in Megabytes).