ABINIT, the main code¶
This document explains the IO parameters and format needed for the main code (abinit) in the ABINIT package.
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. Many user guides are also present on the Web. As an example, for calculating response properties using abinit, the complementary respfn help file is needed. Some additional specialized documentation is not available on the Web, but inside the package, but this should be for advanced users only.
1 How to run the code?¶
The main executable file is called abinit. Supposing that the input file is called run.abi, and that the executable is placed in your working directory, abinit is run interactively (in Unix) with the command
abinit run.abi >& log
or, in the background, with the command
abinit run.abi >& log &
where standard out and standard error are piped to the log file called “log” (piping the standard error, thanks to the ‘&’ sign placed after ‘>’ is really important for the analysis of eventual failures, when not due to ABINIT, but to other sources, like disk quota issues). It is also possible to separate the standard error file from the standard output file with the command
abinit run.abi > log 2> err &
or even more explicitly
abinit run.abi > run.log 2> run.err &
Actually, the user can specify any names he/she wishes for any of these files. The suffix .abi is the most usual for the abinit input file, likewise the suffix .abo for the main output file. Nevertheless, the names of all input/output/temporary abinit files can also be tuned. See e.g. output_file or topic_Control.
Variations of the above commands could be needed, depending on the flavor of Unix that is used
on the platform that is considered for running the code.
If you do not underderstand the standard Unix syntax above, please get familiarized with Unix before continuing.
2 The underlying theoretical framework and algorithms¶
The methods employed in this computer code to solve the electronic structure problem are described in part in different review papers as well as research papers. The code is an implementation of the Local Density Approximation to the Density Functional Theory, based upon a plane wave basis set and separable pseudopotentials. The iterative minimization algorithm is a combination of fixed potential preconditioned conjugate gradient optimization of wavefunction and a choice of different algorithms for the update of the potential, one of which is a potentialbased conjugate gradient algorithm.
The representation of potential, density and wavefunctions in real space will be done on a regular 3D grid of points. Its spacing will be determined by the cutoff energy ecut of the planewave basis in reciprocal space. This grid of points will also be the starting point of Fast Fourier Transforms between real and reciprocal space. The number of such points, called ngfft, should be sufficiently large for adequate representation of the functions, but not too large, for reasons of computational efficiency. The tradeoff between accuracy and computational efficiency is present in many places of the code, and addressed briefly at the end of the present help file.
We recommend a good introduction to many different concepts valid for this code, available in this Reviews of Modern Physics article. Note that this paper does NOT reflect the present status of the code. ABINIT is closer in spirit to the paper of Kresse and Furthmuller. If you have never used another electronic structure code or a Quantum Chemistry package, you should browse through the Chaps. 1 to 13, and appendices L and M of this book by R. M. Martin
3 The input file¶
3.1 Format of the input file¶
Note that this input file was called ab_in in the example of the introduction. We first explain the content of the input file without use of the multidataset possibility that will be explained in section 3.3.
The parameters are input to the code from a single input file. Each parameter is provided by giving the name of the input variable and then placing the numerical value(s) after the name, separated by one or more spaces, or even an equal sign (equal signs are replaced by blanks by the parser). Depending on the input variable, the numerical value may be an integer or a real number (internal representation as double precision number), or a character string (delimited with double quotes) and may actually represent an array of values. If it represents an array, the next set of numbers separated by spaces are taken as the values for the array.
 Do NOT separate a minus sign from the number to which it applies.
 Do NOT use tabs.
 NOTE THAT NO LINE OF THE INPUT FILE MAY EXCEED 132 CHARACTERS. That is, only the first 132 characters of each line of the input file will be read and parsed for input variables and their values.
The names of all the parameters can be found in the input variable database. The list of input variables present in the latter file links them to their definitions, contained in different “variable set” files, some of which are listed here:
 Basic variables, basic varset
 Files handling variables, files varset
 Groundstate calculation variables, gstate varset
 GW variables, gw varset
 Parallelisation variables, paral varset
 Density Functional Perturbation Theory variables, dfpt varset
In the actual input file, these parameters may be given in any desired order,
and more than one may be given per line.
Spaces are used to separate values and additional spaces are ignored.
An as example of input, the parameter for length scales is called acell
and is an array acell(3) for the lengths of the primitive translations in
Bohr atomic units.
To input a typical Si diamond lattice one would have the line
acell 10.25311 10.25311 10.25311
in the input file. This may equivalently be written
acell 3*10.25311
and will still be parsed correctly: it is equivalent to the above line. Even
acell *10.25311
will work. In the latter case the ‘‘ sign means that the parser should use
the given value to fill the array, by repeating it as **many time as needed*.
Multiple spaces are ignored, as is any text which does not contain the
character strings which correspond to some input parameters. In case of
arrays, only the needed numbers will be considered, and the eventual numbers
after those needed will also be ignored. For example,
natom 3 # This gives the number of atoms
typat 1 1 2 2 3 # typat(1:natom) gives the type of each atom: only
# the first three data are read, since [[natom]]=3 `
A given variable is identified by the parser by having at least one blank
before it and after it (again, multiple blanks are irrelevant).
ABINIT has also some (very limited) interpretor capabilities:

It can identify one slash sign (/) being placed between two numbers (without a separating blank) as being the definition of a fraction (e.g. ⅓ will be interpreted as 0.33333333333333d0);

It can identify sqrt(…) or sqrt(…) as being the definition of a square root, when applied to one valid number  also without a separating blank  (e.g. sqrt(0.75) will be interpreted as 0.8660254038d0);

Note, however, that these capabilities are NOT recursive. At most, a sqrt identifier can contain an expression that uses a fraction (e.g. sqrt(¾) is OK), but two fractions (or two sqrt) cannot be used in one expression, and a sqrt cannot be present in the numerator or denominator of a fraction.

A string might be formed by concatenating two strings with the // operator (even blanks before or after // are accepted)

Environment variables are accepted, in the form
$VAR
where VAR is the name of the environment variable. The end of the name of the environment variable, VAR, is determined by finding the closest separator among a blank, a slash, or a double quote. The parser will automatically substitute the value of the environment variable to the$VAR
string. As an example, “$PSPDIR/PseudosHGH_pwteter” with PSPDIR being Psps_for_tests will give “Psps_for_tests/PseudosHGH_pwteter”.
Comments should be placed to the right of the comment characters # or !; anything to the right of a “#” or a “!” on any line is simply ignored by the parser. Additional text, not preceded by a “#” or a “!” would not otherwise cause trouble unless the text inadvertently contained character strings which were the same as variable names (e.g. acell). The characters “#” or “!” can also be used to “store” old values of variables or place anything else of convenience into the file in such a way as to be ignored by the parser when the data is read.
Case is irrelevant as the entire input string is mapped to upper case before
parsing, to remove case sensitivity.
More than one parameter per line may be given. If a given parameter name is
given more than once in the input file, an error message is printed, and the code stops.
External input files can be included with the syntax:
include "geometry.inc"
where geometry.inc gives the crystalline structure in the Abinit format:
cat geometry.in
# Si in diamond structure
acell 3*10.25
rprim
0.0 0.5 0.5
0.5 0.0 0.5
0.5 0.5 0.0
natom 2
ntypat 1
typat 2*1
xred 0.00 0.00 0.00
0.25 0.25 0.25
znucl 14.0
Alternatively, one can use the structure variable to read the structure from POSCAR files, netcdf files produced by Abinit or text files with a reduced set of ABINIT variables.
3.2 File names in ABINIT¶
File names in ABINIT are either given automatically by ABINIT, or build from different input variables, like output_file, pseudos, indata_prefix, outdata_prefix or tmpdata_prefix.
Filename of the main file in which formatted output will be placed (the main output file). Error messages and other diagnostics will NOT be placed in this file, but sent to unit 06 (terminal or log file); the unit 06 output can be ignored unless something goes wrong. The code repeats a lot of information to both unit 06 and to the main output file. The unit 06 output is intended to be discarded if the run completes successfully, with the main output file keeping the record of the run in a nicer looking format.
Give filenames of the different pseudopotential input files. The pseudopotential data files are formatted. There must be as many filenames provided sequentially here as there are types of atoms in the system, and the order in which the names are given establishes the identity of the atoms in the unit cell, as listed in typat.
abi or indata_prefix
The other files READ by the code will have a name that is constructed from the root “abi” or another root defined by indata_prefix. This apply to optionally read wavefunction, density or potential files. In the multidataset mode, this root will be complemented by _DS and the dataset index. The list of possible input files, with their name created from the root ‘abi’, is the following (a similar list exist when _DS and the dataset index are appended to ‘abi’):

abi_WFK filename of file containing input wavefunction coefficients created from an earlier run (with nqpt=0). Will be opened and read if irdwfk is 1. The wavefunction file is unformatted and can be very large. Warning: in the multi dataset mode, if getwfk is nonzero, a wavefunction file build from abo will be read.

abi_WFQ filename of file containing input wavefunction coefficients created from an earlier run (with nqpt=1), as needed for response function calculations. The wavefunction file is unformatted and can be very large. Warning: in the multi dataset mode, if getwfk is nonzero, a wavefunction file build from abo will be read.

abi_1WFxx filename of file containing input firstorder wavefunctions created from an earlier RF run. xx is the index of the perturbation

abi_DEN filename of file containing density created from an earlier run. See explanations related to negative values of iscf. This file is also unformatted. Warning: in the multi dataset mode, if getwfk is nonzero, a density file build from abo will be read.

abi_HES filename of file containing an approximate hessian, for eventual (re)initialisation of Broyden minimisation. See brdmin.F90 routine. The use of restartxf is preferred.
abo or outdata_prefix
Except output_file and “log”, the other files WRITTEN by the code will have a name that is constructed from the root “abo” or outdata_prefix. This applies to optionally written wavefunction, density, potential, or density of states files. In the multi dataset mode, this root will be complemented by _DS and the dataset index. Also in the multidataset mode, the root “abo” can be used to build the name of input files, thanks to the ‘get’ variables. The list of possible output files, with their name created from the root ‘abo’ is the following (a similar list exists when _DS and the dataset index are appended to ‘abo’):

abo_WFK Filename of file containing output wavefunction coefficients, if nqpt=0. The wavefunction file is unformatted and can be very large.

abo_WFQ Same as abo_WFK, but for the case nqpt=1. The wavefunctions are always output, either with the name abo_WFK, or with the name abo_WFQ.

abo_1WFxx Same as abo_WFK, but for firstorder wavefunctions, xx is the index of the perturbation, see the section section 6.3 of the respfn help file.

abo_DDB The derivative database, produced by a responsefunction dataset, see this section of the respfn help file.

abo_DEN filename of file containing density, in the case ionmov=0. See the keyword prtden. This file is unformatted, but can be read by cut3d.

abo_TIMx_DEN filenames of files containing density, in the case ionmov/=0. The value of “x” after ” TIM ” is described hereafter. See the keyword prtden. This file is unformatted, but can be read by cut3d.

abo_POT filename of file containing KohnSham potential See the keyword prtpot. This file is unformatted, but can be read by cut3d.

abo_TIMx_POT filenames of files containing KohnSham potential in the case ionmov/=0. The value of “x” after “TIM” is described hereafter. See the keyword prtpot. This file is unformatted, but can be read by cut3d.

abo_DOS filename of file containing density of states. See the keyword prtdos. This file is formatted.

abo_TIMx_DOS filenames of files containing the density of states in the case prtdos=2 and ionmov=1 or 2. The value of “x” after “TIM” is described hereafter. See also the keyword prtdos. This file is formatted.

abo_GEO filename of file containing the geometrical analysis (bond lengths and bond angles) in the case ionmov=0. See the keyword prtgeo. This file is formatted.

abo_TIMx_GEO filenames of files containing the geometrical analysis (bond lengths and bond angles) in the case ionmov=1 or 2. The value of “x” after “TIM” is described hereafter. See also the keyword prtgeo. This file is formatted.

abo_KSS filename of file containing output wavefunction coefficients, if nbandkss/=0. This wavefunction file is unformatted and can be very large. Its purpose is to start a GW calculation using M.Torrent’s code. A different format than for abo_WFK is used, see the file ~abinit/doc/developers/format_KSS.txt.

abo_EIG A file containing the electronic eigenvalues, for subsequent plotting of band structure.
When ionmov/=0, the POT, DEN, or GEO files are output each time that a SCF cycle is finished. The ” x ” of TIMx aims at giving each of these files a different name. It is attributed as follows:  case ionmov==1: there is an initialization phase, that takes 4 calls to the SCF calculation. The value of x will be A, B, C, and D. Then, x will be 1, 2, 3 …, actually in agreement with the value of itime (see the keyword ntime)  other ionmov cases: the initialisation phase take only one SCF call. The value of x will be 0 for that call. Then, the value of x is 1, 2, 3… in agreement with the value of itime (see the keyword ntime)
tmp or tmpdata_prefix
The temporary files created by the codes will have a name that is constructed
from the root ” tmp ” or tmpdata_prefix. tmp should usually be chosen such as to give access
to a disk of the machine that is running the job, not a remote (NFS) disk.
Under Unix, the name might be something like /tmp/user_name/temp
. As an
example, tmp_STATUS
gives the status of advancement of the calculation, and is updated very
frequently
3.3 More about ABINIT input variables¶
In each section of the ABINIT input variables files, a generic information on the input variable is given: a mnemonics, possibly some characteristics, the variable type (integer, real, string), and the default value. Then, follows the description of the variable.
The characteristics can be one of the following:
 DEVELOP,
 NO_MULTI
 INTERNAL_ONLY
 INPUT_ONLY
 EVOLVING
 ENERGY
 LENGTH
 MAGNETIC FIELD
Physical information¶
The ENERGY, LENGTH and MAGNETIC FIELD characteristics indicate that the physical meaning of the variable is known by ABINIT, so that ABINIT can treat its physical dimensions with some units.
The use of the atomic unit system (e.g. the Hartree for energy, about 27.211 eV, and the Bohr for lengths about 0.529 Angstroms) is strictly enforced within the code. However, the dimension of some input variables can be specified and read correctly.
At present, this applies to three types of variables: those that have the dimension of an energy, those that have a dimension of length, and those that have a dimension of magnetic field. The first class of variables have the characteristics ENERGY, and can be specified in atomic units (Hartree), or electronvolts, or Rydbergs, or even Kelvin.
The second class of variables have the characteristics LENGTH, and can be specified in atomic units (Bohr), nm (nanometer) and angstrom. The third class of variables have the characteristics MAGNETIC FIELD, and can be specified in atomic units and Tesla. The abinit parser recognize a dimension if it is specified after the list of numbers following the input variable keyword, in the input file. The specification can be upper or lower case, or a mix thereof. Here is the list of recognized chains of characters:
 Ry or Rydberg or Rydbergs → Rydberg (for energies)
 eV → electronvolts (for energies)
 K → Kelvin (for energies)
 Angstr → Angstrom (for lengths)
 nm → nanometer (for lengths)
 T or Tesla → Tesla (for magnetic fields)
 S or Sec or Second → second (for time)
Other character chains, like “au” (for atomic units) or “Hartree”, or “Bohr” are not recognized, but make the parser choose (by default) atomic units, which is the correct behaviour. Example:
acell 8 8 8 angstrom
ecut 8 Ry
tsmear 1000 K
or
acell 3*10 Bohr ecut 270 eV tsmear 0.01
The use of the atomic units is mandatory for other dimensioned input variables, like the tolerance on forces (toldff), parameters that define an ‘object’ (objaax, objaax, objbax, objatr, objbtr), and the initial velocity of atoms (vel if needed).
The initial atomic positions can be input in Bohr or Angstrom through xcart (possibly use the Angstrom unit), or even in reduced coordinates, through xred.
Flow information¶
Most of the variables can be used in the multidataset mode (see section 3.3), but those that must have a unique value throughout all the datasets are signaled with the indication NO_MULTI.
Some of the input variables, with characteristics INPUT_ONLY are only used by the parser, to initialize other input variables, but are not transmitted inside the code, beyond the parser. In particular, they are not echoed in the output file.
At variance, some internal variables, with characteristics INTERNAL_ONLY are documented in the help files, but are not accessible as input variables. The documentation is provided because such variables are sometimes mentioned in the output file.
Most of the input variables do not change while a run is performed. Some of them, by contrast, may evolve, like the atomic positions, the atomic velocities, the cell shape, and the occupation numbers. Their echo, after the run has proceeded, will of course differ from their input value. They are signaled by the indication EVOLVING.
Other information¶
DEVELOP refers to input variables that are not used in production runs, but have been introduced during development time, of a feature that is likely not finalized. For non ABINIT developers, it is strongly advised to skip them.
In addition to giving the input variables, the input file can be useful for another purpose: placing the word ” exit ” on the top line will cause the job to end smoothly on the very next iteration, if the chkexit input variable is nonzero. This functions because the program closes and reopens the input file on every iteration and checks the top line for the keyword “exit”. THE WORD MUST BE PLACED WITH SPACES (BLANKS) ON BOTH SIDES. Thus placing exit on the top line of the input file WHILE THE JOB IS ALREADY RUNNING will force the job to end smoothly on the very next iteration. On some machines, this does not work always (we do not know why…). Another possibility is offered: one can create a file named “abinit.exit” in the directory where the job was started. The code should also smoothly end. In both cases, the stop is not immediate. It can take a significant fraction (about 20% at most) of one SCF step to execute properly the instruction still needed.
3.4 The multidataset mode¶
Until now, we have assumed that the user wants to make computations corresponding to one set of data: for example, determination of the total energy for some geometry, with some set of plane waves and some set of kpoints.
It is often needed to redo the calculations for different values of some parameter, letting all the other things equal. As typical examples, we have convergence studies needed to determine which cutoff energy gives the needed accuracy. In other cases, one makes chains of calculations in order to compute the band structure: first a selfconsistent calculation of the density and potential, then the eigenenergy computation along different lines.
For that purpose, the multidataset mode has been implemented.
It allows the code to treat, in one run, different sets of data, and to chain them. The number of datasets to be treated is specified by the variable ndtset, while the indices of the datasets (by default 1, 2, 3, and so on) can be eventually provided by the array jdtset.
For each dataset to be treated, characterized by some index, each input variable will determined by the following rules (actually, it is easier to understand when one looks at examples, see below):

(1) ABINIT looks whether the variable name (e.g. ecut ), appended with the index of the dataset (e.g. jdtset=2), exists (e.g. “ecut2” ). It will take the data that follows this keyword, if it exists.

(2) If this modified variable name does not exist, it will look whether a metacharacter, a series or a doubleloop data set has been defined, see sections 3.4 or 3.5.

(3) If the variable name appended with the index of the dataset does not exist, and if there is no series nor doubleloop dataset for this keyword, it looks for an occurrence of the variable name without any index appended, and take the corresponding data. (This corresponds to the single dataset mode)

(4) If such occurrences do not exist, it takes the default value. (Also, similar to the single dataset mode)
 1st example. ndtset 2 acell 8 8 8 ecut1 10 ecut2 15
means that there are 2 datasets: a first in which
acell 8 8 8 ecut 10
has to be used, and a second in which
acell 8 8 8 ecut 15
has to be used.

2nd example
ndtset 2 jdtset 4 5
acell 8 8 8
acell5 10 10 10
ecut1 10
ecut2 15
ecut3 20
ecut4 25
ecut5 30
this means that there are still two datasets, but now characterized by the indices 4 and 5, so that the first run will use the generic “acell”, and “ecut4”:
acell 8 8 8 ecut 25
and the second run will use “acell5” and “ecut5”:
acell 10 10 10 ecut 30
Note that ecut1, ecut2 and ecut3 are not used.
3.5 Defining a series¶
Rule (2) is split in three parts: (2a), (2b) and (2c). Series relate with (2b):
(2b) If the variable name appended with the index of the dataset does not exist, the code looks whether a series has been defined for this keyword.
There are two kinds of series:

arithmetic series (constant increment between terms of the series)

geometric series (constant ratio between terms of the series)
The first term of the series is defined by the keyword appended with a colon (e.g. ecut: ), while the increment of an arithmetic series is defined by the keyword appended with a plus (e.g. ecut+ ), and the factor of a geometric series is defined by the keyword appended with a times (e.g. ecut* ).
If the index of the dataset is 1, the first term of the series is used, while for index N, the appropriate input data is obtained by considering the Nth term of the series.

3rd example
ndtset 6
ecut1 10
ecut2 15
ecut3 20
ecut4 25
ecut5 30
ecut6 35
is equivalent to
ndtset 6 ecut: 10 ecut+ 5
In both cases, there are six datasets, with increasing values of ecut.
3.6 Defining a double loop dataset¶
To define a double loop dataset, one has first to define the upper limit of two loop counters, thanks to the variable udtset. The inner loop will execute from 1 to udtset(2), and the outer loop will execute from 1 to udtset(1). Note that the largest value for udtset(1) is presently 999, while it is 9 for udtset(2) (so, only the last digit for the inner loop).
The value of ndtset must be coherent with udtset (it must equal the
product udtset(1) * udtset(2)
).
A dataset index is created by the concatenation of the outer loop index and
the inner loop index.
For example, if udtset(1) is 2 and udtset(2) is 4, the index will
assume the following values: 11, 12, 13, 14, 21, 22, 23, and 24
.
Independently of the use of udtset, rules (2a) and (2c) will be used to define the value of an input variable:
(2a) The question mark ” ? ” can be used as a metacharacter, replacing any
digit from 1 to 9, to define an index of a dataset.
For example, ecut? 1 means that the input value that follows it can be
used for ecut for the datasets 01, 11, 21, 31, 41, 51, 61, 71, 81, and 91
.
(2c) If the variable name appended with the index of the dataset does not
exist, the code looks whether a doubleloop series has been defined for this
keyword. Series can be defined for the inner loop index or the outer loop
index. Two signs will be appended to the variable name (instead of one in the
simple series case). One of these signs must be a question mark ” ? “,
again used as a metacharacter able to assume the values 1 to 9.
If it is found in the first of the two positions, it means that the series
does not care about the outer loop index (so the values generated are equal
for all outer loop index values). If it is found in the second of the two
positions, the series does not care about the inner loop index. The other sign
can be a colon, a plus or a times, as in the case of the series defined in
(2a), with the same meaning.
Rule (1) has precedence over them, they have precedence over rules (3) or (4), rule (2a) has precedence over rules (2b) or (2c) and the two latter cannot be used simultaneously for the same variable.

4th example
ndtset 6 udtset 2 3
acell1? 10 10 10
acell2? 15 15 15
ecut?: 5 ecut?+ 1
is equivalent to
ndtset 6 jdtset 11 12 13 21 22 23
acell11 10 10 10 ecut11 5
acell12 10 10 10 ecut12 6
acell13 10 10 10 ecut13 7
acell21 15 15 15 ecut21 5
acell22 15 15 15 ecut22 6
acell23 15 15 15 ecut23 7
Tip
More examples can be found in the directory ~abinit/tests/v1, cases 59 and later.
3.7 File names in the multidataset mode¶
The root names for input and output files (potential, density, wavefunctions and so on) will receive an appendix: _DS followed by the index of the dataset. See section 4.
The get variables can be used to chain the calculations.
Let us mention a few of them: getwfk, getwfq, getddk, get1wf, getden, getcell, getxred and getxcart.
 getwfk allows to take the output wavefunctions of a previous dataset and use them as input wavefunctions
 getwfq, getddk and get1wf do similar things for response function calculations
 getden does the same for the density; getcell does the same for acell and rprim
 getxred and getxcart do the same for the atomic positions, either in reduced coordinates, or in cartesian coordinates.
The different variables corresponding to each dataset are echoed using the same indexing convention as for the input step. For the last echo of the code variables, some output variables are also summarized, using the same conventions:
 etotal (total energy)
 fcart (cartesian forces)
 strten (the stress tensor).
4 The pseudopotential files and PAW atomic data files¶
Actually, no real understanding of these files is needed to run the code. The recommended pseudopotentials can be downloaded from the ABINIT Web site at https://www.abinit.org/psptables. Documentation is provided there as well as in the dedicated topic_PseudosPAW. Note that it is not possible to mix normconserving pseudopotentials and PAW atomic data sets in the same run. Also, every such file has been generated for a particular choice of the exchangecorrelation functional ixc. It is in principle incorrect to use a pseudopotential (or PAW data) with another exchangecorrelation functional than the one it has been generated for, but ABINIT will only send a warning.
For different other reasons, it might nevertheless be useful to be able to grasp some information from the file. For normconserving pseudopotentials different format are possible (labelled 1 to 8 presently). The associated internal variable is called pspcod. Information on the header of these pseudopotential files can be found in the abinit wiki at https://wiki.abinit.org/doku.php?id=developers:pseudos, that you should read now (do not pursue with the description of each format, though).
5 The different output files¶
Explanation of the output from the code
Output from the code goes to several places listed below.
5.1 The log file¶
The “log” file (this is the standard UNIX output file, and corresponds to Fortran unit number 06): a file which echoes the values of the input parameters and describes various steps of the calculation, typically in much more detail than is desired as a permanent record of the run. This log file is intended to be informative in case of an error or for a fuller description of the run. For a successful run the user will generally delete the log file afterwards. There are four types of exception messages: ERROR, BUG, WARNING and COMMENT messages.
ERROR and BUG messages cause the code to stop, immediately, or after a very small delay. An ERROR is attributed to the user, while a BUG is attributed to the developer.
A WARNING message indicates that something happened that is not as expected, but this something is not so important as to make the code stop. A COMMENT message gives some information to the user, concerning something unusual. None of them should appear when the run is completely normal.
After a run is completed, always have a look at the end of the log file, to see whether an ERROR or a BUG occurred.
Also, the code gives the number of WARNING or COMMENT it issued. It is advised to read at least the WARNING messages, during the first month of ABINIT use.
5.2 The main output file¶
The main output file is a formatted output file to be kept as the permanent record of the run.
Note that it is expected not to exist at the beginning of the run:
If a file with the same name as the one specified by the input variable output_file
(see also the default output file name description) already exists, the code
will generate, from the given one, another name, appended with .A. If
this new name already exists, it will try to append .B, and so on, until .Z.
Then, the code stops, and asks you to clean the directory.
The main output file starts with a heading:
 version number and specified platform
 copyright notice and distribution licence
 date
 echo of “files” file (except pseudopotential name)
Then, for each dataset, it reports the point symmetry group and Bravais lattice, and the expected memory needs. It echoes the input data, and report on checks of data consistency for each dataset.
5.3 More on the main output file¶
Then, for each dataset, the real computation is done, and the code will report on some initialisations, the SCF convergence, and the final analysis of results for this dataset. Each of these phases is now described in more details.
The code reports:
 The real and reciprocal space translation vectors ( Note: the definition of the reciprocal vector is such that R_i.G_j= \delta_{ij}).
 The volume of the unit cell.
 The ratio between linear dimension of the FFT box and the sphere of plane waves, called boxcut. It must be above 2 for exact treatment of convolutions by FFT. ngfft has been automatically chosen to give a boxcut value larger than 2, but not much larger, since more CPU time is needed for larger FFT grids.
 The code also mention that for the same FFT grid you might treat (slightly) larger ecut (so, with a rather small increase of CPU time).
 The heading for each pseudopotential which has been input.
 From the inwffil subroutine, a description of the wavefunction initialization (random number initialization or input from a disk file), that is, a report of the number of planewaves (npw) in the basis at each k point
 From the setup2 subroutine, the average number of planewaves over all k points is reported in two forms, arithmetic average and geometric average.
Until here, the output of a groundstate computation is identical to the one of a responsefunction calculation. See the respfn help file for the latter, especially section 6.2.
Next the code reports information for each SCF iteration:
 The iteration number.
 The (pseudo) total energy (Etot) in Hartree [This is not the total energy of the system, since the pseudopotential approximation has been made: a constant energy (in the frozencore approximation) should be added to the present pseudo total energy in order to obtain a total energy, that includes the contributions from the core electrons. Since only differences of total energy matter (except is extremely rare cases), one can work with this pseudo energy like if it were the true total energy, except that the missing constant depends on the pseudopotential that has been used. Thus one has to perform differences of pseudo energies between simulations that use the same pseudopotentials].
 The change in Etot since last iteration (deltaE).
 The maximum squared residual residm over all bands and k points (residm  the residual measures the quality of the wavefunction convergence)
 The squared residual of the potential in the SCF procedure (vres2).
 The maximum change in the gradients of Etot with respect to fractional coordinates (diffor, in Hartree).
 The rms value of the gradients of Etot with respect to fractional coordinates (maxfor, in Hartree).
The latter two are directly related to forces on each atom.  Then comes an assessment of the SCF convergence: the criterion for fulfillment of the SCF criterion (defined by toldfe, toldff, tolwfr or tolvrs) might be satisfied or not…
 Then the stresses are reported.
This ends the content of a fixed atomic position calculation.
Many such blocks can follow.
When the atomic positions have been eventually relaxed, according to the value of ntime, the code output more information:
 The squared residuals for each band are reported, k point by k point.
 Then the fractional or reduced coordinates are given,
 followed by the energy gradients,
 followed by the cartesian coordinates in Angstroms,
 followed by the cartesian forces in Hartree/Bohr and eV/Angstrom.
 Also are given the rms force ( frms ) and the maximum absolute value of any force component ( max ).
 Next are the length scales of the unit cell in Bohr and in Angstroms.
 Next are the eigenvalues of each band for each k point, in eV or Hartree or both depending on the choice of enunit.
NOTE that the average electrostatic potential of a periodically repeated cell is UNDEFINED.
In the present implementation, the average Hartree potential and local
potential are imposed to be zero, but not the average exchangecorrelation
potential. This definition gives some meaning to the absolute values of
eigenenergies, thanks to Janak’s theorem: they are derivatives of the total
energy with respect to occupation number. Indeed, the G=0 contributions of the
Hartree, local potential and ionion to the total energy is independent of the
occupation number in the present implementation. With this noticeable
exception, one should always work with differences in eigenenergies, as
well as differences between eigenenergies and the potential. For example,
the absolute eigenenergies of a bulk cell should not be used to try to predict
a work function. The latter quantity should be obtained in a supercell
geometry, by comparing the Fermi energy in a slab and the potential in the
vacuum in the same supercell.

Next are the minimum and maximum values for charge density, and next smaller or larger values (in order to see degeneracies).

Next are the total energy (Ha and eV) and its components:
 kinetic,
 Hartree,
 exchange and correlation (xc),
 Ewald (ionion energy),
 core correction to the local pseudopotential,
 local pseudopotential, and
 nonlocal pseudopotential. The sum of the KohnSham energies (termed “band energy”) is also given.

Next is the stress tensor, (1/ucvol) d(Etot)/d(strain(a,b)) for Etot=total energy per unit cell and a, b are x, y, or z components of strain. The stress tensor is given in cartesian coordinates in Hartree/Bohr 3 and GPa. The basics of the stress tensor are described in [Nielsen1985] and [Nielsen1985a].
Having finished all the calculations for the different datasets, the code echoes the parameters listed in the input file, using the latest values e.g. for xred, vel, and xcart, and supplement them with the values obtained for the total energy, the forces and stresses, as well as occupation numbers. The latter echoes are very convenient for a quick look at the result of calculation!
This is followed finally by the timing output: both “cpu” time and “wall clock” time as provided by calls within the code. The total cpu and wall clock times are reported first, in seconds, minutes, and hours for convenient checking at a glance. Next are the cpu and wall times for the principal timeconsuming subroutine calls, each of which is independent of the others. The sum of these times usually accounts for about 90% of the run time.
The main subroutines, for BIG jobs, are
 fourwf: the subroutine which performs the fast Fourier transform for the wavefunctions:
 fourdp: the subroutine which performs the fast Fourier transform related to density and potential
 rhohxc: computes the Hartree and exchangecorrelation energy and potential and sometimes
derivative of potential; only the XC timing is reported, excluding time connected to the FFTs:
xc:pot/=fourdp.
 nonlop: computes the matrix elements of the nonlocal pseudopotential: \langle GV_{nonlocal}C \rangle
 projbd: GramSchmidt orthogonalisation
In case of small jobs, other (initialisation) routines may take a larger share, and the sum of the times for the principal timeconsuming subroutine calls will not make 90% of the run time.
If the long printing option has been selected (prtvol=1), the code gives
much more information in the whole output file. These should be rather
selfexplanatory, usually. Some need more explanation.
In particular the cpu and wall times for major subroutines which are NOT
independent of each other; for example vtorho conducts the loop over k points
and calls practically everything else. In case of a ground state calculation,
at fixed atomic positions, these subroutines are:
 abinit: the main routine
 driver: select ground state or response calculations
 gstate: the driver of the ground state calculations
 scfcv_core: the SCF cycle driver
 vtorho: compute the density from the potential (it includes a loop over spins and kpoints)
 vtowfk: compute the wavefunctions at one particular kpoint (includes a non self consistent loop, and a loop over bands)
 cgwf: optimize one wavefunction in a fixed potential
 getghc: computes \langle GHC\rangle, that is, applies the Hamiltonian operator to an input vector.
5.4 The header¶
The wavefunction files, density files, and potential files are not plain text files (as they are quite big, and need some kind of compressed storage mode), but either binary files (from FORTRAN write statements), or netcdf files. See the input variable iomode. FORTRAN binary format is the default, and will be the focus of the following sections. Reading a netcdf file is actually much easier than reading a FORTRAN binary file, as netcdf files can be addressed by content.
In order to provide the relevant metadata about the run that generated such numerical data,
the wavefunction files, density files, and potential files (irrespective of the format)
contain standardized basic information. In case of FORTRAN,
such files begin with the same records, called the “header”.
This header is treated using the hdr_type Fortran data structure inside ABINIT.
There are dedicated routines inside ABINIT for initializing a header, updating it,
reading the header of an unformatted disk file, writing a header to an unformatted disk file,
echoing a header to a formatted disk file, cleaning a header data structure.
The header is made of 4 + ntypat unformatted records, obtained by the following Fortran90 instructions (format 9.0):
write(unit=header) codvsn,headform,fform
write(unit=header) bantot,date,intxc,ixc,natom,ngfft(1:3),&
nkpt,nspden,nspinor,nsppol,nsym,npsp,ntypat,occopt,pertcase,usepaw,&
ecut,ecutdg,ecutsm,ecut_eff,qptn(1:3),rprimd(1:3,1:3),stmbias,&
tphysel,tsmear,usewvl,nshiftk_orig,nshiftk,mband
write(unit=header) istwfk(1:nkpt),nband(1:nkpt*nsppol),&
npwarr(1:nkpt),so_psp(1:npsp),symafm(1:nsym),symrel(1:3,1:3,1:nsym),&
typat(1:natom),kpt(1:3,1:nkpt),occ(1:bantot),tnons(1:3,1:nsym),&
znucltypat(1:ntypat),wtk(1:nkpt)
write(unit=unit) residm,xred(1:3,1:natom),etotal,fermie,amu(1:ntypat)
write(unit=unit) kptopt,pawcpxocc,nelect,cellcharge,icoulomb,&
kptrlatt(3,3),kptrlatt_orig(3,3),shiftk_orig(3),shiftk(3)
do ipsp=1,npsp
! (npsp lines, 1 for each pseudo; npsp=ntypat, except if alchemical pseudoatoms)
write(unit=unit) title,znuclpsp,zionpsp,pspso,&
pspdat,pspcod,pspxc,lmn_size,md5_pseudos
enddo
!(in case of usepaw==1, there are some additional records)
if (usepaw==1)then
write(unit=unit)( pawrhoij(iatom)%nrhoijsel(1:nspden),iatom=1,natom), cplex, nspden
write(unit=unit)((pawrhoij(iatom)%rhoijselect(1: nrhoijsel(ispden),ispden),ispden=1,nspden),iatom=1,natom),&
((pawrhoij(iatom)%rhoijp (1:cplex*nrhoijsel(ispden),ispden),ispden=1,nspden),iatom=1,natom)
endif
where the type of the different variables is:
character*8 :: codvsn
integer :: headform,fform
integer :: bantot,cplex,date,icoulomb,intxc,ixc,kptopt,mband,natom,ngfft(3),&
nkpt,npsp,nshiftk_orig,nshiftk,nspden,nspinor,nsppol,nsym,ntypat,occopt,&
pawcpxocc,pertcase,usepaw,usewvl
double precision :: acell(3),cellcharge,ecut,ecutdg,ecutsm,ecut_eff,etotal,&
fermie,nelect,qptn(3),residm,rprimd(3,3),shiftk(3),shiftk_orig(3),&
stmbias,tphysel,tsmear
integer :: istwfk(nkpt),kptrlatt(3,3),kptrlatt_orig(3,3),nband(nkpt*nsppol),&
npwarr(nkpt),so_psp(npsp),symafm(nsym),symrel(3,3,nsym),typat(natom),&
nrhoijsel(nspden),rhoijselect(*,nspden)
double precision :: amu(ntypat),kpt(3,nkpt),occ(bantot),tnons(3,nsym),&
znucltypat(ntypat),wtk(nkpt),xred(3,natom)
character*132 :: title
character*32 :: md5_pseudos
double precision :: znuclpsp,zionpsp
integer :: pspso,pspdat,pspcod,pspxc,lmn_size
double precision :: rhoij(*,nspden)
NOTE: etotal is set to its true value only for density and potential files.
For other files, it is set to 1.0d20
NOTE: ecut_eff= ecut*(dilatmx)^2
NOTE: For all cases where occupation numbers are defined (that is, positive
iscf, and iscf=3), and for nonmetallic occupation numbers, the Fermi energy
is set to the highest occupied eigenenergy. This might not correspond to the
expected Fermi energy for a later nonselfconsistent calculation (e.g. the band structure)
The header might differ for different versions of ABINIT. One prev5.3 format is described below. Note however, that the current version of ABINIT should be able to read all the previous formats (not to write them), with the exception of wavefunction files for which the ecutsm value was nonzero (there has been a change of definition of the smearing function in v4.4).
The format for ABINIT versions 8.0 to 8.11 was:
write(unit=header) codvsn,headform,fform
write(unit=header) bantot,date,intxc,ixc,natom,ngfft(1:3),&
& nkpt,nspden,nspinor,nsppol,nsym,npsp,ntypat,occopt,pertcase,usepaw,&
& ecut,ecutdg,ecutsm,ecut_eff,qptn(1:3),rprimd(1:3,1:3),stmbias,tphysel,tsmear,usewvl
write(unit=header) istwfk(1:nkpt),nband(1:nkpt*nsppol),&
& npwarr(1:nkpt),so_psp(1:npsp),symafm(1:nsym),symrel(1:3,1:3,1:nsym),typat(1:natom),&
& kpt(1:3,1:nkpt),occ(1:bantot),tnons(1:3,1:nsym),znucltypat(1:ntypat),wtk(1:nkpt)
do ipsp=1,npsp
! (npsp lines, 1 for each pseudopotential; npsp=ntypat, except if alchemical pseudoatoms)
write(unit=unit) title,znuclpsp,zionpsp,pspso,pspdat,pspcod,pspxc,lmn_size
enddo
!(in case of usepaw==0, final record: residm, coordinates, total energy, Fermi energy)
write(unit=unit) residm,xred(1:3,1:natom),etotal,fermie
!(in case of usepaw==1, there are some additional records)
if (usepaw==1)then
write(unit=unit)( pawrhoij(iatom)%nrhoijsel(1:nspden),iatom=1,natom), cplex, nspden
write(unit=unit)((pawrhoij(iatom)%rhoijselect(1: nrhoijsel(ispden),ispden),ispden=1,nspden),iatom=1,natom),&
& ((pawrhoij(iatom)%rhoijp (1:cplex*nrhoijsel(ispden),ispden),ispden=1,nspden),iatom=1,natom)
endif
where the type of the different variables was:
character*6 :: codvsn
integer :: headform,fform
integer :: bantot,date,intxc,ixc,natom,ngfft(3),nkpt,npsp,
nspden,nspinor,nsppol,nsym,ntypat,occopt,pertcase,usepaw
integer :: usewvl, cplex, nspden
double precision :: acell(3),ecut,ecutdg,ecutsm,ecut_eff,qptn(3),rprimd(3,3),stmbias,tphysel,tsmear
integer :: istwfk(nkpt),nband(nkpt*nsppol),npwarr(nkpt),so_psp(npsp),&
& symafm(nsym),symrel(3,3,nsym),typat(natom),nrhoijsel(nspden),rhoijselect(*,nspden)
double precision :: kpt(3,nkpt),occ(bantot),tnons(3,nsym),znucltypat(ntypat),wtk(nkpt)
character*132 :: title
double precision :: znuclpsp,zionpsp
integer :: pspso,pspdat,pspcod,pspxc,lmax,lloc,mmax=integers
double precision :: residm,xred(3,natom),etotal,fermie,rhoij(*,nspden)
5.5 The density output file¶
This is an unformatted data file containing the electron density on the real space FFT grid. It consists of the header records followed by
do ispden=1,nspden
write(unit) (rhor(ir),ir=1,cplex*ngfft(1)*ngfft(2)*ngfft(3))
enddo
where rhor is the electron density in electrons/Bohr^3, and cplex is the number of complex components of the density (cplex=1 for GS calculations the density is real, and cplex=1 or 2 for RF). The input variable nspden describes the number of components of the density. The first component (the only one present when nspden=1) is always the total charge density. When nspden=2, the second component is the density associated with spinup electrons. When nspden=4, the second, third and fourth components correspond to the x, y and z projections of the local magnetization, in units of \hbar/2. Note that the meaning of the different components of the density differs for the density array (rhor) and for the different potential arrays (vxc…), see the next section.
To identify the points in real space which correspond with the index “ir”
above, consider the following.
The first array value (ir=1) corresponds with the first grid point which is at
the origin of the unit cell, (x=0, y=0, z=0).
The next grid point (ir=2) lies along the first primitive translation at the
next fft grid point, which is (1/ngfft(1))acell(1)rprim(mu,1).
This is 1/ngfft(1) of the way along the first primitive translation.
The rest of the values up to ir=ngfft(1) lie along this vector, at
(ir1)/ngfft(1) of the way along the first primitive translation. The
point at ir=ngfft(1)+1 lies at 1/ngfft(2) along the second primitive translation.
The next points up to ir=ngfft(1)+ngfft(1) are displaced in the
direction of the second primitive translation by 1/ngfft(2) and in the
first translation by (irngfft(1)1)/ngfft(1).
This pattern continues until ir=ngfft(1)ngfft(2).
The next point after that is displaced along the third primitive translation
by 1/ngfft(3), and so forth until ir varies all the way from 1 to
ngfft(1)ngfft(2)*ngfft(3). This last point is in the corner
diagonally opposite from the origin, or right alongside the origin if the
whole grid is viewed as being periodically repeated.
5.6 The potential files¶
Also unformatted files consisting of the header records and
do ispden=1,nspden
write(unit) (potential(ir),ir=1,cplex*ngfft(1)*ngfft(2)*ngfft(3))
enddo
where potential can be either the sum of the Hartree potential, exchange correlation and local pseudopotential (see prtpot), the Hartree potential (see prtvha), the Hartree+XC potential (see prtvhxc), the local pseudopotential (see prtvpsp) or the XC potential (see prtvxc), These are defined on the real space grid in Hartree energy units. The underlying grid is as described above. If nspden=2, the different components are the spinup potential and the spindown potential. In the case nspden=4, the components correspond to the upup potential, the downdown potential, the real part of the updown potential, and the imaginary part of the updown potential. Note that the Hartree potential is NOT spindependent, but in order to use the same format as for the other potential files, the spinindependent array is written twice, once for spinup and one for spindown.
5.7 The wavefunction output file¶
This is an unformatted data file containing the planewaves coefficients of all the wavefunctions, and different supplementary data.
The groundstate wf file consists of the header records, and data written with the following lines of FORTRAN:
bantot=0 < counts over all bands
index=0 < index for the wavefunction location
do isppol=1,nsppol
do ikpt=1,nkpt
write(unit) npw,nspinor,nband < for each k point
write(unit) kg(1:3,1:npw) < plane wave reduced coordinates
write(unit) eigen(1+bantot:nband+bantot), < eigenvalues for this k point
occ(1+bantot:nband+bantot) < occupation numbers for this k point
do iband=1,nband
write(unit) (cg(ii+index),ii=1,2*npw*nspinor) < wavefunction coefficients
enddo for a single band and k point
bantot=bantot+nband
index=index+2*npw*nspinor*nband
enddo
enddo
If the job ended without problem, a few supplementary lines are added, in order to give the history of atomic positions and corresponding forces. The integer nxfh gives the number of pairs (x,f) of positions and forces in reduced coordinates:
write(unit)nxfh
do ixfh=1,nxfh
write(unit) xred(1:3,1:natom,ixfh),dummy(1:3,1:4),&
& fred(1:3,1:natom,ixfh),dummy(1:3,1:4)
enddo
The dummy variables might contain, in the future, the description of the unit cell, and the stresses. The type of the different variables is:
integer :: kg,nband,npw,nspinor,nxfh
double precision :: cg,dummy,eigen,fred,occ,xred
The responsefunction wf file consists of the header records, and data written with the following lines of FORTRAN:
bantot=0 < counts over all bands
do isppol=1,nsppol
do ikpt=1,nkpt
write(unit) npw,nspinor,nband < for each k point
write(unit) kg(1:3,1:npw) < plane wave reduced coordinates
do iband=1,nband
write(unit) (eigen(jband+(iband1)*nband+bantot),jband=1,2*nband) < column of eigenvalue matrix
write(unit) (cg(ii+index),ii=1,2*npw*nspinor) < wavefunction coefficients
enddo for a single band and k point
bantot=bantot+nband
enddo
enddo
Note that there is an alternative format (_KSS) for the output of the wavefunction coefficients, activated by a nonzero value of nbandkss.
5.8 Other output files¶
There are many other output files, optionally written, all formatted files at present. Their use is usually governed by a specific input variable. Please consult the description of this input variable, in order to have more information on such files:
 prtdos to print a file with the electronic DensityOfStates
 prteig to print a file with the list of k points and eigenenergies
 prtgeo to print a file with a geometrical analysis (bond lengths and bond angles), that also contains an XMOL section
 prt1dm to print a onedimensional projection of potential and density, for the three axes.
5.9 Control of output in the parallel case¶
For massively parallel runs, one cannot afford to have some of the output files that are usually created. Explicitly, the log file and also the status file become problematic. By default, with N=2 processors or less than N=2 processors, they are created, but beyond N=2 processors, they are deactivated except for the main log file (master processor).
This default behaviour can be changed as follows. If a file named “_NOLOG” exists in the current directory, then no log file and no status file will be created, even with less than N=2 processors. By contrast, if a file “_LOG” exists in the current directory, then a log file and the status files will be created, even with more than N=2 processors. Alternatively, if a file named “_MAINLOG” exists and there are less than N=2 processors, only the master processor writes the log and status files (this mimic the default behavior when using more than N=2 processors but with less than N=2 processors)
In ABINITv7, N was set at N=100. However, with ABINITv8, N has been switched to 2. It can be changed “by hand”, though: modify NPROC_NO_EXTRA_LOG in src/10_defs/defs_basis.F90 and recompile. See 44_abitypes_defs/m_dtfil.F90 for more explanation.
6 Numerical quality of the calculations¶
The following section describes various parameters which affect convergence and the numerical quality of calculations.
The list of these input parameters is
 (1) ecut
 (2) toldfe, toldff, tolwfr, and tolvrs, as well as nstep
 (3) nkpt
 (4) ngfft
 (5) tolmxf, as well as amu, dtion, vis, ntime
 (6) acell and rprim
The technical design of the pseudopotential also affects the quality of the results.
(1) The first issue regarding convergence is the number of planewaves in the basis for a given set of atoms. Some atoms (notably those in the first row or first transition series row) have relatively deep pseudopotentials which require many planewaves for convergence. In contrast are atoms like Si for which fewer planewaves are needed. A typical value of ecut for silicon with a normconserving pseudopotential might be 10…15 Hartree for quite good convergence, while the value for oxygen might be 35…40 Hartree or more depending on the convergence desired and the design of the pseudopotential. With PAW, the ecut to be used is usually smaller, e.g. 17 Hartree for oxygen. The pseudodojo http://www.pseudodojo.org/ provides hints with the available pseudopotentials.
NOTE: It is necessary in every new problem to TEST the convergence, by RAISING ecut for a given calculation, until the results being computed are constant to within some tolerance. This is up to the user and is very important. For a given acell and rprim, ecut is the parameter which controls the number of planewaves. Of course if rprim or acell is varied then the number of planewaves will also change.
Let us reiterate that extremely careful pseudopotential design can optimize the convergence of e.g. the total energy within some range of planewave number or ecut. It is appropriate to attempt to optimize this convergence, especially for difficult atoms like oxygen or copper, as long as one does not significantly compromise the quality or transferability of the pseudopotential.
For information on optimizing the convergence of pseudopotentials, see [Rappe1990]. For the generation of ONCVPSP pseudopotentials, see [Hamann2013]. For the pseudodojo, see [Vansetten2018]. For the JTH table of PAW atomic data, see [Jollet2014].
(2) In addition to achieving convergence in the number of planewaves in the basis, one must ensure that the SCF iterations which solve the electronic structure for a given set of atomic coordinates are also converged. This convergence is controlled by the parameters toldfe, toldff, tolwfr, and tolvrs, as well as the parameter nstep. One of the “tolerance” parameters must be chosen, and, when the required level of tolerance is fulfilled, the SCF cycles will stop. The nstep variable also controls convergence in preconditioned conjugate gradient iterations by forcing the calculation to stop whenever the number of such iterations exceeds nstep. Usually one wants nstep to be set larger than needed to reach a given tolerance, or else one wants to restart insufficiently converged calculations until the required tolerance is reached.
Note that, if the gap in the system closes (e.g. due to defect formation or if the system is metallic in the first place), the presently coded algorithm will be slower to converge than for insulating materials. Convergence trouble during iterations usually signals closure of the gap. The code will suggest to treat at least one unoccupied state (or band) in order to be able to monitor such a closure.
(3) For self consistent calculations (iscf positive) it is important to test the adequacy of the k point integration. If symmetry is used then one usually tests a set of “special point” grids. Otherwise one tests the addition of more and more k points, presumably on uniform grids, to ensure that a sufficient number has been included for good k point integration. The parameter nkpt indicates how many k points are being used, and their coordinates are given by kpt and kptnrm, described above. The weight given to each k point is provided by input variable wtk. Systematic tests of k point integration are much more difficult than tests of the adequacy of the number of planewaves. The difficulty I refer to is simply the lack of a very systematic method for generating k point grids for tests.
(4) It is possible to run calculations for which the fft box is not quite large enough to avoid aliasing error in fft convolutions. An aliasing error, or a Fourier filter approximation, is occurring when the output variable ” boxcut ” is less than 2. boxcut is the smallest ratio of the fft box side to the planewave basis sphere diameter. If this ratio is 2 or larger then e.g. the calculation of the Hartree potential from the charge density is done without approximation.
NOTE: the values of ngfft(1:3) are chosen automatically by the code to give boxcut > 2, if ngfft has not been set by hand. At ratios smaller than 2, certain of the highest Fourier components are corrupted in the convolution. If the basis is nearly complete, this Fourier filter can be an excellent approximation. In this case values of boxcut can be as small as about 1.5 without incurring significant error. For a given ecut, acell, and rprim, one should run tests for which ngfft is large enough to give boxcut >= 2, and then one may try smaller values of ngfft if the results are not significantly altered. See the descriptions of these variables above.
(5) If you are running calculations to relax or equilibrate structures, i.e.
with ionmov=1 and possibly vis>0, then the quality of your molecular
dynamics or relaxation will be affected by the parameters amu, dtion,
vis, ntime, tolmxf. Clearly if you want a relaxed structure you
must either run long enough or make repeated runs until the largest force in
the problem (output as fmax) is smaller than what you will tolerate (see
tolmxf).
If dtion is too large for the given values of masses (amu) and
viscosity (vis) then the molecular dynamics will be unstable. If dtion
is too small, then the molecular dynamics will move inefficiently slowly. A
consensus exists in the community that forces larger than about 0.1
eV/Angstrom are really too large to consider the relaxation to be converged.
It is best for the user to get experience with this in his/her own
application.
The option ionmov=2, 3 or 7 are also available This uses the Broyden
(BFGS) scheme for structural optimization and is much more efficient than
viscous damping for structural relaxation.
(6) If you are running supercell calculations (i.e. an isolated atom or molecule in a big box, or a defect in a solid, or a slab calculation) you must check the convergence of your calculation with respect to the supercell and system size.
 For an isolated molecule in a big box: increase concurrently the three dimensions of your supercell (acell), and check the convergence of your physical property.
 For a defect in a solid: your supercell must be a multiple of the primitive cell of the bulk solid, so you have less freedom. Still, be sure that your supercell is large enough for your properties of interest to be accurate at the level you want it to be.
 For a slab calculation: you must increase the vacuum in the cell, but also the thickness of your slab systematically…