The Input File (CTRL)
This guide aims to detail the structure and use of the main input file (called the ctrl file) and related topics. As this guide explains, data is organized by categories, and the contents of each category is documented. A more careful description of the input file’s syntax can be found in this manual.
Table of Contents
 1. Input File Structure
 2. Help with finding tokens
 3. The EXPRESS category
 4. Input File Categories
1. Input File Structure
Introduction
Here is a sample input file for the compound Bi_{2}Te_{3} written for the lmf code.
categories tokens
VERS LM:7 FP:7
HAM AUTOBAS[PNU=1 LOC=1 LMTO=3 MTO=1 GW=0]
GMAX=8.1
ITER MIX=B2,b=.3 NIT=10 CONVC=1e5
BZ NKABC=3 METAL=5 N=2 W=.01
STRUC
NSPEC=2 NBAS=5 NL=4
ALAT=4.7825489
PLAT= 1 0 4.0154392
0.5 0.8660254 4.0154392
0.5 0.8660254 4.0154392
SPEC
ATOM=Te Z= 52 R= 2.870279
ATOM=Bi Z= 83 R= 2.856141
SITE
ATOM=Te POS= 0.0000000 0.0000000 0.0000000
ATOM=Te POS= 0.5000000 0.8660254 1.4616199
ATOM=Te POS= 0.5000000 0.8660254 1.4616199
ATOM=Bi POS= 0.5000000 0.8660254 0.8030878
ATOM=Bi POS= 0.5000000 0.8660254 0.8030878
Each element of data follows a token. The token tells the reader what the data signifies.
Each token belongs to a category. VERS, ITER, BZ, STRUC, SPEC, SITE are categories that organize the input by topic. Any text that begins in the first column is a category.
The full identifier (tag) consists of a sequence of branches, usually trunk and branch e.g. BZ_METAL. The leading component (trunk) is the category; the last is the token, which points to actual data. Thus the category groups tags into themes, the token identifies a particular type of data within the theme. Sometimes a tag has three branches, e.g. HAM_AUTOBAS_LOC.
Note: input files described here (ctrl.ext) can be automatically constructed from init files using the blm utility. init files and ctrl files are structured with categories and tokens in essentially the same way. For a another description of categories and tokens, see the init file documentation.
Tags, Categories and Tokens
The input file offers a very flexible free format: tags identify data to be read by a program, e.g.
W=.01
reads a number (.01) from token W=. In this case W= belongs to the BZ category, so the full tag name is BZ_W.
A category holds information for a family of data, for example BZ contains parameters associated with Brillouin zone integration. The entire input system has at present a grand total of 17 categories, though any one program uses only a subset of them.
Consider the Brillouin zone integration category. You plan to carry out the BZ integration using the MethfesselPaxton sampling method. MP integration has two parameters: polynomial order n and gaussian width w. Two tags are used to identify them: BZ_N and BZ_W; they are usually expressed in the input file as follows:
BZ N=2 W=.01
This format style is the most commonly used because it is clean and easy to read; but it conceals the tree structure a little. The same data can equally be written:
BZ[ N=2 W=.01]
Now the tree structure is apparent: [..] delimits the scope of tag BZ.
Any tag that starts in the first column is a category, so any nonwhite character appearing in the first column automatically starts a new category, and also terminates any prior category. N= and W= mark tokens BZ_N and BZ_W.
Apart from the special use of the first column to identify categories, data is largely freeformat, though there are a few mild exceptions. Thus:
BZ N=2
W=.01
BZ W=.01 N=2
BZ[ W=.01 N=2]
all represent the same information.
Note: if two categories appear in an input file, only the first is used. Subsequent categories are ignored. Generally, only the first tag is used when more than one appears within a given scope.
Usually the tag tree has only two levels (category and token) but not always. For example, data associated with atomic sites must be supplied for each site. In this case the tree has three levels, e.g. SITE_ATOM_POS. Site data is typically represented in a format along the following lines:
SITE ATOM=Ga POS= 0 0 0 RELAX=T
ATOM=As POS= .25 .25 .25
ATOM=...
...
END
The scope of SITE starts at “SITE” and terminates just before “END”. There will be multiple instances of the SITE_ATOM tag, one for each site. The scope of the first instance begins with the first occurrence of ATOM and terminates just before the second:
ATOM=Ga POS= 0 0 0 RELAX=T
And the scope of the second SITE_ATOM is
ATOM=As POS= .25 .25 .25
Note that ATOM simultaneously acts like a token pointing to data (e.g. Ga) and as a tag holding tokens within it, in this case SITE_ATOM_POS and (for the first site) SITE_ATOM_RELAX.
Some tags are required; others are optional; still others (in fact most) may not be used at all by a particular program. If a code needs site data, SITE_ATOM_POS is required, but SITE_ATOM_RELAX is probably optional, or not read at all.
Note: this manual contains a more careful description of the input file’s syntax.
Preprocessor
Input lines are passed through a preprocessor, which provides a wide flexibility in how input files are structured. The preprocessor has many features in common with a programming language, including the ability to declare and assign variables, evaluate algebraic expressions; and it has constructs for branching and looping, to make possible multiple or conditional reading of input lines.
For example, supposing through a prior preprocessor instruction you have declared a variable range, and it has been assigned the value 3. This line in the input file:
RMAX={range+1/4}
is turned in to:
RMAX=3.25
The preprocessor treats text inside brackets {…} as an expression (usually an algebraic expression), which is evaluated and rendered back as an ASCII string. See this annotated lmf output for an example.
The preprocessor’s programming language makes it possible for a single file to serve as input for many materials systems in the manner of a database; or as documentation. Also you can easily vary input conditions in a parameteric fashion.
Other files besides ctrl.ext are first parsed by the preprocessor — files for site positions, Euler angles for noncollinear magnetism are read through the preprocessor, among others.
2. Help with finding tokens
 Seeing the effect of the preprocessor
 The preprocessor can act in nontrivial ways. To see the effect of the preprocessor, use the
showp
commandline option.
See this annotated output for an example.  Finding what tags the parser seeks
 It is often the case that you want to input some information but don’t know the name of the tag you need. Try searching this page for a keyword.
 You can list each tag a particular tool reads, together with a synopsis of its function, by adding
input
to the commandline.
Search for keywords in the text to find what you need.
Take for an example:
lmchk input
This switch tells the parser not to try and read anything, but print out information about what it would try to read. Several useful bits of information are given, including a brief description of each tag in the following format. A snippet of the output is reproduced below:
Tag Input cast (size,min)

IO_VERBOS opt i4v 5, 1 default = 35
Verbosity stack for printout.
May also be set from the commandline: pr#1[,#2]
IO_IACTIV opt i4 1, 1 default = 0
Turn on interactive mode.
May also be controlled from the commandline: iactiv or iactiv=no
...
STRUC_FILE opt chr 1, 0
(Not used if data read from EXPRESS_file)
Name of site file containing basis and lattice information.
Read NBAS, PLAT, and optionally ALAT from site file, if specified.
Otherwise, they are read from the ctrl file.
...
STRUC_PLAT reqd r8v 9, 9
Primitive lattice vectors, in units of alat
...
SPEC_ATOM_LMX opt i4 1, 1 (default depends on prior input)
lcutoff for basis
...
SITE_ATOM_POS reqd* r8v 3, 1
Atom coordinates, in units of alat
 If preceding token is not parsed, attempt to read the following:
SITE_ATOM_XPOS reqd r8v 3, 1
Atom coordinates, as (fractional) multiples of the lattice vectors
The table tells you IO_VERBOS and IO_IACTIV are optional tags; default values are 35 and 0, respectively. A single integer will be read from the latter tag, and between one and five integers will be read from IO_VERBOS.
There is a brief synopsis explaining the functions of each. For these particular cases, the output gives alternative means to perform equivalent functions through commandline switches.
STRUC_FILE=fname is optional. Here fname is a character string: it should be the site file name fname.ext from which lattice information is read. If you do use this tag, other tags in the STRUC category (NBAS, PLAT, ALAT) may be omitted. Otherwise, STRUC_PLAT is required input; the parser requires 9 numbers.
The synopsis also tells you that you can specify the same information using EXPRESS_file=fname (see EXPRESS category below).
SPEC_ATOM_LMX is optional input whose default value depends on other input (in this case, atomic number).
SITE_ATOM_POS is required input in the sense that you must supply either it or SITE_ATOM_XPOS. The * in reqd* the information in SITE_ATOM_POS can be supplied by an alternate tag – SITE_ATOM_XPOS in this case.
Note: if site data is given through a site file, all the other tags in the SITE category will be ignored.
The cast (real, integer, character) of each tag is indicated, and also how many numbers are to be read. Sometimes tags will look for more than one number, but allow you to supply fewer. For example, BZ_NKABC in the snippet below looks for three numbers to determine the kmesh, which are the number of divisions only each of the reciprocal lattice vectors. If you supply only one number it is copied to elements 2 and 3.
BZ_NKABC reqd i4v 3, 1
(Not used if data read from EXPRESS_nkabc)
No. qp along each of 3 lattice vectors.
Supply one number for all vectors or a separate number for each vector.
 Commandline options
help
performs a similar function for the command line arguments: it prints out a brief summary of arguments effective in the executable you are using. A more complete description of generalpurpose command line options can be found on this page.
See this annotated lmfa output for an example. Displaying tags read by the parser
 To see what is actually read by a particular tool, run your tool with
show=2
orshow
.
See the annotated lmf output for an example.
These special modes are summarized here.
3. The EXPRESS category
Section 3 provides some description of the input and purpose of tags in each category.
There is one special category, EXPRESS, whose purpose is to simplify and streamline input files. Tags in EXPRESS are effectively aliases for tags in other categories, e.g. reading EXPRESS_gmax reads the same input as HAM_GMAX.
If you put a tag into EXPRESS, it will be read there and any tag appearing in its usual location will be ignored. Thus including GMAX in HAM would have no effect if gmax is present in EXPRESS.
EXPRESS collects the most commonly used tags in one place. There is usually a onetoone correspondence between the tag in EXPRESS and its usual location. The sole exception to this is EXPRESS_file, which performs the same function as the pair of tags, STRUC_FILE and SITE_FILE. Thus in using EXPRESS_file all structural data is supplied through the site file.
4. Input File Categories
This section details the various categories and tokens used in the input file.
Preliminaries
Note: The tables below list the input systems’ tokens and their function. Tables are organized by category.
 The Arguments column refers to the cast belonging to the token (“l”, “i”, “r”, and “c” refer to logical, integer, floatingpoint and character data, respectively)
 The Program column indicates which programs the token is specific to, if any
 The Optional column indicates whether the token is optional (Y) or required (N)
 The Default column indicates the default value, if any
 The Explanation column describes the token’s function.
BZ
Category BZ holds information concerning the numerical integration of quantities such as energy bands over the Brillouin Zone (BZ). The LMTO programs permit both sampling and tetrahedron integration methods. Both are described in bzintegration, and the relative merits of the two different methods are discussed. As implemented both methods use a uniform, regularly spaced mesh of kpoints, which divides the BZ into microcells as described here. Normally you specify this mesh by the number of divisions of each of the three primitive reciprocal lattice vectors (which are the inverse, transpose of the lattice vectors PLAT); NKABC below.
These tokens are read by programs that make hamiltonians in periodic crystals (lmf,lm,lmgf,lmpg,tbe). Some tokens apply only to codes that make energy bands, (lmf,lm,tbe).
Token  Arguments  Program  Optional  Default  Explanation 

GETQP  l  Y  F  Read list of kpoints from disk file qpts.ext. This is a special mode, and you normally would let the program choose its own mesh by specifying the number of divisions (see NKABC). If token is not parsed, the program will attempt to parse NKABC.  
NKABC  1 to 3 i  N  The number of divisions in the three directions of the reciprocal lattice vectors. kpoints are generated along a uniform mesh on each of these axes. (This is the optimal general purpose quadrature for periodic functions as it integrates the largest number of sine and cosine functions exactly for a specified number of points.) The parser will attempt to read three integers. If only one number is read, the missing second and third entries assume the value of the first. Information from NKABC, together with BZJOB below, contains specifications equivalent to the widely used “Monkhorst Pack” scheme. But it is more transparent and easier to understand. The number of kpoints in the full BZ is the product of these numbers; the number of irreducible kpoints may be reduced by symmetry operations.  
PUTQP  l  Y  F  If T, write out the list of irreducible kpoints to file qpts, and the weights for tetrahedron integration if available.  
BZJOB  1 to 3 i  Y  0  Controls the centering of the kpoints in the BZ: 0: the mesh is centered so that one point lies at the origin. 1: points symmetrically straddle the origin. Three numbers are supplied, corresponding to each of the three primitive reciprocal lattice vectors. As with NKABC if only one number is read the missing second and third entries assume the value of the first.  
METAL  i  lmf, lm, tbe  Y  5  Specifies how the weights are generated for Brillouin zone integration. For a detailed description, see this page. The METAL token accepts the following: 0. System assumed to be an insulator; weights determined a priori. 1. Eigenvectors are written to disk, in which case the integration for the charge density can be deferred until all the bands are obtained. 2. Integration weights are read from file wkp.ext, which will have been generated in a prior band pass. If wkp.ext is unavailable, the program will temporarily switch to METAL=3. 3. Two band passes are made; the first generates only eigenvalues to determine E_{F}. It is slower than METAL=2, but it is more stable which can be important in difficult cases. 4. Three distinct Fermi levels are assumed and weights generated for each. After E_{F} is determined, the actual weights are calculated by quadratic interpolation through the three points. 5. Like METAL=3 in which two passes are made but eigenvectors are kept in memory, so there is no additional overhead in making the first pass. This is the recommended mode for lmf unless you are working with a large system and need to conserve memory. The ASA implements METAL=0,1,2; the FP codes METAL=0,2,3,4,5. 
TETRA  i  lmf,lm,tbe  Y  1  Selects BZ integration method. 0: MethfesselPaxton sampling integration. Tokens NPTS, N, W, EF0, DELEF, DOS (see below) are relevant to this integration scheme. 1: tetrahedron integration, with Bloechl weights 
N  i  lmf,lm,tbe  Y  0  Polynomial order for MP sampling integration. (Not used with tetrahedron integration or for insulators). 0: integration uses standard gaussian method. >0: integration uses generalized gaussian functions, i.e. polynomial of order N × gaussian to generate integration weights. −1: use the Fermi function rather than gaussians to broaden the δfunction. This generates the actual electron (fermi) distribution for a finite temperature. Add 100: by default, if a gap is found separating occupied and unoccupied states, the program will treat the system as an insulator, even when METAL>0. To suppress this, add 100 to N (use −101 for Fermi distribution). 
W  r  lmf,lm,tbe  Y  5e3  Case BZ_N≥0 : broadening (Gaussian width) for Gaussian sampling integration (Ry). Case BZ_N<0 : k_{B}T (Ry) where k_{B} is the Boltzmann constant and T the temperature. W is not used for insulators or when using tetrahedron integration. 
EF0  r  lmf,lm,tbe  Y  0  Initial guess at Fermi energy. Used when TETRA=0, or when BZ_METAL=4 (which does not use the tetrahedron method for the density). 
DELEF  r  lmf,lm,tbe  Y  0.05  Initial uncertainty in Fermi level for sampling integration. Used when TETRA=0, or when BZ_METAL=4 (which does not use the tetrahedron method for the density). As the system approaches selfconsistency this window is reduced. 
ZBAK  r  lmf,lm  Y  0  Homogeneous background charge 
SAVDOS  i  lmf,lm,tbe  Y  0  0: does not save dos on disk. 1: writes the total density of states on NPTS mesh points to disk file dos.ext. 2: Write weights to disk for partial DOS (does not work for lmf; in the ASA this occurs automatically). 4: Same as (2), but write weights mresolved (ASA). Notes: 1. SAVDOS>0 uses DOS and NPTS tags also. 2. You may also cause lm or lmf to generate mresolved dos the from commandline (see pdos). 
DOS  2 r  Y  1,0  Energy window over which DOS accumulated (Ry). Needed either for sampling integration or if SAVDOS>0.  
NPTS  i  Y  1001  Number of points in the densityofstates mesh used in conjunction with sampling integration. Needed either for sampling integration or if SAVDOS>0.  
EFMAX  r  lmf,lm,tbe  Y  2  Only eigenvectors whose eigenvalues are less than EFMAX are computed; this improves execution efficiency. 
NEVMX  i  lmf,lm,tbe  Y  0  >0 : Find at most NEVMX eigenvectors. =0 : program uses internal default. <0 : no eigenvectors are generated (and correspondingly, nothing associated with eigenvectors such as density). Caution: if you want to look at partial DOS well above the Fermi level (which usually comes out around 0), you must set EFMAX and NEVMX high enough to encompass the range of interest. 
ZVAL  r  Y  all LDA  Number of electrons to accumulate in BZ integration. Normally zval is computed by the program.  
NOINV  l  lmf,lm,tbe  Y  F  Suppress the automatic addition of the inversion to the list of point group operations. Usually the inversion symmetry can be included in the determination of the irreducible part of the BZ because of time reversal symmetry. There may be cases where this symmetry is broken: e.g. when spinorbit coupling is included or when the (beyond LDA) selfenergy breaks timereversal symmetry. In most cases, the program will automatically disable this addition in cases that it knows the symmetry is broken. 
FSMOM  2 r  lmf,lm  Y  0 0  Set the global magnetic moment (collinear magnetic case). In the fixedspin moment method, a spindependent potential shift Beff is added to constrain the total magnetic moment to value assigned by FSMOM. No constraint is imposed if this value is zero (the default). Optional second argument #2 supplies an initial Beff. It is applied whether or not the first argument #1 is 0. If #1 ≠ 0, Beff is made consistent with it. 
DMATK  l  lmf,lmgf  Y  F  Calculate the density matrix. Implementation still not ready. 
INVIT  l  lmf,lm  Y  T  Generate eigenvectors by inverse iteration (this is the default). It is more efficient than the QL method, but occasionally fails to find all the vectors. When this happens, the program stops with the message: DIAGNO: tinvit cannot find all evecs If you encounter this message set INVIT=F. 
EMESH  r  lmgf,lmpg  Y  10,0,1,…  Parameters defining contour integration for Green’s function methods. See also the GF documentation. Element: 1. number of energy points n. 2. contour type: 0: Uniform mesh of nz points: Real part of z between emin and emax 1: Same as 0, but reverse sign of Im z 10: elliptical contour 11: same as 10, but reverse sign of Im z 100s digit used for special modifications Add 100 for nonequil part using Im(z)=delne Add 200 for nonequil part using Im(z)=del00 Add 300 for mixed elliptical contour + real axis to find fermi level Add 1000 to set nonequil part only. 3. lower bound for energy contour emin (on the real axis). 4. upper bound for energy contour emax, e.g. Fermi level (on the real axis). 5. (elliptical contour) eccentricity: ranges between 0 (circle) and 1 (line) (uniform contour) Im z. 6. (elliptical contour) bunching parameter eps : ranges between 0 (distributed symmetrically) and 1 (bunched toward emax) (uniform contour) not used. 7. (nonequilibrium GF, lmpg) nzne = number of points on nonequilibrium contour. 8. (nonequilibrium GF, lmpg) vne = difference in fermi energies of right and left leads. 9. (nonequilibrium GF, lmpg) delne = Im part of E for nonequilibrium contour. 10 (nonequilibrium GF, lmpg) substitutes for delne when making the surface selfenergy. 
MULL  i  tbe  Y  0  Mulliken population analysis. Mulliken population analysis is also implemented in lmf, but you specify the analysis with a commandline argument. 
CONST
This category enables users to declare variables in algebraic expressions. The syntax is a string of declarations inside the category, e.g:
CONST a=10.69 nspec=4+2
Variables declared this way are similar to, but distinct from variables declared for the preprocessor, such as
% const nbas=5
In the latter case the preprocessor makes a pass, and may use expressions involving variables declared by e.g. “% const nbas=5” to alter the structure of the input file.
Variables declared for use by the preprocessor lose their definition after the preprocessor completes.
The following code segment illustrates both types:
% const nbas=5
CONST a=10.69 nspec=4
STRUC ALAT=a NSPEC=nspec NBAS={nbas}
After the preprocessor compiles, the input file appears as:
CONST a=10.69 nspec=4
STRUC ALAT=a NSPEC=nspec NBAS=5
When the CONST category is read (it is read before other categories), variables a and nspec are defined and used in the SPEC category.
DYN
Contains parameters for molecular statics and dynamics.
Token  Arguments  Program  Optional  Default  Explanation 

NIT  i  lmf, lmmc, tbe  Y  maximum number of relaxation steps (molecular statics).  
SSTAT[…]  lm, lmgf  Y  (noncollinear magnetism) parameters specifying how spin statics (rotation of quantization axes to minimze energy) is carried out.  
SSTAT_MODE  i  lm, lmgf  N  0  0: no spin statics or dynamics. 1: LandauGilbert spin dynamics. 1: spin statics: quantization axis determined by making output density matrix diagonal. 2: spin statics: size and direction of relaxation determined from spin torque. Add 10 to mix angles independently of P,Q (Euler angles are mixed with prior iterations to accelerate convergence). Add 1000 to mix Euler angles independently of P,Q. 
SSTAT_SCALE  i  lm, lmgf  N  0  (used with mode=2) scale factor amplifying magnetic forces. 
SSTAT_MAXT  i  lm, lmgf  N  0  maximum allowed change in angle. 
SSTAT_TAU  i  lm, lmgf  N  0  (used with mode=1) time step. 
SSTAT_ETOL  i  lm, lmgf  N  0  (used with mode=1) Set tau=0 this iter if etotehf>ETOL. 
MSTAT[…]  lmf, lmmc, tbe  Y  (molecular statics) parameters specifiying how site positions are relaxed given the internuclear forces.  
MSTAT_MODE  i  lmf, lmmc, tbe  N  0  0: no relaxation. 4: relax with conjugate gradients algorithm (not generally recommended). 5: relax with FletcherPowell alogirithm. Find minimum along a line; a new line is chosen. The Hessian matrix is updated only at the start of a new line minimization. FletcherPowell is more stable but usually less efficient then Broyden. 6: relax with Broyden algorithm. This is essentially a NewtonRaphson algorithm, where Hessian matrix and direction of descent are updated each iteration. 
MSTAT_HESS  l  lmf, lmmc, tbe  N  T  T: Read hessian matrix from file, if it exists. F: assume initial hessian is the unit matrix. 
MSTAT_XTOL  r  lmf, lmmc, tbe  Y  1e3  Convergence criterion for change in atomic displacements. >0: criterion satisfied when xtol > net shift (shifts summed over all sites). <0: criterion satisfied when xtol > max shift of any site. 0: Do not use this criterion to check convergence. Note: When molecular statics are performed, either GTOL or XTOL must be specified. Both may be specified. 
MSTAT_GTOL  r  lmf,lmmc,tbe  Y  0  Convergence criterion for tolerance in forces. >0: criterion satisfied when gtol > “net” force (forces summed over all sites). <0: criterion satisfied when xtol > max absolute force at any site. 0: Do not use this criterion to check convergence. Note: When molecular statics are performed, either GTOL or XTOL must be specified. Both may be specified. 
MSTAT_STEP  r  lmf, lmmc, tbe  Y  0.015  Initial (and maximum) step length. 
MSTAT_NKILL  i  lmf, lmmc, tbe  Y  0  Remove hessian after NKILL iterations. Never remove Hessian if NKILL=0 
MSTAT_PDEF=  r  lmf, lmmc, tbe  Y  0 0 0 …  Lattice deformation modes (not documented). 
MD[…]  lmmc, tbe  Y  Parameters for molecular dynamics.  
MD_MODE  i  lmmc  N  0  0: no MD 1: NVE 2: NVT 3: NPT 
MD_TSTEP  r  lmmc  Y  20.671  Time step (a.u.) NB: 1 fs = 20.67098 a.u. 
MD_TEMP  r  lmmc  Y  0.00189999  Temperature (a.u.) NB: 1 deg K = 6.3333e6 a.u. 
MD_TAUP  r  lmmc  Y  206.71  Thermostat relaxation time (a.u.) 
MD_TIME  r  lmmc  N  20671000  Total MD time (a.u.) 
MD_TAUB  r  lmmc  Y  2067.1  Barostat relaxation time (a.u.) 
EWALD
Category EWALD holds information controlling the Ewald sums for structure constants entering into, e.g. the Madelung summations and Bloch summed structure constants (lmf). Most programs use quantities in this category to carry out Ewald sums (exceptions are lmstr and the molecules code lmmc).
Token  Arguments  Program  Optional  Default  Explanation 

AS  r  Y  2  Controls the relative number of lattice vectors in the real and reciprocal space.  
TOL  r  Y  1e8  Tolerance in the Ewald sums.  
NKDMX  i  Y  800  The maximum number of realspace lattice vectors entering into the Ewald sum, used for memory allocation. Normally you should not need this token. Increase NKDMX if you encounter an error message like this one: xlgen: too many vectors, n=…  
RPAD  r  Y  0  Scale rcutoff by RPAD when lattice vectors padded in oblong geometries. 
HAM
This category contains parameters defining the oneparticle hamiltonian.
Portions of HAM are read by these codes:
lm, lmfa, lmfgwd, lmfgws, lmf, lmgf, lmpg, lmdos, lmchk, lmscell, lmstr, lmctl, lmmc, tbe, lmmag.
Token  Arguments  Program  Optional  Default  Explanation 

NSPIN  i  ALL  Y  1  1 for nonspinpolarized calculations. 2 for spinpolarized calculations. NB: For the magnetic parameters below to be active, use NSPIN=2. 
REL  i  ALL  Y  1  0: for nonrelativistic Schrödinger equation. 1: for scalar relativistic approximation to the Dirac equation. 2: for Dirac equation (ASA only). 11: compute cores with the Dirac equation (lmfa only). 
SO  i  ALL  Y  0  0: no SO coupling. 1: Add L·S to hamiltonian. However, only the spindiagonal part of the density is retained. 2: Add Lz·Sz only to the hamiltonian, so the spin channels remain distinct. 3: Like 2, but also L·S−LzSz is included perturbatively in the eigenvalues only and in a manner that preserves independence in the spin channels. This generates eigenvalues very close to LS for a given potential, but the eigenfunctions are generated from H+LzSz only. As a result the eigenfunctions (and then the density) remain spindiagonal. There is some effect on the density, but the approximation seems to be rather good since the error on the eigenfunctions is of 2nd order in the perturbation. 11: Same as 1, but additionally decompose SO by site. See here for analysis and description of the different approximations. GWbased codes at present requires the spin channels to be kept separated and works, then, with SO=2,3 only. 
NONCOL  l  ASA  Y  F  F: collinear magnetism. T: noncollinear magnetism. 
SS  4 r  ASA  Y  0  Magnetic spin spiral, direction vector and angle. Example: nc/test/test.nc 1 
BFIELD  i  lm, lmf  Y  0  0: no external magnetic field applied. 1: add sitedependent constant external Zeeman field (requires NONCOL=T). Fields are read from file bfield.ext. 2: add Bz·Sz only to hamiltonian. Examples: fp/test/test.fp gdn nc/test/test.nc 5 
BXCSCAL  i  lm, lmgf  Y  0  This tag provides an alternative means to add an effective external magnetic field in the LDA. 0: no special scaling of the exchangecorrelation field. 1: scale the magnetic part of the LDA XC field by a sitedependent factor 1 + sbxci as described below. 2: scale the magnetic part of the LDA XC field by a sitedependent factor $(1{+}\mathrm{sbxc}_i^2)^{1/2}$ as described below. This is a special mode used to impose constraining fields on rotations, used, e.g. by the CPA code. Sitedependent scalings sbxci are read from file bxc.ext. 
XCFUN  i  ALL  Y  2  Specifies local part exchangecorrelation functional. 0,#2,#3: Use libxc exchange functional #2 and correlation functional #3 1: CeperlyAlder 2: BarthHedin (ASW fit) 
GGA  i  ALL  Y  0  Specifies gradient additions to exchangecorrelation functional (not used when XCFUN=0,#2,#3). 0. No GGA (LDA only) 1. LangrethMehl 2. PW91 3. PBE 4. PBE with Becke exchange This tutorial uses the PBE functional. To compare the internally coded PBE functional with libxc, try fp/test/test.fp te 
PWMODE  i  lmf, lmfgwd  Y  0  Controls how APWs are added to the LMTO basis. 1s digit: 0. LMTO basis only 1. Mixed LMTO+PW 2. PW basis only Examples: fp/test/test.fp srtio3 and fp/test/test.fp felz 4 10s digit: 0. PW basis fixed to $\exp(i\mathbf{G}\cdot\mathbf{r})$ (less accurate, but simpler) 1. PW basis $\exp(i(\mathbf{k{+}G})\cdot\mathbf{r})$ symmetryconsistent, but basis depends on k. Example: fp/test/test.fp te 
PWEMIN  r  lmf, lmfgwd  Y  0  Include APWs with energy E > PWEMIN (Ry) 
PWEMAX  r  lmf, lmfgwd  Y  Include APWs with energy E < PWEMAX (Ry)  
NPWPAD  i  lmf, lmfgwd  Y  1  If >0, overrides default padding of variable basis dimension. Certain arrays have fixed dimension that must be at least as large as the rank of the hamiltonian. The APW basis is depends on k if PWMODE>10, so some padding must be added to this fixed dimesion to ensure that these arrays can accommodate any k. Normally the code will internally select a sensible default. In the event it is not large enough (the program will stop), you can enlarge the padding with this token. 
RDSIG  i  lmf, lmfgwd, lm, lmgf  Y  0  Controls how the QSGW selfenergy Σ^{0} substitutes for the LDA exchange correlation functional. Note: the GW codes store $\Sigma^0{}V_{xc}^\mathrm{LDA}$ in file sigm.ext. 1s digit: 0 do not read Σ^{0} 1 read file sigm.ext, if it exists, and add it to the LDA potential 2 same as 1 but symmetrize sigm after reading Add 4 to retain only real part of realspace sigma 10s digit: 0 simple interpolation (not recommended). 1 approximate high energy parts of sigm by diagonal. Optionally add the following (the same functionality using rsig on the command line): 10000 to indicate the sigma file was stored in the full BZ (no symmetry operations are assumed). 20000 to use the minimum neighbor table (only one translation vector at the surfaces or edges; cannot be used with symmetrization). 40000 to allow mismatch between expected kpoints and file values. 
RSSTOL  r  ALL  Y  5e6  Max tolerance in Bloch sum error for realspace Σ^{0}. Σ^{0} is read in kspace and is immediately converted to real space by inverse Bloch transform. The real space form is forward Bloch summed and checked against the original kspace Σ^{0}. If the difference exceeds RSSTOL the program will abort. The conversion should be exact to machine precision unless the range of Σ^{0} is truncated. You can control the range of realspace Σ^{0} with RSRNGE below. 
RSRNGE  r  ALL  Y  5  Maximum range of connecting vectors for realspace Σ^{0} (units of ALAT). 
NMTO  i  ASA  Y  0  Order of polynomial approximation for NMTO hamiltonian. 
KMTO  r  ASA  Y  Corresponding NMTO kinetic energies. Read NMTO values, or skip if NMTO=0.  
EWALD  l  lm  Y  F  Make strux by Ewald summation (NMTO only). 
VMTZ  r  ASA  Y  0  Muffintin zero defining wave functions. 
QASA  i  ASA  Y  3  A parameter specifying the definition of ASA moments Q_{0},Q_{1},Q_{2} 0. band code accumulates Q_{1}, Q_{2} from true energy moments of sphere charges (KKR style). Sphere code generates density from Q_{0}×${\phi}^2$ + Q_{2}×$[\dot{\phi}^2{+}{\phi}{\ddot{\phi}}]$. This (Methfessel convention) is approximate but decouples potential parameters from charges. 1. Sphere code generates density from Q_{0}×${\phi}^2$ + Q_{2}×$[\dot{\phi}^2{}\langle{\dot{\phi}}^2\rangle\phi^2]$; thus Q_{0} is the sphere charge. 2. Q_{1},Q_{2} accumulated from $\langle \phi \dot{\phi} \rangle$ and $\langle{\dot{\phi}}^2\rangle$, rather than power moments (not applicable to lmgf, lmpg). 3. 1+2 (Standard conventions). Add 4 to cause the sphere integrator to construct $\dot{\phi}$ and $\ddot{\phi}$ by outward radial integration only. 
PMIN  r,r,…  ALL  Y  0 0 0 …  Global minimum in fractional part of the continuous principal quantum number $P_l$. Enter values for l=0,..lmx. 0: no minimum constraint. # : with #<1, fractional part of $P_l$. 1: use freeelectron value as minimum. Note: lmf always uses a minimum constraint, the freeelectron value (or slightly higher if AUTOBAS_GW is set). You can set the floor still higher with PMIN=#. 
PMAX  r,r,…  ALL  Y  0 0 0 …  Global maximum in fractional part of the continuous principal quantum number $P_l$. Enter values for l=0,..lmx. 0 : no maximum constraint. #: with #<1, uppper bound of fractional P is #. 
OVEPS  r  ALL  Y  0  The overlap is diagonalized and the hilbert space is contracted, discarding the part with eigenvalues of overlap < OVEPS. Especially useful with the PMT basis, where the combination of smooth Hankel functions and APWs has a tendency to make the basis overcomplete. 
OVNCUT  i  ALL  Y  0  This tag has a similar objective to OVEPS. The overlap is diagonalized and the hilbert space is contracted, discarding the part belonging to lowest OVNCUT evals of overlap. Supersedes OVEPS, if present. 
GMAX  r  lmf, lmfgwd  N  Gvector cutoff used to create the mesh for the interstitial density (Ry^{1/2}). A uniform mesh with spacing between points in the three directions as homogeneous as possible, with G vectors G < GMAX. This input is required; but you may omit it if you supply information with the FTMESH token.  
FTMESH  i1 [i2 i3]  FP  N  The number of divisions specifying the uniform mesh density along the three lattice vectors. The second and third arguments default to the value of the first one, if they are not specified. This input is used only if the parser failed to read the GMAX token.  
TOL  r  FP  Y  1e6  Specifies the precision to which the generalized LMTO envelope functions are expanded in a Fourier expansion of G vectors. 
FRZWF  l  FP  Y  F  Set to T to freeze the shape of the augmented part of the wave functions. Normally their shape is updated as the potential changes, but with FRZWF=t the potential used to make augmentation wave functions is frozen at what is read from the restart file (or freeatom potential if starting from superposing free atoms). This is not normally necessary, and freezing wave functions makes the basis slightly less accurate. However, there are slight inconsistencies when these orbitals are allowed to change shape. Notably the calculated forces do not take this shape change into account, and they will be slightly inconsistent with the total energy. 
FORCES  i  FP  Y  0  Controls how forces are to be calculated, and how the secondorder corrections are to be evaluated. Through the variational principle, the total energy is correct to second order in deviations from selfconsistency, but forces are correct only to first order. To obtain forces to second order, it is necessary to know how the density would change with a (virtual) displacement of the core+nucleus, which requires a linear response treatment. lmf estimates this change using one of ansatz:1. the freeatom density is subtracted from the total density for nuclei centered at the original position and added back again at the (virtually) displaced position. The core+nucleus is shifted and screened assuming a Lindhard dielectric response. You also must specify ELIND, below. 
ELIND  r  lmf  Y  1  A parameter in the Lindhard response function, (the Fermi level for a freeelectron gas relative to the bottom of the band). You can specify this energy directly, by using a positive number for the parameter. If you instead use a negative number, the program will choose a default value from the total number of valence electrons and assuming a freeelectron gas, scale that default by the absolute value of the number you specify. If you have a simple sp bonded system, the default value is a good choice. If you have d or f electrons, it tends to overestimate the response. Use something smaller, e.g. ELIND=0.7. ELIND is used in three contexts: (1) in the force correction term; see FORCES= above. (2) to estimate a selfconsistent density from the input and output densities after a band pass. (3) to estimate a reasonable smooth density from a starting density after atoms are moved in a relaxation step. 
SIGP[…]  lmf, lmfgwd  Y  Parameters used to interpolate the selfenergy $\Sigma^0$. Used in conjunction with the GW package. See gw for description. Default: not used.  
SIGP_MODE  i  lmf, lmfgwd  Y  4  Specifies the linear function used for matrix elements of $\Sigma^0$ at highlylying energies. Highlying states should be far enough away from the Fermi level that their effect should be small, and the result should depend very little on the choice of the constraint. By approximating $\Sigma^0$ for these states, one ensures that the LDA and quasiparticle eigenvectors for those states are the same. 0. constrain $\Sigma^0_{ii}$ to be greater than $a+b \varepsilon^{lda}_i$. 1. constrain $\Sigma^0_{ii}$ to be equal to $a+b \varepsilon^{lda}_i$. 2. constrain $\Sigma^0_{ii}$ to be defined in the interval $a < \Sigma^0_{ii} < b$. 3. constrain $\Sigma^0_{ii}$ as in SIGP_MODE=1. The difference between modes 1 and 3 are merely informational. 4. constrain $\Sigma^0_{ii}$ to be a constant. Its value is calculated by the GW package and read from sigm.ext. This mode requires no information from the user. It is the recommended mode, available in version 7.7 or later. 
SIGP_NMAX  i  lmf, lmfgwd  Y  0  Integer specifying which of the highest selfenergy matrix elements are to be approximated. States higher than SIGP_NMAX have the offdiagonal part of sigma stripped; unlike the lowlying states, the diagonal part of $\Sigma^0$ is constrained (see SIGP_MODE above). If SIGP_NMAX is lower or equal to 0, it is not used; see SIGP_EMAX below. 
SIGP_EMAX  r  lmf, lmfgwd  Y  2.0  Alternative way to specify approximation of highlying elements of the selfenergy matrix. It is only used if SIGP_NMAX is lower or equal to 0, which case SIGP_EMAX is an energy cutoff: states above SIGP_EMAX are approximated. 
SIGP_NMIN  i  lmf, lmfgwd  Y  0  Integer specifying how many of the lowestlying states are approximated by discarding the offdiagonal parts in the basis of LDA functions. If SIGP_NMIN is zero, no lowlying states are approximated. 
SIGP_EMIN  r  lmf, lmfgwd  Y  0.0  Alternative way to specify approximations of lowlying elements of the selfenergy matrix. It is only used if SIGP_NMIN<0, which case SIGP_EMIN is an energy cutoff: states below SIGP_EMIN are approximated. 
SIGP_A  r  lmf, lmfgwd  Y  0.02  Coefficient $a$ in the linear fit $a+b \varepsilon^{lda}_i$ (see SIGP_MODE=0,…,3). If SIGP_MODE=4, SIGP_A is not used. In the linear constraints (SIGP_MODE=0,1) it is the constant coefficient; for SIGP_MODE=2, it is the lower bound. Note that its default value is a good estimate for Si. 
SIGP_B  r  lmf, lmfgwd  Y  0.06  Coefficient $b$ in the linear fit $a+b \varepsilon^{lda}_i$ (see SIGP_MODE=0,…,3). If SIGP_MODE=4, SIGP_B is not used. In the linear constraints (SIGP_MODE=0,1) it is the linear coefficient; for SIGP_MODE=2, it is the upper bound. Note that its default value is a good estimate for Si. 
SIGP_EFIT  r  lmf, lmfgwd  Y  0  Lower bound for the least squares fit required for a reasonable evaluation of the above coefficients SIGP_A and SIGP_B when SIGP_MODE=0,…,3. For SIGP_MODE<3, lmf will make a leastsquares fit to $\Sigma^0_{ii}$ for states higher than SIGP_EFIT. For SIGP_MODE=3, lmf will make a leastsquares fit for states between SIGP_EFIT and SIGP_EMAX, which must be used if one is going to evaluate $\Sigma^0_{ii}$ for states above some SIGP_EMAX. For the case SIGP_MODE<3 one must invoke lmf one the mesh of kpoints for which the selfenergy is known (there appear to be fewer problems with interpolation on that mesh). lmf accumulates the minimum, maximum, and leastsquares fit for the for all the states above the cutoff. Look in the output for a line beginning with “hambls:”. Also, setting the verbosity above 45, lmf will print out the calculated $\Sigma^0_{ii}$ for each of these states, together with the constrained value. lmf will write to file sigii.ext the data used to make the fit, and summarize the fit and the end of the file. If SIGP_MODE=4, SIGP_EFIT is not needed. 
AUTOBAS[…]  lmfa, lmf, lmfgwd  Y  Parameters associated with the automatic determination of the basis set. These switches greatly simplify the creation of an input file for lmf. Note: Programs lmfa and lmf both use tokens in the AUTOBAS tag but they mean different things, as described below. This is because lmfa generates the parameters while lmf uses them. Default: not used.  
AUTOBAS_GW  i  lmfa  Y  0  Set to 1 to tailor the autogenerated basis set file basp0.ext to a somewhat larger basis, better suited for GW. 
AUTOBAS_GW  i  lmf  Y  0  Set to 1 to float log derivatives $D_l$ a bit more conservatively — better suited to GW calculations. 
AUTOBAS_LMTO  i  lmfa  Y  0  lmfa autogenerates a trial basis set, saving the result into basp0.ext. LMTO is used in an algorithm to determine how large a basis it should construct: the number of orbitals increases as you increase LMTO. This algorithm also depends on which states in the free atom carry charge. Let lq be the highest l which carries charge in the free atom. There are the following choices for LMTO: 0. standard minimal basis; same as LMTO=3. 1. The hyperminimal basis, which consists of envelope functions corresponding those l which carry charge in the free atom, e.g. Ga sp and Mo sd (this basis is only sensible when used in conjunction with APWs). 2. All l up to lq+1 if lq<2; otherwise all l up to lq. 3. All l up to min(lq+1, 3). For elements lighter than Kr, restrict l≤2. For elements heavier than Kr, include l to 3. 4. (Standard basis) Same as LMTO=3, but restrict l≤2 for elements lighter than Ar. 5. (Large basis) All l up to max(lq+1,3) except for H, He, Li, B (use l=spd). Use the MTO token (see below) in combination with this one. MTO controls whether the LMTO basis is 1κ or 2κ, meaning whether 1 or 2 envelope functions are allowed per l channel. 
AUTOBAS_MTO  i  lmfa  Y  0  Autogenerate parameters that control which LMTO basis functions are to be included, and their shape. Tokens RSMH,EH (and possibly RSMH2,EH2) determine the shape of the MTO basis. lmfa will determine a reasonable set of RSMH,EH automatically (and RSMH2,EH2 for a 2κ basis), fitting to radial wave functions of the free atom. Note: lmfa can generate parameters and write them to file basp0.ext. lmf can read parameters from basp.ext. You must manually create basp.ext, e.g. by copying basp0.ext into basp.ext. You can tailor basp.ext with a text editor. Here are the following choices for MTO: 0: do not autogenerate basis parameters. 1 or 3 : 1κ parameters with Zdependent LMX. 2 or 4: 2κ parameters with Zdependent LMX. For lmfa 1 and 3 are equivalent, as are 2 and 4. 
AUTOBAS_MTO  i  lmf, lmfgwd  Y  0  Read parameters RSMH,EH,RSMH2,EH2 that control which LMTO basis functions enter the basis. Once initial values have been generated you can tune these parameters automatically for the solid, using lmf with the –optbas switch; see here (or for a simple input file guide, here) and here. The –optbas step is not essential, especially for large basis sets, but it is a way to improve on the basis without increasing the size. Here are the following choices for MTO: 0 Parameters not read from basp.ext; they are specified in the input file ctrl.ext. 1 or 3: 1κ parameters may be read from the basis file basp.ext, if they exist. 2 or 4: 2κ parameters may be read from the basis file basp.ext, if they exist. 1 or 2: Parameters read from ctrl.ext take precedence over basp.ext. 3 or 4: Parameters read from basp.ext take precedence over those read from ctrl.ext. 
AUTOBAS_PNU  i  lmfa  Y  0  Autoset boundary condition for augmentation part of basis, through specification of the continuous principal quantum number $P_l$. 0 do not make P 1 Find P for l < SPEC_LMXA from free atom wave function; save in basp0.ext. 
AUTOBAS_PNU  i  lmf, lmfgwd  Y  0  Autoset boundary condition for augmentation part of basis, through specification of the continuous principal quantum number $P_l$. 0 do not attempt to read P from basp.ext. 1 Read P from basp.ext, for species which P is supplied. 
AUTOBAS_LOC  i  lmfa, lmf, lmfgwd  Y  0  Autoset local orbital parameters PZ, which determine which deep or highlying states are to be included as local orbitals. Used by lmfa to control whether parameters PZ are to be sought: 0: do not autogenerate PZ. 1 or 2: autogenerate PZ. Default: 0 Used by lmf and lmfgwd to control how PZ is read: 1 or 2: read parameters PZ. 1: Nonzero values from ctrl file take precedence over basis file input. Default: 1 
AUTOBAS_RSMMX  r  lmfa  Y  2/3  sets an upper bound to LMTO smoothing radius RSMH, when autogenerating a basis set. Value is a multiple of the MT radius. 
AUTOBAS_EHMX  r  lmfa  Y  –  sets an upper bound to LMTO smoothed Hankel energy EH, when autogenerating a basis set. Default depends on whether AUTOBAS_GW is set. 
AUTOBAS_ELOC  r  lmfa  Y  2 Ry  The first of two criteria to decide which orbitals should be included in the valence as local orbitals. If the energy of the free atom wave function exceeds (is more shallow than) ELOC, the orbital is included as a local orbital. 
AUTOBAS_QLOC  r  lmfa  Y  0.005  The second of two criteria to decide which orbitals should be included in the valence as local orbitals. If the fraction of the free atom wave function’s charge outside the augmentation radius exceeds QLOC, the orbital is included as a local orbital. 
AUTOBAS_PFLOAT  i1 i2  lmf, lmfgwd  y  1 1  Governs how the Pnu are set and floated in the course of a selfconsistency cycle. The 1st argument controls default starting values of P and lower bounds to P when it is floated. 0: Use pre2002 (i.e. version 6) lower bound for P (lmf only). 1: Use defaults and float lower bound designed for LDA. 2: Use defaults and float lower bound designed for GW. The 2nd argument controls how the band center of gravity (CG) is determined — used when floating P. 0: band CG is found by a traditional method. 1: band CG is found from the true energy moment of the density. 
GF
Category GF is intended for parameters specific to the Green’s function code lmgf. It is read by lmgf.
Token  Arguments  Program  Optional  Default  Explanation 

MODE  i  ASA  Y  0  0: do nothing. 1: selfconsistent cycle. 10: Transverse magnetic exchange interactions J(q). 11: Read J(q) from disk and analyze results. 14: Longitudinal exchange interactions. 20: Transverse χ^{+−} from ASA Green’s function. 21: Read χ from disk and analyze results. 20: Transverse χ^{++}, χ^{−−} from ASA Green’s function Caution: Modes 14 and higher have not been maintained. 
GFOPTS  c  ASA  Y  ASCII string with switches governing execution of lmgf. Use ’;’ to separate the switches. Available switches: p1 First order of potential function. p3 Third order of potential function. pz Exact potential function (some problems; not recommended). Use only one of the above; if none are used, the code generates second order potential functions. idos integrated DOS (by principal layer in the lmpg case). noidos suppress calculation of integrated DOS pdos accumulate partial DOS. emom accumulate output moments; use noemom to suppress. noemom suppresss accumulation of output moments. sdmat make site densitymatrix. dmat make densitymatrix. frzvc do not update potential shift needed to obtain charge neutrality. padtol Tolerance in Pade correction to charge. If tolerance exceeded, lmgf will repeat the band pass with an updated Fermi level. omgtol (CPA) tolerance criterion for convergence in coherent potential. omgmix (CPA) linear mixing parameter for iterating convergence in coherent potential. nitmax (CPA) maximum number of iterations to iterate for coherent potential. lotf (CPA). dz (CPA).  
DLM  i  ALL  Y  0  Disordered local moments for CPA. Governs selfconsistency for both chemical CPA and magnetic CPA. 12 : normal CPA/DLM calculation: charge and coherent potential Ω both iterated to selfconsistency. 32 : Ω alone is iterated to selfconsistency. 
BXY  1  ALL  Y  F  (DLM) Setting this switch to T generates a sitedependent constraining field to properly align magnetic moments. In this context constraining field is applied by scaling the LDA exchangecorrelation field. The scaling factor is [1+bxc(ib)^2]^{1/2}. A table of bxc is kept for each site in the first column of file shfac.ext. See also HAM_BXCSCAL 
TEMP  r  ALL  Y  0  (DLM) spin temperature. 
GW
Category GW holds parameters specific to GW calculations, particularly for the GW driver lmfgwd. Most of these tokens supply values for tags in the GWinput template when lmfgwd generates it (jobgw 1).
Token  Arguments  Program  Optional  Default  Explanation 

CODE  i  lmfgwd  Y  2  This token tells what GW code you are creating input files for. lmfgwd serves as a driver to several GW codes. 0. First GW version v033a5 (code still works but it is no longer maintained) . 2. Current version of GW codes . 1. Driver for the Julich spex code (not fully debugged or maintained). 
NKABC  1 to 3 i  Y  Defines the kmesh for GW. This token serves the same function for GW as BZ_NKABC does for the LDA codes, and the input format is the same. When generating a GWinput template, lmfgwd passes the contents of NKABC to the n1n2n3 tag. Note: Shell scripts lmgw and lmgwsc used for the GW codes may also use this token. When invoked with switches –getsigp or –getnk, they will modify the n1n2n3 in GWinput. The data they use is taken from GW_NKABC.  
MKSIG  i  lmfgwd  Y  3  (selfconsistent calculations only). Controls the form of $\Sigma^0_{nn'}$ (the QSGW approximation to the dynamical selfenergy $\Sigma^0_{nn'}(E)$, where $\Sigma_{nn'}(E)$ refers to a matrix element of Σ between eigenstates n and n′, at energy E relative to E_{F}. When generating a GWinput template, lmfgwd passes MKSIG to the iSigMode tag. Values of this tag have the following meanings. 0. do not make Σ^{0} 1. Σ^{0} = Σ_{nn’} (E_{F}) if n≠n’, and Σ_{nn}(E_{n}) if n=n’: mode B, Eq.(11) in Phys. Rev. B76, 165106 (2007) 3. Σ^{0} = 1/2[Σ_{nn’} (E_{n}) + Σ_{nn’} (E_{n’})]: mode A, Eq.(10) in Phys. Rev. B76, 165106 (2007) 5. “eigenvalue only” selfconsistency Σ^{0} = δ_{nn’}Σ_{nn‘ }(E_{n}) 
GCUTB  r  lmfgwd  Y  2.7  Gvector cutoff for basis envelope functions as used in the GW package (Ry^{1/2}). When generating a GWinput template, lmfgwd passes GCUTB to the QpGcut_psi tag in GWinput.. 
GCUTX  r  lmfgwd  Y  2.2  Gvector cutoff for interstitial part of twoparticle objects such as the screened coulomb interaction (Ry^{1/2}). When generating a GWinput template, lmfgwd passes GCUTX to the QpGcut_cou tag. 
ECUTS  r  lmfgwd  Y  2.5 Ry  (for selfconsistent calculations only). Maximum energy for which to calculate the $V^{xc}$ described in MKSIG above. This energy should be larger than HAM_SIGP_EMAX which is used to interpolate $V^{xc}$. When generating a GWinput template, lmfgwd passes ECUTS+1/2 to the emax_sigm tag in the GWinput file. 
NIME  i  lmfgwd  Y  6  Number of frequencies on the imaginary integration axis when making the correlation part of Σ. When generating a GWinput template, lmfgwd passes NIME to the new tag. 
DELRE  r  lmfgwd  Y  0.01, 0.1  Frequency mesh parameters DW and OMG defining the real axis mesh in the calculation of Im $\chi_0$. The i^{th} mesh point is given by: ω_{i}=DW×(i−1) + [DW×(i−1)]^{2}/OMG/2 Points are approximately uniformly spaced, separated by DW, up to frequency OMG, around which point the spacing begins to increase linearly with frequency. When generating a GWinput template, lmfgwd passes DELRE(1) to the dw tag and DELRE(2) to the omg_c tag. Note: the similarity to OPTICS_DW used by the optics part of lmf and lm. 
DELTA  r  lmfgwd  Y  1e4  δfunction broadening for calculating χ_{0}, in atomic units. Tetrahedron integration is used if DELTA<0. When generating a GWinput template, lmfgwd passes DELTA to the delta tag. 
GSMEAR  r  lmfgwd  Y  .003  Broadening width for smearing pole in the Green’s function when calculating Σ. This parameter is sometimes important in metals, e.g. Fe. When generating a GWinput template, lmfgwd passes GSMEAR to the esmr tag. The tag is described in this manual 
PBTOL  r  lmfgwd  Y  .001  Overlap criterion for product basis functions inside augmentation spheres. The overlap matrix of the basis of product functions generated and diagonalized for each l. Functions with overlaps less than PBTOL are removed from the product basis. When generating a GWinput template, lmfgwd passes PBTOL to the second line after the start of the PRODUCT_BASIS section. 
HEADER
This category is optional, and merely prints to the standard output whatever text is in the category. For example:
HEADER This line and the following one are printed to
standard output whenever a program is run.
NEXT
Alternately:
HEADER [ In this form only two lines reside within the
category delimiters,]
and only two lines are printed.
IO
This optional category controls what kind of information, and how much, is written to the standard output file.
Token  Arguments  Program  Optional  Default  Explanation 

SHOW  1  all  Y  F  Echo lines as they are read from input file and parsed by the proprocessor. Commandline argument show provides the same functionality. 
HELP  1  all  Y  F  Show what input would be sought, without attempting to read data. Commandline argument input provides the same functionality. 
VERBOS  1 to 3  all  Y  30  Sets the verbosity. 20 is terse, 30 slightly terse, 40 slightly verbose, 50 verbose, and so on. If more than one number is given, later numbers control verbosity in subsections of the code, notably the parts dealing with augmentation spheres. May also be set from the commandline: pr#1[,#2] 
IACTIV  1  all  Y  F  Turn on interactive mode. Programs will prompt you with queries, in various contexts. May also be controlled from the commandline: iactiv or iactiv=no. 
TIM  1 or 2  all  Y  0, 0  Prints out CPU usage of blocks of code in a tree format. First value sets tree depth. Second value, if present, prints timings on the fly. May also be controlled from the commandline: time=#1[,#2] 
ITER
The ITER category contains parameters that control the requirements to reach selfconsistency.
It applies to all programs that iterate to selfconsistency: lm, lmf, lmmc, lmgf, lmpg, tbe, lmfa.
A detailed discussion can be found at the end of this document.
Token  Arguments  Program  Optional  Default  Explanation 

NIT  i  all  Y  1  Maximum number of iterations in the selfconsistency cycle. 
MIX  c  all  Y  A string of mixing rules for mixing input, output density in the selfconsistency cycle. The syntax is given below. See here for detailed description of the mixing.  
CONV  r  all  Y  1e5  Maximum energy change from the prior iteration for selfconsistency to be reached. See annotated lmf output. 
CONVC  r  all  Y  3e5  Maximum in the RMS difference in the density n^{out}−n^{in}. See below. 
UMIX  r  all  Y  1  Mixing parameter for density matrix; used with LDA+U 
TOLU  r  all  Y  0  Tolerance for density matrix; used with LDA+U 
NITU  i  all  Y  0  Maximum number of LDA+U iterations of density matrix 
AMIX  c  ASA  Y  Mixing rules when extra degrees of freedom, e.g. Euler angles, are mixed independently. Uses the same syntax as MIX.  
NRMIX  i1 i2  ASA, lmfa  Y  80, 2  Used when selfconsistency is needed inside an augmentation sphere. This occurs when the density is determined from the momentsQ0,Q1,Q2 in the ASA; or in the free atom code, just Q0. i1: max number of iterations i2: number of prior iterations for Anderson mixing 2 of the sphere density Note: You will probably never need to use this token. 
OPTICS
Optics functions available with the ASA extension packages OPTICS.
It is read by lm and lmf.
Token  Arguments  Program  Optional  Default  Explanation 

MODE  i  OPTICS  Y  0  0: make no optics calculations 1: generate linear $\varepsilon_2$ 20: generate second harmonic ε Example: optics/test/test.optics sic The following cases (MODE<0) generate joint or single densityofstates. Note: MODE<0 works only with LTET=3 described below. −1: generate joint densityofstates Examples: (ASA) optics/test/test.optics all 4 (FP) fp/test/test.fp zbgan −2: generate joint densityofstates, spin 2 Example:optics/test/test.optics fe 6 −3: generate updown joint densityofstates −4: generate downup joint densityofstates −5: generate spinup single densityofstates Example: optics/test/test.optics all 7 −6: generate spindn single densityofstates 
LTET  i  OPTICS  Y  0  0: Integration by MethfesselPaxton sampling 1: standard tetrahedron integration 3: enhanced tetrahedron integration Note: In the metallic case, states near the Fermi level must be treated with partial occupancy. LTET=3 is the only scheme that handles this properly. It was adapted from the GW package and has extensions, e.g. the ability to handle nonvertical transitions $k^{occ} \ne k^{unocc}$. 
WINDOW  r1 r2  OPTICS  N  0 1  Energy (frequency) window over which to calculate Im[ε(ω)]. Im ε is calculated on a mesh of points $\omega_i$. The mesh spacing is specified by NPTS or DW, below. 
NPTS  i  OPTICS  N  501  Number of mesh points in the energy (frequency) window. Together with WINDOW, NPTS specifies the frequency mesh as: $\omega_i$ = WINDOW(1) + DW×(i−1) where DW = (WINDOW(2)−WINDOW(1))/(NPTS−1) Note: you may alternatively specify DW below. 
DW  r1 [r2]  OPTICS  Y  Frequency mesh spacing DW[,OMG]. You can supply either one argument, or two. If one argument (DW) is supplied, the mesh will consist of evenly spaced points separated by DW. If a second argument (OMG) is supplied, points are spaced quadratically as: $\omega_i$ = WINDOW(1) + DW×(i−1) + [DW×(i−1)]^{2}/OMG/2 Spacing is approximately uniform up to frequency OMG; beyond which it increases linearly. Note: The quadratic spacing can be used only with LTET=3.  
FILBND  i1 [i2]  OPTICS  Y  0 no. electrons  i1[,i2] occupied energy bands from which to calculate ε using first order perturbation theory, without local fields. i1 = lowest occupied band i2 = highest occupied band (defaults to no. electrons) 
EMPBND  i1 [i2]  OPTICS  Y  0 no. bands  i1[,i2] empty energy bands from which to calculate ε using first order perturbation theory, without local fields. i1 = lowest unoccupied band i2 = highest unoccupied band (defaults to no. bands) 
PART  i  OPTICS  Y  0  Resolve ε or joint DOS into bandtoband contributions, or by k. Result is output into file popt.ext. 0. No decomposition 1. Resolve ε or DOS into individual (occ,unocc) contributions Example: optics/test/test.optics ogan 5 2. Resolve ε or DOS by k Example: optics/test/test.optics all 6 3. Both 1 and 2 Add 10 to write popt as a binary file. 
CHI2[..]  OPTICS  Y  Tag containing parameters for second harmonic generation. Not calculated unless tag is parsed. Example: optics/test/test.optics sic  
CHI2_NCHI2  i  OPTICS  N  0  Number of direction vectors for which to calculate χ_{2}, i.e. the nonlinear susceptibility tensor. 
CHI2_AXES  i1, i2, i3  OPTICS  N  Direction vectors for each of the NCHI2 sets  
ESCISS  r  OPTICS  Y  0  Scissors operator (constant energy added to unoccupied levels, in Ry) 
ECUT  r  OPTICS  Y  0.2  Energy safety margin for determining (occ,unocc) window. lmf will attempt to reduce the number of (occ,unocc) pairs by restricting, for each k, transitions that contribute to the response, i.e. to those inside the optics WINDOW. The window is padded by ECUT to include states outside, but near the edge of the window. States outside window may nevertheless make contribution, e.g. because they can be part of a tetrahedron that does contribute. If you do not want lmf to restrict the range, use ECUT<0. 
OPTIONS
Portions of OPTIONS are read by these codes:
lm, lmfa, lmfgwd, lmfgws, lmf, lmmc, lmgf, lmdos, lmstr, lmctl, lmpg, tbe.
Token  Arguments  Program  Optional  Default  Explanation 

HF  1  lm, lmf  Y  F  If T, use the HarrisFoulkes functional only; do not evaluate output density. 
SHARM  1  ASA, lmf, lmfgwd  Y  F  If T, use true spherical harmonics, rather than real harmonics. 
FRZ  l  all  Y  F  (ASA) If T, freezes core wave functions. (FP) If T, freezes the potential used to make augmented partial waves, so that the basis set does not change with potential. 
SAVVEC  1  lm  Y  F  Save eigenvectors on disk. (This may be enabled automatically in some circumstances) 
Q=strn  c  all  Y  Q=SHOW, Q=ATOM, Q=HAM, Q=POT, Q=BAND, Q=DOS, Q=RHO make the program stop at selected points without completing a full iteration.  
SCR  i  ASA  Y  0  Is connected with the generation or use of the q>0 ASA dielectric response function. It is useful in cases when there is difficulty in making the density selfconsistent. See here for documentation. 0. Do not screen qout−qin. 1. Make the ASA response function P0. 2. Use P0 to screen qout−qin and the change in ves. 3. 1+2 (lmgf only). 4. Screen qout−qin from a model P0. 5. Illegal input. 6. Use P0 to screen the change in ves only. P0 and U should be updated every iteration, but this is expensive and not worth the cost. However, you can: Add 10k to recompute intrasite contribution U every kth iteration, 0<k≤9. Add 100k to recompute P0 every kth iteration (lmgf only). Examples: testing/test.scr and gf/test/test.gf mnpt 6 
ASA[…]  r  ASA  N  Parameters associated with ASAspecific input.  
ASA_ADNF  1  ASA  Y  F  Enables automatic downfolding of orbitals. 
ASA_NSPH  1  ASA  Y  0  Set to 1 to generate l>0 contributions (from neighboring sites) to l=0 electrostatic potential 
ASA_TWOC  i  ASA  Y  0  Set to 1 to use the twocenter approximation ASA hamiltonian 
ASA_GAMMA  i  ASA  Y  0  Set to 1 to rotate to the (orthogonal) gamma representation. This should have no effect on the eigenvalues for the usual threecenter hamiltonian, but converts the twocenter hamiltonian from first order to second order. Set to 2 to rotate to the spinaveraged gamma representation. The lm code does not allow downfolding with GAMMA≠0. 
ASA_CCOR  1  lm  Y  T  If F, suppresses the combined correction. By default it is enabled. Note: NB: if any orbitals are downfolded, CCOR is automatically enabled. 
ASA_NEWREP  1  lm  Y  F  Set to 1 to rotate structure constants to a userspecified representation. It requires special compilation to be effective 
ASA_NOHYB  1  lm  Y  F  Set to 1 to turn off hybridization 
ASA_MTCOR  1  lm  Y  F  Set to T to turn on Ewald MT correction 
ASA_QMT  r  NC  Y  0  Override standard background charge for Ewald MT correction Input only meaningful if MTCOR=T 
RMINES  r  lmchk  N  1  Minimum augmentation radius when finding new empty sites (getwsr) 
RMAXES  r  lmchk  N  2  Maximum augmentation radius when finding new empty sites (getwsr) 
NESABC  i,i,i  lmchk  N  100  Number of mesh divisions when searching for empty spheres (getwsr) 
PGF
Category PGF concerns calculations with the layer Green’s function program lmpg.
It is read by lmpg and lmstr.
Token  Arguments  Program  Optional  Default  Explanation 

MODE  i  ASA  Y  0: do nothing. 1: diagonal layer GF. Examples: pgf/test/test.pgf all 5 and pgf/test/test.pgf all 6 2: left and rightbulk GF. 3: find k(E) for left bulk. Example: pgf/test/test.pgf 2 4: find k(E) for right bulk. 5: Calculate ballistic current. Example: pgf/test/test.pgf femgo  
SPARSE  i  ASA  Y  0  0: Calculate G layer by layer using Dyson’s equation Example: pgf/test/test.pgf all 5 1: Calculate G using LU decomposition Example: pgf/test/test.pgf all 6 
PLATL  r  ASA  N  The third lattice vector of left bulk region  
PLATR  r  ASA  N  The third lattice vector of right bulk region  
GFOPTS  c  ASA  Y  ASCII string with switches governing execution of lmgf or lmpg. Use ‘;’ to separate the switches. Available switches: p1 First order of potential function p3 Third order of potential function pz Exact potential function (some problems; not recommended) Use only one of the above; if none are used, the code makes second order potential functions idos integrated DOS (by principal layer in the lmpg case) noidos suppress calculation of integrated DOS pdos accumulate partial DOS emom accumulate output moments; use noemom to suppress noemom suppresss accumulation of output moments sdmat make site densitymatrix dmat make densitymatrix frzvc do not update potential shift needed to obtain charge neutrality ‘padtol** Tolerance in Pade correction to charge. If tolerance exceeded, lmgf will repeat the band pass with an updated Fermi level omgtol (CPA) tolerance criterion for convergence in coherent potential omgmix (CPA) linear mixing parameter for iterating convergence in coherent potential nitmax (CPA) maximum number of iterations to iterate for coherent potential lotf (CPA) dz (CPA) 
SITE
Category SITE holds site information. As in the SPEC category, tokens must read for each site entry; a similar restriction applies to the order of tokens. Token ATOM= must be the first token for each site, and all tokens defining parameters for that site must occur before a subsequent ATOM=.
Token  Arguments  Program  Optional  Default  Explanation 

FILE  c  all  Y  Provides a mechanism to read site data from a separate file. File subs/iosite.f documents the syntax of the site file structure. The reccommended (standard) format has the following syntax: The first line should contain a ‘%’ in the first column, and a `version’ token vn=#. Structural data (see category STRUC documentation) may also be included in this line. Each subsequent line supplies input for one site. In the simplest format, a line would have the following: spid x y z where spid is the species identifier (same information would otherwise be specified by token ATOM= below) and x y z are the site positions. Examples: fp/test/test.fp er and fp/test/test.fp tio2 Bug: when you read site data from an alternate file, the reader doesn’t compute the reference energy. Kotani format (documented here but no longer maintained). In this alternative format the first four lines always specify data read in the STRUC category; see FILE= in STRUC. Then follow lines, one line for each site ib iclass spid x y z The first number is merely a basis index and should increment 1,2,3,4,… in successive lines. The second class index is ignored by these programs. The remaining columns are the species identifier for the site positions. If SITE_FILE is missing, the following are read from the ctrl file:  
ATOM  c  all  N  Identifies the species (by label) to which this atom belongs. It is a fatal error for the species not to have been defined.  
ATOM_POS  r1 r2 r3  all  N  The basis vector (3 elements), in dimensionless Cartesian coordinates. As with the primitive lattice translation vectors, the true vectors (in atomic units) are scaled from these by ALAT in category STRUC. NB: XPOS and POS are alternative forms of input. One or the other is required.  
ATMOM_XPOS  r1 r2 r3  all  N  Atom coordinates, as (fractional) multiples of the lattice vectors. NB: XPOS and POS are alternative forms of input. One or the other is required.  
ATOM_DPOS  r1 r2 r3  all  Y  0 0 0  Shift in atom coordinates to POS 
ATOM_RELAX  i1 i2 i3  all  Y  1 1 1  Relax site positions (lattice dynamics or molecular statics) or Euler angles (spin dynamics) 
ATOM_RMAXS  r  FP  Y  Sitedependent radial cutoff for structure constants, in a.u.  
ATOM_ROT  c  ASA  Y  Rotation of spin quantization axis at this site  
ATOM_PL  i  lmpg  Y  0  (lmpg) Assign principal layer number to this site 
SPEC
Category SPEC contains speciesspecific information. Because data must be read for each species, tokens are repeated (once for each species). For this reason, there is some restriction as to the order of tokens. Data for a specific species (Z=, R=, R/W=, LMX=, IDXDN= and the like described below) begins with a token ATOM=; input of tokens specific to that species must precede the next occurence of ATOM=.
The following tokens apply to the automatic sphere resizer:
Token  Arguments  Program  Optional  Default  Explanation 

SCLWSR  r  ALL  Y  0  SCLWSR>0 turns on the automatic sphere resizer. It defaults to 0, which turns off the resizer. The 10’s digit tells the resizer how to deal with resizing empty spheres; see this page. 
OMAX1  r1 r2 r3  ALL  Y  0.16, 0.18, 0.2  Constrains maximum allowed values of sphere overlaps. This overlap is defined as $(r_1{+}r_2{}d)/d$, where $r_1$ and $r_2$ ae the two sphere radii, and $d$ is the bond length. See this page. You may input up to three numbers, which correspond to atomatom, atomemptysphere, and emptysphereemptysphere overlaps respectively. 
OMAX2  r1 r2 r3  ALL  Y  0.4, 0.45, 0.5  Constrains maximum allowed values of sphere overlaps defined as $(r_1{+}r_2{}d)/\min(r_1,r_2)$; see this page. Both constraints are applied. 
WSRMAX  r  ALL  Y  0  Imposes an upper limit to any one sphere radius 
The following tokens are input for each species. Data sandwiched between successive occurences of ATOM apply to one species.
Token  Arguments  Program  Optional  Default  Explanation 

ATOM  c  all  N  A character string (8 characters or fewer) that labels this species. This label is used, e.g. by the SITE category to associate a species with an atom at a given site. ASAspecific: The species ID also names a disk file with information about that atom (potential parameters, moments, potential and some sundry other information). More precisely, species are split into classes, the program differentiates class names by appending integers to the species label. The first class associated with the species has the species label; subsequent ones have integers appended. Example: testing/test.ovlp 3  
Z  r  all  N  Nuclear charge. Normally an integer, but Z can be a fractional number. A fractional number implies a virtual crystal approximation to an alloy with some Z intermediate between the two integers sandwiching it.  
R  r  all  N  The augmentation sphere radius, in atomic units. This is a required input for most programs: choose one of R=, R/W= or R/A=. Read descriptions of the R/W AND R/A below for further remarks; also see this page for a more complete discussion on the choice of sphere radii. lmchk can find sphere radii automatically. Invoke lmchk with \–getwsr. You can also rescale asgiven radii to meet constraints with the SCLWSR token.  
R/W  r  all  N  R/W= ratio of the augmentation sphere radius to the average Wigner Seitz radius W. W is the radius of a sphere such that (4πW^{3<\sup>/3) = V/N, where V/N is the volume per atom.Thus if all radii are equal with R/W=1, the sum of sphere volumes would fill space, as is usual in the ASA.ASAspecific:You must choose the radii so that the sum of sphere volumes (4π/3ΣiRi3) equals the unit cell volume V; otherwise results may become unreliable. The spacefilling requirement means sphere may overlap quite a lot, particularly in open systems. If sphere overlaps get too large, (>20% or so) accuracy becomes an issue. In such a case you should add “empty spheres” to fill space. Use lmchk to print out sphere overlaps. lmchk also has an automatic empty spheres finder, which you invoke with the –findes switch; see here for a discussion.Example: testing/test.ovlp 3FPspecific:FP results are much less sensitive to the choice of sphere radii. Strictly, the spheres should not overlap, but because of lmf’s unique augmentation scheme, overlaps of up to 10% cause negligibly small errors as a rule.(This does not apply to GW calculations!)Even so, it is not advisable to let the overlaps get too large. As a general rule the Lcutoff should increase as the sphere radius increases. Also it has been found in practice that selfconsistency is harder to accomplish when spheres overlap significantly.}  
R/A  r  all  N  R/A = ratio of the aumentation sphere radius to the lattice constant  
A  r  all  Y  0.025  Radial mesh point spacing parameter. All programs dealing with augmentation spheres represent the density on a shifted logarithmic radial mesh. The ith point on the mesh is $r_i = b(e^{a(i1)}1)$. b is determined from the number of radial mesh points specified by NR. 
NR  i  all  Y  Depends on other input  Number of radial mesh points 
LMX  i  all  Y  NL1  Basis lcutoff inside the sphere. If not specified, it defaults to NL−1 
RSMH  r,r,…  lmf, lmfgwd  Y  0  Smoothing radii defining basis (a.u.), one radius for each l. RSMH and EH together define the shape of basis function in lmf. To optimize, try running lmf with optbas. 
EH  r,r,…  lmf, lmfgwd  Y  Hankel energies for basis (Ry), one energy for each l. RSMH and EH together define the shape of basis function in lmf.  
RSMH2  r,r,…  lmf, lmfgwd  Y  0  Basis smoothing radii, second group 
EH2  r,r,…  lmf, lmfgwd  Y  Basis Hankel function energies, second group  
LMXA  i  FP  Y  NL  1  Angular momentum lcutoff for projection of wave functions tails centered at other sites in this sphere. Must be at least the basis lcutoff (specified by LMX=). 
IDXDN  i  ASA  Y  1  A set of integers, one for each lchannel marking which orbitals should be downfolded. 0 use automatic downfolding in this channel. 1 leaves the orbitals in the basis. 2 folds down about the inverse potential function at $E_\nu$ 3 folds down about the screening constant alpha. In the FP case, 1 includes the orbital in the basis; >1 removes it 
KMXA  i  lmf, lmfgwd  Y  3  Polynomial cutoff for projection of wave functions in sphere. Smoothed Hankels are expanded in polynomials around other sites instead of Bessel functions as in the case of normal Hankels. 
RSMA  r  lmf, lmfgwd  Y  R * 0.4  Smoothing radius for projection of smoothed Hankel tails onto augmentation spheres. These functions are expanded in polynomials by integrating with Gaussians of radius RSMA at that site. RSMA very small reduces the polynomial expansion to a Taylor series expansion about the origin. For large KMXA the choice is irrelevant, but RSMA is best chosen that maximizes the convergence of smooth Hankel functions with KMXA. 
LMXL  i  lmf, lmfgwd  Y  NL  1  Angular momentum lcutoff for explicit representation of local charge on a radial mesh. 
RSMG  r  lmf, lmfgwd  Y  R/4  Smoothing radius for Gaussians added to sphere densities to correct multipole moments needed for electrostatics. Value should be as large as possible but small enough that the Gaussian doesn’t spill out significantly beyond the Radius of the MuffinTin (RMT). 
LFOCA  i  FP  Y  1  Prescribes how the core density is treated. 0 confines core to within RMT. Usually the least accurate. 1 treats the core as frozen but lets it spill into the interstitial 2 same as 1, but interstitial contribution to vxc treated perturbatively. 
RFOCA  r  FP  Y  R × 0.4  Smoothing radius fitting tails of core density. A large radius produces smoother interstitial charge, but less accurate fit. 
RSMFA  r  FP  Y  R/2  Smoothing radius for tails of freeatom charge density. Irrelevant except first iteration only (nonselfconsistent calculations using Harris functional). A large radius produces smoother interstitial charge, but somewhat less accurate fit. 
RS3  r  FP  Y  1  Minimum allowed smoothing radius for local orbital 
HCR  r  lm  Y  Hard sphere radii for structure constants. If token is not parsed, attempt to read HCR/R below  
HCR/R  r  lm  Y  0.7  Hard sphere radii for structure constants, in units of R 
ALPHA  r  ASA  Y  Screening parameters for structure constants  
DV  r  ASA  Y  0  Artificial constant potential shift added to spheres belonging to this species 
MIX  1  ASA  Y  F  Set to suppress selfconsistency of classes in this species 
IDMOD  i  all  Y  0  0 : floats P_{l} aka continuous principal quantum number to band center of gravity 1 : freezes $P_l$ 2 : freezes linearization energy $E_\nu$. 
CSTRMX  1  all  Y  F  Set to T to exclude this species when automatically resizing sphere radii 
GRP2  i  ASA  Y  0  Species with a common nonzero value of GRP2 are symmetrized, independent of symmetry operations. The sign of GRP2 is used as a switch, so species with negative GRP2 are symmetrized but with spins flipped (NSPIN=2) 
FRZWF  1  FP  Y  F  Set to T to freeze augmentation wave functions for this species 
IDU  i  all  Y  0  LDA+U mode: 0 No LDA+U 1 LDA+U with Around Mean Field limit double counting 2 LDA+U with Fully Localized Limit double counting 3 LDA+U with mixed double counting. IDU is a vector, with one number for each l. 
UH  r  all  Y  0  Hubbard U for LDA+U (Ry). UH is a vector, with one number for each l. 
JH  r  all  Y  0  Exchange parameter J for LDA+U (Ry). JH is a vector, with one number for each l. 
EREF=  r  all  Y  0  Reference energy subtracted from total energy 
AMASS=  r  FP  Y  Nuclear mass in a.u. (for dynamics)  
CHOLE  c  lmf, lm  Y  Channel for core hole. You can force partial core occupation. Syntax consists of two characters, the principal quantum number and the second one of ‘s’, ‘p’, ‘d’, ‘f’ for the l quantum number, e.g. ‘2s’ See Partially occupied core holes for description and examples. Default: nothing  
CHQ  r[,r]  all  Y  1 0  First number specifies the number of electrons to remove from the l channel specified by CHOLE=. Second (optional) number specifies the hole magnetic moment. See Partially occupied core holes for description and examples. 
P  r,r,…  all  Y  Starting values for P_{l}, aka “continuous principal quantum number”, one for each l=0..LMXA Default: taken from an internal table.  
PZ  r,r,…  FP  Y  0  starting values for local orbital’s potential functions, one for each of l=0..LMX. Setting PZ=0 for any l means that no local orbital is specified for this l. Each integer part of PZ must be either one less than P (semicore state) or one greater (highlying state). 
Q  r,r,…  all  Y  Charges for each lchannel making up freeatom density Default: taken from an internal table.  
MMOM  r,r,…  all  Y  0  Magnetic moments for each lchannel making up freeatom density Relevant only for the spinpolarized case. 
STR
Category STR contains information connected with realspace structure constants, used by the ASA programs. It is read by lmstr, lmxbs, lmchk, and tbe.
Token  Arguments  Program  Optional  Default  Explanation 

RMAXS  r  all  Y  Radial cutoff for strux, in a.u. If token is not parsed, attempt to read RMAX, below  
RMAX  r  all  Y  0  The maximum sphere radius (in units of the average WSR) over which neighbors will be included in the generation of structure constants. This takes a default value and is not required input. It is an interesting exercise to see how much the structure constants and eigenvalues change when this radius is increased. 
NEIGHB  i  FP  Y  30  Minimum number of neighbors in cluster 
ENV_MODE  i  all  Y  0  Type of envelope functions: 0 2nd generation 1 SSSW (3rd generation) 2 NMTO 3 SSSW and vallap basis 
ENV_NEL  i  lm, lmstr  Y  (NMTO only) Number of NMTO energies  
ENV_EL  r  lm, lmstr  N  0  SSSW of NMTO energies, in a.u. 
DELRX  r  ASA  Y  3  Range of screened function beyond last site in cluster 
TOLG  r  FP  Y  1e6  Tolerance in l=0 gaussians, which determines their range 
RVL/R  r  all  Y  0.7  Radial cutoff for vallap basis (this is experimental) 
VLFUN  i  all  Y  0  Functions for vallap basis (this is experimental) 0 G0 + G1 1 G0 + Hsm 2 G0 + Hsmdot 
MXNBR  i  ASA  Y  0  Make lmstr allocate enough memory in dimensioning arrays for MXNBR neighbors in the neighbor table. This is rarely needed. 
SHOW  1  lmstr  Y  F  Show strux after generating them 
EQUIV  1  lmstr  Y  F  If true, try to find equivalent neighbor tables, to reduce the computational effort in generating strux. Not generally recommended 
LMAXW  i  lmstr  Y  1  lcutoff for (optional) Watson sphere, used to help localize strux 
DELRW  r  lmstr  Y  0.1  Range extending beyond cluster radius for Watson sphere 
IINV_NIT=  i  lmstr  Y  0  Number of iterations 
IINV_NCUT  i  lmstr  Y  0  Number of sites for inner block 
IINV_TOL  r  lmstr  Y  0  Tolerance in errors 
*IINV parameters govern iterative solutions to screened strux
START
Category START is specific to the ASA. It controls whether the code starts with moments P,Q or potential parameters; also the moments P,Q may be input in this category. It is read by lm, lmgf, lmpg, and tbe.
Token  Arguments  Program  Optional  Default  Explanation 

BEGMOM  i  ASA  Y  1  When true, causes program lm to begin with moments from which potential parameters are generated. If false, the potential parameters are used and the program proceeds directly to the band calculation. 
FREE  1  ASA  Y  F  Is intended to facilitate a selfconsistent freeatom calculation. When FREE is true, the program uses rmax=30 for the sphere radius rather than whatever rmax is passed to it; the boundary conditions at rmax are taken to be value=slope=0 (rmax=30 should be large enough that these boundary conditions are sufficiently close to that of a free atom.); subroutine atscpp does not calculate potential parameters or save anything to disk; and lm terminates after all the atoms have been calculated. 
CNTROL  1  ASA  Y  F  When CONTRL=T, the parser attempts to read the “continuously variable principal quantum numbers” P and moments Q0,Q1,Q2 for each l channel; see P,Q below. 
ATOM  c  ASA  Y  Class label. P,Q (and possibly other data) is given by class. Tokens following a class label and preceding the next class label belong to that class.  
ATOM_P= and ATOM_Q  c  ASA  Y  Read “continuously variable principal quantum numbers” for this class (P=…), or energy moments Q0,Q1,Q2 (Q=…). P consists of one number per l channel, Q of three numbers (Q0,Q1,Q2) for each l. Note In spin polarized calculations, a second set of parameters must follow the first, and the moments should all be half of what they are in nonspin polarized calculations. In this sample input file for Si, P,Q is given as: ATOM=SI P=3.5 3.5 3.5 Q=1 0 0 2 0 0 0 0 0 ATOM=ES P=1.5 2.5 3.5 Q=.5 0 0 .5 0 0 0 0 0 One electron is put in the Si s orbital, 2 in the p and none in the d, while 0.5 electrons are put in the s and p channels for the empty sphere. All first and second moments are zero. This rough guess produces a correspondingly rough potential. You do not have to supply information here for every class; but for classes you do, you must supply all of (P,Q0,Q1,Q2). Data read in START supersedes whatever may have been read from disk. Remarks below provide further information about how P,Q is read and printed.  
RDVES  1  ASA  Y  F  Read Ves(RMT) from the START category along with P,Q 
ATOM_ENU  r  ASA  Y  Linearization energies 
Sample START category
The following is taken for the distribution’s test of La_{2} Cu O_{4}.
START BEGMOM=T CNTROL=T RDVES=t
ATOM=LA P= 6.3055046 6.3000000 5.2308707
Q= 0.4770507 0.0000000 0.0610692
0.9882047 0.3905638 0.2327244
2.0252993 0.0000000 0.1272500
ATOM=CU P= 4.6331214 4.3438861 3.8947075
Q= 0.4910799 0.0000000 0.0974578
0.6087341 0.0000000 0.1140513
9.4164169 0.0000000 0.2018023
ATOM=OX P= 2.8833091 2.8438183 3.1896353
Q= 1.6741779 0.0000000 0.0653497
4.2304006 0.0000000 0.1036699
0.0404676 0.0000000 0.0023966
ATOM=OX2 P= 2.8840328 2.8447249 3.1806967
Q= 1.6660490 0.0000000 0.0257208
4.1318836 0.0000000 0.0365535
0.0083512 0.0000000 0.0003608
Notes on parsing P and Q
In the ASA, knowledge of P and Q is sufficient to completely determine the ASA density. Several ways are available to read these important quantities. The parser returns (P,Q) as a set according the following priorities:

The (P,Q) set is read from the disk, if supplied, (along possibly with other quantities such as potential parameters El, C, Δ, γ.) One file is created for each class that contains this data and other classspecific information. Some or all of the data may be missing from the disk files. Alternatively, you may read these data from a restart file rsta.ext, which if it exists contains data for all classes in one file. The program will not read this data by default; use rs=1 to have it read from the rsta file. To write class data to rsta, use rs=,1 ( must be 0 or 1)

If START_CONTRL=T, (P,Q) (and possibly other quantities) are read from START for classes you supply (usually all classes). Data read from this category supersedes any that might have been read from disk. If class data read from either of these sources, the input system returns it. For classes where none is available the parser will pick a default:

If data from a different class but in the same species is available, use it.

Otherwise use some preset default values for (P,Q).

After a calculation finishes you can run lmctl to read (P,Q) from disk and format it in a form ready to insert into the START category,e.g.
ATOM=SI P= 3.8303101 3.7074067 3.2545634
Q= 1.1694276 0.0000000 0.0297168
1.8803181 0.0000000 0.0489234
0.1742629 0.0000000 0.0063520
ATOM=ES P= 1.4162942 2.2521617 3.1546386
Q= 0.2873686 0.0000000 0.0129888
0.3485430 0.0000000 0.0165416
0.1400664 0.0000000 0.0055459
Thus all the information needed to generate a selfconsistent ASA density can be embedded in the ctrl file.
Because the P’s float to the band centerof gravity (i.e. center of gravity of the occupied states for a particular site and l channel) the corresponding first moments Q_{1} vanish. P’s are floated by default since it minimizes the linearization error.
Caution: Sometimes it is necessary to override this default: If the band CG (of the occupied states) is far removed from the natural CG of a particular channel, you must restrict how far P can be shifted to the band CG. In some cases, allowing P to float completely will result in “ghost bands”.
The highlying Ga 4d state is a classic example. To restrict P to a fixed value, see SPEC_ATOM_IDMOD.
In such cases, you want to pick the fractional part of P to be small, but not so low as to cause problems (about 0.5 for s orbitals and 0.15 for d orbitals; see here).
STRUC
By default structural information is read through the ctrl file. But some of the essential data can be read in multiple ways, in particular from site file. Questaal has utilities that will import this information from other formats such as cif files.
Token  Arguments  Program  Optional  Default  Explanation 

FILE  c  all  Y  Read structural data (ALAT, NBAS, PLAT) from an independent site file. The file structure is documented here; see also this tutorial. Note: EXPRESS_file performs the same function as STRUC_FILE, and supersedes STRUC_FILE if it is present.  
NBAS  i  all  N†  Number of the sites in the primitive unit cell.  
NSPEC  i  all  Y  Number of atom species  
ALAT  r  all  N†  A scaling, in atomic units, of the primitive lattice and basis vectors  
DALAT  r  all  Y  0  is added to ALAT. It can be useful in contexts certain quantities that depend on ALAT are to be kept fixed (e.g. SPEC_ATOM_R/A) while ALAT varies. 
PLAT  r,r,…  all  N†  (dimensionless) primitive translation vectors  
SLAT  r,r,…  lmscell  N  Superlattice vectors  
NL  i  all  Y  3  Sets a global default value for lcutoffs l_{cut} = NL−1. NL is used for both basis set and augmentation cutoffs. 
SHEAR  r,r,r,r  all  Y  Enables shearing of the lattice in a volumeconserving manner. If SHEAR=#1,#2,#3,#4, #1,#2,#3=direction vector; #4=distortion amplitude. Example: SHEAR=0,0,1,0.01 distorts a lattice in initially cubic symmetry to tetragonal symmetry, with 0.01 shear.  
ROT  c  all  Y  Rotates the lattice and basis vectors, and the symmetry group operations by a unitary matrix. Example: ROT=z:pi/4,y:pi/3,z:pi/2 generates a rotation matrix corresponding to the Euler angles α=π/4, β=π/3, γ=π/2. See this document for the general syntax. Lattice and basis vectors, and point group operations (SYMGRP) are all rotated.  
DEFGRD  r,r,…  all  Y  A 3×3 matrix defining a general linear transformation of the lattice vectors.  
STRAIN  r,r,…  all  Y  A sequence of six numbers defining a general distortion of the lattice vectors  
ALPHA  r  all  N  Amount of Voigt strain 
†Information may be obtained from a site file
SYMGRP
Category SYMGRP provides symmetry information; it helps in two ways. First, it provides the relevant information to find which sites are equivalent, this makes for a simpler and more accurate band calculations. Secondly, it reduces the number of kpoints needed in Brillouin zone integrations.
Normally you don’t need SYMGRP; the program is capable of finding its own symmetry operations. However, there are cases where it is useful or even necessary to manually specify them. For example when including spinorbit coupling or noncollinear magnetism where the symmetry group isn’t only specified by the atomic positions. In this case you need to supply extra information.
You can use SYMGRP to explicitly declare a set of generators from which the entire group can be created. For example, the three operations R4X, MX and R3D are sufficient to generate all 48 elements of cubic symmetry.
Unless conditions are set for noncollinear magnetism and/or SO coupling, the inversion is assumed by default as a consequence of timereversal symmetry.
A tag describing a generator for a point group operation has the form O(nx,ny,nz) where O is one of M, I or Rj, or E, for mirror, inversion jfold rotation and identity operation, respectively. nx,ny,nz are a triplet of indices specifying the axis of rotation. You may use X, Y, Z or D as shorthand for (1,0,0), (0,1,0), (0,0,1), and (1,1,1) respectively. You may also enter products of rotations, such as I*R4X.
Thus
SYMGRP R4X MX R3D
specifies three generators (4fold rotation around x, mirror in x, 3fold rotation around (1,1,1)). Generating all possible combinations of these rotations will result in the 48 symmetry operations of the cube.
To suppress all symmetry operations, use
SYMGRP E
In the ASA, owing to the spherical approximation to the potential only the point group is required for selfconsistency.
But in general you must specify the full space group. The translation part gets appended to rotation part in one of the following forms: :(x1,x2,x3) or alternatively ::(p1,p2,p3) with the double ‘::’. The first defines the translation in Cartesian coordinates; the second as fractional multiples of lattice vectors.
These two lines (taken from testing/ctrl.cr3si6) provide equivalent specifications:
SYMGRP r6z:(0,0,0.4778973) r2(1/2,sqrt(3)/2,0)
SYMGRP r6z::(0,0,1/3) r2(1/2,sqrt(3)/2,0)
Keywords in the SYMGRP category
SYMGRP accepts, in addition to symmetry operations the following keywords:

find tells the program to determine its own symmetry operations. Thus:
SYMGRP find
amounts to the same as not incuding a SYMGRP category in the input at all
You can also specify a mix of generators you supply, and tell the program to find any others that might exist. For example:
SYMGRP r4x find
specifies that 4fold rotation be included, and find tells the program to look for any additional symops that might exist.

AFM: For certain antiferromagnets, certain translation operations exist provided the rotation/shift is accompanied by a spin flip. Say a translation of (1/2,1/2,1/2)a restores the crystal structure, but all atoms after translation have opposite spin. Specify this symmetry with:
SYMGRP ... AFM::1/2,1/2,1/2
This operation is used only by lmf.

SOC or SOC=2: Tells the symmetry group generator to exclude operations that do not preserve the z axis. This is used particularly for spinorbit coupling where the crystal symmetry is reduced (z is the quantization axis). SOC=2 is like SOC but allows operations that preserve z or flip z to −z. This works in some cases.
Note: This keyword is only active when the two spin channels are linked, e.g. SO coupling or noncollinear magnetism.

GRP2 turns on a switch that can force the density among inequivalent classes that share a common species to be averaged. In the ASA codes the density is spherical and the averaging is complete; in the FP case only the spherical part of the densities can be averaged. This helps sometimes with stabilizing difficult cases in the path to selfconsistency. You specify which species are to be averaged with the SPEC_ATOM_GRP2 token.
GRP2 averages the input density; GRP2=2 averages the output density; GRP2=3 averages both the input and the output density.

RHOPOS turns on a switch that forces the density positive at all points.
You can also accomplish this with the commandline switch rhopos.
VERS
This category is used for version control. As of version 7, the input file must have the following tokens for any program in the suite:
VERS LM:7
It tells the input system that you have a v7 style input file.
For a particular program you need an additional token to tell the parser that this file is set up for that program. Thus your VERS category should read:
VERS LM:7 ASA:7 for lm, lmgf or lmpg
VERS LM:7 FP:7 for lmf or lmfgwd
VERS LM:7 MOL:3 for a molecules codes such as lmmc
VERS LM:7 TB:9 for the empirical tightbinding tbe
and so on.
Add version control tokens for whatever programs your input file supports.
Notes on gradient corrected functionals
The semilocal exchangecorrelation potential is defined as
$v_{j}^{GGA}(n,\mathbf{r}) = v_{j}^{HEG}(n,\mathbf{r}) * F_{j}^{GGA}(n,\mathbf{r}) + \int d\mathbf{r}' \varepsilon^{HEG}_j(n,\mathbf{r}') \frac{\partial F_{j}^{GGA}(n,\mathbf{r}')}{\partial s(n,\mathbf{r}')} \frac{\partial s(n,\mathbf{r}')}{\partial n(\mathbf{r})} \quad\quad \text{with } j=x,c$where $v_{j}^{HEG}[n](r)$ and $\varepsilon^{HEG}_j[n](r)$ are the exchange or correlation potentials and energy densities, respectively, for the Homogeneous Electron Gas (HEG), which are determined by the options XCFUN=1 or XCFUN=2. $F_{j}^{GGA}[n](r)$ is the exchange or correlation GGA enhancement factor which introduce semilocal effects and can be choosen using the token GGA. $s(n,r)$ is the reduced gradient. Then, the final exchangecorrelation potential/functional is determined using two different tokens, i.e. one for the local (HEG) part and the other one for the semilocal (GGA) correction.
Note that hybrid and metaGGA exchangecorrelation functional schemes are not implemented in the QUESTAAL package. As consequence, only LDA and GGA libXC functionals can be used.
Interactive mode
For the most part Questaal codes are designed to run without receiving information through standard input. The various editors are an exception (though even editor instructions can be run in batch mode; see e.g. dynamical selfenergy editor tutorial).
It is often convenient to have some interactive facility, e.g. whether to limit the number of iterations in a selfconsistency cycle. The Questaal codes have a interactive mode, which you can turn on with token IO_IACTIV in the ctrl file, or on the commandline with the iactive switch.
For example, if you run lm or lmf interactively you will be prompted with
QUERY: beta (def=0.3)?
and the program will wait for input. To see what your options are, enter ? <RET> You should see
(A)bort (I)active (V)erb (C)pu (T)iming (W)ork (S)et value
QUERY: beta (def=0.3)?
Here it is asking if you want to modify the existing value for the charge mixing parameter beta.
Enter one of:
 a Program aborts execution
 i Toggles interactive mode
 v # Sets verbosity to #
 c Prints out CPU usage so far
 t Turns on timing printout
 w Not used now
 s # Sets parameter to #
If you enter
s .5 <RET>
in this instance, the program will modify its value for beta to 0.5 and prompt you again. If you don’t want to make any (further) changes, just enter <RET>.
The most commonly changed parameter is the number of iterations, called maxit. You can increase or decrease it; if you decrease maxit below the current iteration number, the program will stop.
Simulacrum of interactive mode
You can enter interactive mode instructions that are normally read by the standard input by entering them into a file iact.ext.
The executable first looks for that file, reads its contents, and executes its instructions before prompting you. It will perform the instruction, e.g. set the value of the parameter, without turning on the true interactive mode. However if you do turn it on, e.g. put i into iact.ext, the program will revert to true interactive mode and prompt you for instructions.
There is an important difference in the normal and simulacrum operations when setting a parameter. In the latter case, you must tell the program which parameter. Do this by naming the parameter after the s. Thus iact.ext would contain a line like
s maxit 3
Interactive mode with MPI
Normal interactive mode is not available when running with multiple processors. The simulacrum mode does work, but only for a subset of parameters. Most importantly, maxit is one parameter that is read; thus you can adjust the number of iterations a job will do after execution starts.
ITER_MIX
ITER_MIX is a token in the ITER category that controls how Questaal codes iterate to selfconsistency. Its contents are a string consisting of mixing options, described here.
Questaal codes follow the usual procedure of mixing a linear combination of input density n^{in} and output density n^{out} to make a trial guess n^{*} for the selfconsistent density (see for example Chapter 9 in Richard Martin’s book [1]. Questaal uses two independent techniques to accelerate convergence to the selfconsistency condition n^{out}→n^{in}. First, the quantities are mixed making use of model for the dielectric function. Second, multiple (n^{in},n^{out}) pairs (taken from prior iterations) can be used to accelerate convergence. The contents of ITER_MIX control options for both kinds of approaches.
Charge mixing, general considerations
In a perfect mixing scheme, n^{*} would be the selfconsistent density. If the static dielectric response is known, n^{*} can be estimated to linear order in n^{out}−n^{in}. It is not difficult to show that
ε is a function of source and field point coordinates r and r′: ε = ε(r,r′) and in any case it is not given by the standard selfconsistency procedure. The Thomas Fermi approximation provides a reasonable, if rough estimate for ε, which reads in reciprocal space
$\epsilon^{1}(q) = \frac{q^2}{q^2 + k_{TF}^2} \quad\quad (2)$Eq.(2) has one free parameter, the Thomas Fermi wave number k_{TF}. It can be estimated given the total number of electrons qval from the free electron gas formula.
If the density were expanded in plane waves n = Σ_{G} C_{G} n_{G}, a simple mixing scheme would be to mix each C_{G} separately according to Eq.(2). This is called the “Kerker mixing” algorithm. One can use the Lindhard function instead. The idea is similar, but the Lindhard function is exact for free electrons. In any case the Questaal codes do not have a plane wave representation so they do something else.
The ASA uses a simplified mixing scheme since the logarithmic derivative parameters P and energy moments of charge Q for each class is sufficient to completely specify the charge density. The density is not explicitly mixed.
lmf, by contrast, uses a density consisting of three parts: a smooth density n_{0} carried on a uniform mesh, defined everywhere in space and two local densities: the true density n_{1} and a onecenter expansion n_{2} of the smooth density The mixing algorithm must mix all of them and it is somewhat involved. See fp/mixrho.f for details.
The mixing process reduces to estimating a vector X^{*} related to the density (e.g. P,Q in the ASA) where δX = X^{out} − X^{in} vanishes at X^{in} = X^{*}.
Mixing algorithms mix linear combinations of (X^{in},X^{out}) pairs taken from the current iteration together with pairs from prior iterations. If there are no prior iterations, then
It is evident from Eq.(1) that beta should be connected with the dielectric function. However, beta is just a number. If beta=1, X^{*} = X^{out}; if beta→0, X^{*} scarcely changes from X^{in}. Thus in that case you move like an “amoeba” downhill towards the selfconsistent solution. For small systems it is usually sufficient to take beta on the order of, but smaller than unity. For large systems charge sloshing becomes a problem so you have to do something different. This is because the potential change goes as δV ~ G^{−2}×δn so small G components of δn determine the rate of mixing. The simplest (but inefficient) choice is to make beta small.
The beauty of the Kerker mixing scheme is that charges in small G components of the density get damped out, while the shortranged, large G components do not. An alternative is to use an estimate ε for the dielectric function. Construct δn = ε^{−1} (n^{out}−n^{in}) and build δX from δn. Then estimate
Now beta can be much larger again, of order unity.
lmf uses a Lindhard function for the uniform mesh density (similar to Thomas Fermi screening; only the Lindhard function is the actual dielectric function for the free electron gas) and attempts to compensate for the contribution from local densities in an approximate way.
The ASA codes (lm, lmgf, lmpg) offer two options:
 A rough ε is obtained from eigenvalues of the Madelung matrix (OPTIONS_SCR=4).
 The q=0 discretized polarization at q=0 is explicitly calculated (see OPTIONS_SCR).
There is some overhead associated with the second option, but it is not too large and having it greatly facilitates convergence in large systems. This is particularly important in magnetic metals, where there are lowenergy degrees of freedom associated with the magnetic parts that require large beta.
The ITER_MIX tag and how to use it
Mixing proceeds through (X^{in},X^{out}) pairs taken from the current iteration together with pairs from prior iterations. As noted in the previous section it is generally better to mix δX than δX; but the mixing scheme works for either.
You can choose between Broyden 3 and Anderson 2 methods. The string belonging to ITER_MIX should begin with one of
MIX=An MIX=Bn
which tells the mixer which scheme to use. slatsm/amix.f describes the mathematics behind the Anderson 2 scheme.
n is the maximum number of prior iterations to include in the mix. As programs proceed to selfconsistency, they dump prior iterations to disk, to read them the next time through. Data is I/O to mixm.ext.
The Anderson scheme 2 is particularly simple to monitor. How much of δX from prior iterations is included in the final mixed vector is printed to stdout as parameter tj, e.g.
tj: 0.47741 ← iteration 2 tj:0.39609 0.44764 ← iteration 3 tj:0.05454 0.01980 ← iteration 4 tj: 0.24975 tj: 0.48650 tj:1.34689
In the second iteration, one prior iteration was mixed; in the third and fourth, two; and after that, only one. (When the normal matrix picks up a small eigenvalue the Anderson mixing 2 algorithm reduces the number of prior iterations).
Consider the case when a single prior iteration was mixed.
 If tj=0, the new X is entirely composed of the current iteration. This means selfconsistency is proceeding in an optimal manner.
 If tj=1, it means that the new X is composed 100% of the prior iteration. This means that the algorithm doesn’t like how the mixing is proceeding, and is discarding the current iteration. If you see successive iterations where tj is close to (or worse, larger than) unity, you should change something, e.g. reduce beta.
 If tj<0, the algorithm thinks you can mix more of X^{out} and less of X^{in}. If you see successive iterations where tj is significantly negative (less than −1), increase beta.
Broyden mixing 3 uses a more sophisticated procedure, in which it tries to build up the Hessian matrix. It usually works better but has more pitfalls than Anderson. Broyden has an additional parameter, wc, that controls how much weight is given to prior iterations in the mix (see below).
The general syntax is for ITER_MIX is
An[,b=beta][,b2=b2][,bv=betv][,n=nit][,w=w1,w2][,fn=name][,k=nkill][,elind=#][;...] or Bn[,b=beta][,b2=b2][,bv=betv][,wc=wc][,n=nit][,w=w1,w2][,fn=name][,elind=#][,k=nkill]
The options are described below. They are parsed in routine subs/parmxp.f. Parameters (b, wc, etc.) may occur in any order.:

An or Bn: maximum number of prior iterations to include in the mix (the mixing file may contain more than n prior iterations).
n=0 implies linear mixing. Default: B2. 
b=beta: the mixing parameter beta in Eq. 4 above. Default: 0.3.

n=nit: the number of iterations to use mix with this set of parameters before passing on to the next set. After the last set is exhausted, it starts over with the first set.

name=fn: mixing file name (mixm is the default). Must be eight characters or fewer.

k=nkill: kill mixing file after nkill iterations. This is helpful when the mixing runs out of steam, or when the mixing parameters change. Default: 7.

wc=wc: (Broyden only) that controls how much weight is given to prior iterations in estimating the Jacobian. wc=1 is fairly conservative. Choosing wc<0 assigns a floating value to the actual wc, proportional to −wc/rmserror. This increases wc as the error becomes small. wc defaults to −1 if it is not specified. See Johnson’s paper 3 for the definition of wc.

w=w1,w2: (spinpolarized calculations only) The up and down spin channels are mixed independently. Instead the sum (up+down) and difference (updown) are mixed. The two combinations are weighted by w1 and w2 in the mixing, more heavily emphasizing the more heavily weighted. As special cases, w1=0 freezes the charge and mixes the magnetic moments only while w2=0 freezes the moments and mixes the charge only.

elind=elind: The Fermi energy entering into the Lindhard dielectric function: $\mathbf{elind} = k_F^2$.
elind<0: Use the freeelectron gas value, scaled by −elind. The default value is −1. 
wa: (ASA only) weight for extra quantities included with P,Q in the mixing procedure. For noncollinear magnetism, includes the Euler angles.

locm: (FP only) not documented yet.

r=expr: continue this block of mixing sequence until rms error < expr.
Example: MIX=A4,b=.2,k=4 uses the Anderson method 2, killing the mixing file each fourth iteration. The mixing beta** is 0.2.
You can string together several rules. One set of rules applies for a certain number of iterations; followed by another set.
Rules are separated by a “ ; ”.
Example: MIX=B10,n=8,w=2,1,fn=mxm,wc=11,k=4;A2,b=1
does 8 iterations of Broyden mixing, followed by Anderson mixing. The Broyden iterations weight the (up+down) double that of (updown) for the magnetic case, and iterations are saved in a file which is deleted at the end of every fourth iteration. wc is 11. beta assumes the default value. The Anderson rules mix two prior iterations with beta=1.
References
[1] R. M. Martin, Electronic Structure, Cambridge University Press (2004).
[2] D. G. Anderson. Iterative procedures for nonlinear integral equations. J. Assoc. Comput. Mach., 12:547–560, 1965
[3] D. D. Johnson, Phys. Rev. B 38, 12807 (1988).
Edit This Page