SCF

From FEFF

Useful

SCF rfms1 [lfms1 nscmt ca nmix]

This card controls feff’s automated self-consistent potential calculations. Thus all fields except rfms1 are optional. If this card is not specified then all calculations are done with the non-self-consistent (overlapped atomic) potential. By default lfms1=0, nscmt=30, ca=0.2, and nmix=1.

rfms1 
This specifies the radius of the cluster for full multiple scattering during the self-consistency loop. Typically one needs about 30 atoms within the sphere specified by rfms1. Usually this value is smaller than the value rfms used in the FMS card, but should be larger than the radius of the second coordination shell.
lfms1 
The default value 0 is appropriate for solids; in this case the sphere defined by rfms1 is located on the atom for which the density of states is calculated. The value 1 is appropriate for molecular calculations and will probably save computation time, but may lead to inaccurate potentials for solids. When lfms1 = 1 the center of the sphere is located on the absorbing atom.
nscmt 
This is the maximum number of iterations the potential will be recalculated. A value of 0 leads to non-self-consistent potentials and Fermi energy estimates. A value of 1 also yields non-self-consistent potentials but the Fermi energy is estimated more reliably from calculations of the LDOS. Otherwise, the value of nscmt sets an upper bound on the number of iterations in the self-consistency loop. Usually self-consistency is reached in about 10 iterations. The default for nscmt is 30 as of FEFF84. The max allowed value of nscmt (given by code parameter nbr) is also 30.
ca 
The convergence accelerator factor. This is needed only for the first iteration, since feff uses the Broyden algorithm to reach self-consistency. A typical value is 0.2; however, you may want to try smaller values if there are problems with convergence. After a new density is calculated from a new Fermi level, the density after the first iteration is \rho_{next}=\mbox{ca}\cdot\rho_{new}+(1-\mbox{ca})\cdot\rho_{old}. ca = 1.0 is extremely unstable and should not be used. The default is 0.2.
nmix 
The nmix specifies how many iterations to do with the mixing algorithm, before starting the Broyden algorithm. The calculations of SCF in materials which contain f-elements may not converge. We encountered such a problem for Pu. However, the SCF procedure converged if we started the Broyden algorithm after 10 iterations with a mixing algorithm with ca=0.05. nmix must be between 1 and 30; a value outside of this range will be ignored, and replaced with an acceptable value. The default value for nmix is 1.
other 
The SCF card also takes two other (obsolete?) arguments which are specified after nmix: ecv, which defaults to -40 and must not be positive, and icoul, which defaults to 0. The default values for these arguments are most likely quite appropriate. The argument ecv specifies the core-valence cut-off (in eV below the Fermi level) and is generally appropriate unless there just happens to be density of states at the specified location. If there is DOS at ecv one may simply increase or decrease ecv a small amount to avoid the DOS which generally has the form of discrete peaks this far below the Fermi level.
* Automated FMS SCF potentials for a molecule of radius 3.1 Angstroms
SCF 3.1 1
* To reach SCF for f-elements and UNFREEZEF we sometimes had to use
SCF 3.7 0 30 0.05 10
developer's resources