next up previous contents
Next: Execution Up: Programs for running an Previous: Input

    
MIXER (adding and mixing of charge densities)

In mixer the electron densities of core, semi-core, and valence states are added to yield the total new (output) density (in some calculations only one or two types will exist). Proper normalization of the densities is checked and enforced (by adding a constant charge density in the interstitial). As it is well known, simply taking the new densities leads to instabilities in the iterative SCF process. It is thus necessary to stabilize the SCF cycle. In WIEN97 this is done by mixing the output density with the (old) input density to obtain the new density to be used in the next iteration. Two mixing schemes are implemented:

1.
straight mixing as originally proposed by Pratt (52) with a mixing factor Q

\begin{displaymath}\rho_{new}(r)=(1-Q)\rho_{old}(r)+Q\rho_{output}(r)\end{displaymath}

2.
the Broyden-II mixing scheme (Singh et al., 86), in which all the expansion coefficients of the density from several preceding iterations are utilized to calculate an optimal mixing fraction for each coefficient in each iteration.

At the outset of a new calculation (for any changed computational parameter such as k-mesh, matrix size, lattice constant etc.), any existing case.broydX files must be deleted (since the iterative history which they contain refers to a ``different`` incompatible calculation). In addition, in some cases better convergence can be achieved, if these files are removed every 15-20 iterations. Usually the Broyden scheme is much better than Pratt's scheme and thus is recommended.

After modifications to the case.struct file (lattice parameters, atomic positions) a run with mixing factor 0.0 can be used to renormalize the case.clmsum_old file from the previous case.

Note: a case.clmval file must also be present, since the LM values and the K-vectors are read from this file.

If the file case.clmsum_old can not be found by MIXER, a ``PRATT-mixing`` with mixing factor 1.0 is done.

The total energy and the atomic forces are computed in mixer by reading the case.scf file and adding the various contributions computed in preceding steps of the same iteration.



 
next up previous contents
Next: Execution Up: Programs for running an Previous: Input

2000-04-11