lp:~nickpapior/siesta/mixing
- Get this branch:
- bzr branch lp:~nickpapior/siesta/mixing
Branch merges
Branch information
Recent revisions
- 600. By Nick Papior
-
Moved m_mixing* routines to mixing*
It makes it easier to work with if files are not
named m_ because then all files are sorted similarly. - 599. By Nick Papior
-
Cleaned mixing calls in mixer to remove ambiguity
- No logic change in the execution
- 598. By Nick Papior
-
Added total charge mixing
- This is simply only taking into account the total
charge when calculating the residual norm. - 596. By Nick Papior
-
Updated mixing routines with bug-fixes and new external methods
- Fixed a few bugs in the mixing routines if one
requests specific mixing options. i.e. the
restart count was always specified and the
history step was erroneously changed.- Enabled external function in mixing routine to
attach a mixing option externally controlled.
This enables one to assign custom external options
related to the parameters etc. One may use this
external function to decide on the norm-function.- Removed spin-components mixing, these are now controlled
via other means.
For instance one can now mix the mulliken charge (DM*S)
the vector components (M(x), M(y), M(z)) which maybe is
more stable than the other ones.- Cleaned up MPI operations in mixing routines
- Updated documentation.
- 595. By Nick Papior
-
Removed DM*H and mixing options and added H*S
Updated mixing options for more streamlined options.
Added H*S option which seems to sometimes behave better than
DM and regular H mixing. - 594. By Nick Papior
-
Streamlined naming convention for mixing variables
- Now all mixing variables are named mix_*
- 593. By Nick Papior
-
Added DM*H mixing by using the DM*H norm for the residual
- The DM*H mixing allows two different methods:
SCF.Mix DM*H
or
SCF.Mix H*DM
where DM*H mixes the density matrix and H*DM mixes
the Hamiltonian.They seem to behave equivalently to their respective
mixing method, H*DM ~ Hamiltonian and DM*H ~ DM.These will not be documented currently because their
use is probably not encouraged. One of the shortcomings
is that the corresponding DM/H for the multiplication is
not from the equivalent cycle, meaning that only the latest
version is used.The idea of this method is to "mix the band-structure energy
residual" in an approximate way.- Moved residual.F90 to mix_norm.F90 which better reflects the
code intent.
Branch metadata
- Branch format:
- Branch format 7
- Repository format:
- Bazaar repository format 2a (needs bzr 1.16 or later)
- Stacked on:
- lp:siesta