mirror of https://gitlab.com/QEF/q-e.git
A document about the INPUT of davidson is added.
git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@10317 c92efa57-630b-4861-b058-cf58834340f0
This commit is contained in:
parent
48a47195cc
commit
8d2653971a
|
@ -0,0 +1,80 @@
|
|||
# Descriptions for the input of turbo_davidson.x
|
||||
# Created by Xiaochaun Ge (Jun. 2013)
|
||||
|
||||
=====================================
|
||||
num_eign Default: 1
|
||||
num_init Default: 2
|
||||
|
||||
Here you indicate the number of eigenstates that you would like to calculate at once.
|
||||
|
||||
=====================================
|
||||
num_basis_max Default: 20
|
||||
|
||||
The maximum number of basis that you allow for the sub basis set. One should notice
|
||||
that the memory requirement of davidson algorithm is mainly determined by this variable.
|
||||
An estimation of the memory is reported at the first of the run.
|
||||
|
||||
=====================================
|
||||
residue_conv_thr Default: 1.0E-4
|
||||
|
||||
The convergence level of the residue.
|
||||
|
||||
|
||||
=====================================
|
||||
precondition Default: .true.
|
||||
|
||||
If or not use the precondition. At this moment, one sees no reason not to use precondition.
|
||||
|
||||
|
||||
=====================================
|
||||
single_pole Default: .false.
|
||||
|
||||
A slightly better way to set the initial trial vectors. But the improvement is really small
|
||||
and this flag at this moment is comparable with only LDA/PBE + NC PPs. So don't use it unless
|
||||
it's really necessary.
|
||||
|
||||
|
||||
=====================================
|
||||
reference Default: 0 (Unit: Ry)
|
||||
|
||||
One of the highlight variable. Use it to constrain the algorithm converge to these eigenstates
|
||||
having energy close to the reference. So one can calculate less eigenstates at once and do multiple
|
||||
calculations with different reference
|
||||
|
||||
|
||||
=====================================
|
||||
broadening=0.005
|
||||
start=0.0d0
|
||||
finish=1.0d0
|
||||
step=0.01d0
|
||||
|
||||
Flags for ploting the spectrum. Meanings of them shoulbd be self-evident. All units are "Ry"
|
||||
|
||||
=====================================
|
||||
if_random_init=.false.
|
||||
|
||||
Using preconditional random vectors as the trail vectors. Used only in special cases that you
|
||||
do hate, and I don't see why, to calculate a few virtual states(Even the precision doesn't need
|
||||
to be high). Using bad trial vectors should cause only slow convergence, and don't affect the final
|
||||
resulta.
|
||||
|
||||
|
||||
=====================================
|
||||
p_nbnd_occ=10
|
||||
p_nbnd_virt=10
|
||||
|
||||
They are useful when there're too many occ/virt states pairs, and you care of only a few of them.
|
||||
p_nbnd_occ and p_nbnd_virt indicate how many occ and virt states you would like to look at.
|
||||
In priciple they affect only the interpretation of the eigenstates, but no influence to their energy
|
||||
and the spectrum.
|
||||
Make sure that min(p_nbnd_occ,nbnd_occ)*min(p_nbnd_virt,nbnd_virt) is lager than the number of initial
|
||||
vectors, so you won't end up using rendom trail vectors which will slow down your convergence.
|
||||
|
||||
|
||||
=====================================
|
||||
poor_of_ram=.false.
|
||||
|
||||
When this set to true, you double the memory used for USPP calculation, but you get rid of applying many
|
||||
times of s_psi and cal_bec calculation, which takes a lot of time (sometimes more than half of the whole
|
||||
calculation) when the size of sub space is more than 100.
|
||||
|
|
@ -17,5 +17,6 @@ CH4-PR: Tests charge response algorithm
|
|||
SiH4: Tests Ultrasoft PP implementation
|
||||
CH4-BOND: This is an example that shows visual analysis of a softened bond in a set of degenerate bonds.
|
||||
tools: Tools to compare data.
|
||||
CH4-DAV: Example of using davidson algorithm
|
||||
|
||||
|
||||
|
|
Loading…
Reference in New Issue