TABLE OF CONTENTS
INTRODUCTION
&LR_INPUT
prefix | outdir | wfcdir | max_seconds | restart | lr_verbosity | disk_io
&LR_DAV
num_eign | num_init | if_random_init | num_basis_max | residue_conv_thr | precondition | single_pole | if_dft_spectrum | reference | broadening | start | finish | step | p_nbnd_occ | p_nbnd_virt | poor_of_ram | poor_of_ram2 | max_iter | no_hxc | pseudo_hermitian | ltammd | lplot_drho | d0psi_rs | lshift_d0psi
INTRODUCTION
Input data format: { } = optional, [ ] = it depends.
All quantities whose dimensions are not explicitly specified are in
RYDBERG ATOMIC UNITS
BEWARE: TABS, DOS <CR><LF> CHARACTERS ARE POTENTIAL SOURCES OF TROUBLE
Comment lines in namelists can be introduced by a "!", exactly as in
fortran code. Comments lines in ``cards'' can be introduced by
either a "!" or a "#" character in the first position of a line.
Structure of the input data:
===============================================================================
&LR_INPUT
...
/
&LR_DAV
...
/
Namelist: &LR_INPUT
|
This namelist is always needed !
prefix |
CHARACTER |
Default: |
'pwscf'
|
Sets the prefix for generated and read files. The files
generated by the ground state pw.x run should have this
same prefix.
|
outdir |
CHARACTER |
Default: |
'./'
|
The directory that contains the run critical files, which
include the files generated by ground state pw.x run.
|
wfcdir |
CHARACTER |
Default: |
'./'
|
The directory that contains the run critical files, which
include the files generated by ground state pw.x run.
|
max_seconds |
REAL |
Default: |
1.D+7, or 150 days, i.e. no time limit
|
jobs stops after max_seconds CPU time. Use this option
in conjunction with option restart if you need to
split a job too long to complete into shorter jobs that
fit into your batch queues.
|
restart |
LOGICAL |
Default: |
.false.
|
When set to .true., turbo_davidson.x will attempt to restart
from a previous interrupted calculation if max_seconds
was specified.
Beware, if set to .false. turbo_davidson.x will OVERWRITE any
previous runs.
|
lr_verbosity |
INTEGER |
Default: |
1
|
This integer variable controls the amount of information
written to standard output.
|
disk_io |
CHARACTER |
Default: |
'default'
|
Fine control of disk usage. Currently only 'reduced' is
supported where no restart files are written, apart from
the 'default' mode.
|
|
|
Namelist: &LR_DAV
|
num_eign |
INTEGER |
Default: |
1
|
Number of eigenstates to be calculated.
|
num_init |
INTEGER |
Default: |
2
|
Number of trial vectors. Usually it is twice as large as
the number of eigenstates to be calculated (see num_eign).
|
if_random_init |
LOGICAL |
Default: |
.false.
|
When set to .true. trial vectors are chosen randomly, otherwise
they are guessed from the ground-state calculation.
If p_nbnd_occ * p_nbnd_virt < num_init, this term
is forced to be .true. The usage of random trial vectors should
cause only a slower convergence, and do not affect the final results.
|
num_basis_max |
INTEGER |
Default: |
20
|
Maximum number of basis vectors allowed in the subspace.
When this number is reached, a discharging routine is called.
The memory requirement of the Davidson algorithm is mainly
determined by this variable (an estimation of the memory
is reported at the beginning of the run).
|
residue_conv_thr |
REAL |
Default: |
1.0E-4
|
Threshold for the convergence. When the square of the
residue is smaller than this value, the convergence
is achieved.
|
precondition |
LOGICAL |
Default: |
.true.
|
If set to .true. a precondition is used. At this moment,
one sees no reason why not to use the precondition.
|
single_pole |
LOGICAL |
Default: |
.false.
|
A slightly better way to set the initial trial vectors,
but the improvement is really small. Currently this flag
can be used only with LDA/PBE + NC PPs. So do not use
it unless it is really necessary.
|
if_dft_spectrum |
LOGICAL |
Default: |
.false.
|
When set to .true. the independent particle approximation
is used, i.e. the Hartree and exchange-correlation response
contributions are neglected.
|
reference |
REAL |
Default: |
0.0d0
|
Reference energy in units of Ry. This variable is used
to constrain the Davidson algorithm to converge the eigenstates
having the energy closest to the reference energy. In this way
one can calculate less eigenstates at once, and to perform multiple
calculations with different reference energies (the post-processing
code tddfpt_calculate_spectrum.x can be used for this purpose).
|
broadening |
REAL |
Default: |
0.005d0
|
Lorentzian broadening (Ry) to broaden the absorption spectrum.
|
start |
REAL |
Default: |
0.0d0
|
The lower limit of the energy (Ry) scale for the spectrum calculation.
|
finish |
REAL |
Default: |
1.0d0
|
The upper limit of the energy (Ry) scale for the spectrum calculation.
|
step |
REAL |
Default: |
0.001d0
|
Energy step (Ry) for the spectrum calculation.
|
p_nbnd_occ |
INTEGER |
Default: |
10
|
Number of occupied states selected from the total number
of occupied states computed by PWscf. This variable is
useful if there are too many occupied states but your
are interested in only some of them.
In priciple this variable and p_nbnd_virt affect only
the interpretation of the eigenstates, but do not effect
their energy and the final absorption spectrum.
Make sure that min(p_nbnd_occ,nbnd_occ)*min(p_nbnd_virt,nbnd_virt)
is lager than the number of initial vectors (num_init),
so you will not end up using random trial vectors which would
slow down the convergence.
|
p_nbnd_virt |
INTEGER |
Default: |
10
|
Number of empty states selected from the total number
of empty states computed by PWscf. This variable is
useful if there are too many empty states but your
are interested in only some of them.
In priciple this variable and p_nbnd_occ affect only
the interpretation of the eigenstates, but do not effect
their energy and the final absorption spectrum.
Make sure that min(p_nbnd_occ,nbnd_occ)*min(p_nbnd_virt,nbnd_virt)
is lager than the number of initial vectors (num_init),
so you will not end up using random trial vectors which would
slow down the convergence.
|
poor_of_ram |
LOGICAL |
Default: |
.false.
|
Use this variable if you do not have enough RAM (only USPP),
i.e. set it to .true. When this variable is set to .false.,
you double the memory used for the USPP calculation, but you
increase a speed of the calculation by getting rid of
applying many times of s_psi and cal_bec in the
calculation, which takes a lot of time (sometimes more than
a half of the whole calculation) when the size of the
subspace is more than 100.
|
poor_of_ram2 |
LOGICAL |
Default: |
.false.
|
Use this variable if you do not have enough RAM (NCPP and USPP),
i.e. set it to .true. When this variable is set to .false.,
you double the memory used for the calculation, but you
increase a speed of the calculation by storing D_ and C_
basis: the calculation will be speeded up a lot when
one is calculating many transitions at the same time.
|
max_iter |
INTEGER |
Default: |
100
|
Maximum number of Davidson iterations allowed. When the
number of iterations arrives this number, the calculation
will stop even if the convergence has not been achieved.
|
no_hxc |
LOGICAL |
Default: |
.false.
|
When set to .true. the change in the internal field
(Hartree and exchange-correlation) is ignored in the
calculation, resulting in the independent electron
approximation.
|
pseudo_hermitian |
LOGICAL |
Default: |
.true.
|
When set to .true. the pseudo-Hermitian Lanczos
algorithm is used. When set to .false. the
non-Hermitian Lanczos biorthogonalization algorithm
is used (which is two times slower).
|
ltammd |
LOGICAL |
Default: |
.false.
|
When set to .true. the Tamm-Dancoff approximation is used
in constructing the Liouvillian.
|
lplot_drho |
LOGICAL |
Default: |
.false.
|
When set to .true. the turbo_davidson.x code will write
files for each eigenstate "drho-of-eign-$i" which are
needed to plot the response charge-density at each resonance.
This implies a calculation using the pp.x post-processing
program with the corresponding input file which must be
prepared. See example "H2O-PLOTRHO".
|
d0psi_rs |
LOGICAL |
Default: |
.false.
|
When set to .true. the dipole is computed in the
real space. When set to .false. the dipole is
computed in the reciprocal space by computing [H,r].
Note, currently the commutator does not contain
a contribution for hybrids [V_EXX,r]. See also
the variable lshift_d0psi.
Important: Treatment of the dipole in the real space
is allowed only if the system is finite.
|
lshift_d0psi |
LOGICAL |
Default: |
.true.
|
This variable is used only when d0psi_rs = .true.
a) If a molecule is placed in the corner of the
supercell, there is a discontinuity problem for the
position operator r, which is not periodic. By setting
lshift_d0psi = .true. the discontinuity problem is
solved by shifting the position operator r such that
it is continuous and well defined.
b) If a molecule is placed in the center of the supercell,
there is no discontinuity problem for the position operator r,
and thus you can set lshift_d0psi = .false. But if you still
set it to .true., this will not harm, because the position
operator will basically remain as it is, since it is always
centered wrt the center of the molecule.
|
|
|
|