VESTA claims to support XSF but it never worked for me.
http://www.geocities.jp/kmo_mma/crystal/en/vesta.html
P.S. XCrysDen is hard to tune but it IS possible to create excellent
quality graphics in it.
Sarah,
I think the problem comes from * instead of -1 in .pdos file.
Try to global replace * for -1 and run those programs again.
Sorry for a late reply.
Qatom of boron is 3.826,is it wrong?
There is nothing wrong here. You should not get integer charges if this
is what disturbs you.
The algorithm of division of charge density into basins belonging to
different atoms can vary. Mulliken is based on counting the
Dear Sarah,
what properties are you interested in, exactly?
What property depends on the amount of atoms?
I see no reason to divide by 2 in your case anyhow.
Because 0.75+1.25=2 e per 2 H atoms
all bases above DZP you have to specify manually with the
%block PAO.Basis
Here is an example of TZTP from recent post in this archive
%block PAO.Basis
Fe 3
n=40 3 P 3
8.0 0. 0.
n=41 3 P 3
8.0 0. 0.
n=32 3 P 3
8.0 0. 0.
Dear SIESTA users,
I'm getting an error when running Denchar:
Can't read MPIRUN_HOST
Do I need to run it in parallel or I just have a compilation problem?
Sincerely,
Alexander
DRHO= RHO_system - sum_RHO_free_atoms
If I'm not mistaken * appears where m should be -1,
i.e. for p orbitals the quantum number m changes as -1,0,1 and it has
nothing to do with spin - basically that are the populations of the
basis orbitals used in your calculations.
To get the meaningful PDOS you have to sum over all m and over
Also the gap found from the DOS plot is about 4eV, which is also
way too large.
DOS is not a good way of looking at Eg since it is based on several
k-points used for calculations.
%block PS.lmax
In 3
As 2
%endblock PS.lmax
Don't cut it. If you have ghost states you might avoid them
1. Regarding the energy of the free atom, as I read the archive, it
seems that we have to consider the ghost atoms surrounding the main atom
to get the enrgy of the free atom?! or, it is enough to have the right
basis of the atom and calculate its energy?
It is very desirable to include
Ruslan,
recently I found a paper Phys. Rev. Lett. 94 (2005) 056103
there was also a more extended follow up to this paper
http://arxiv.org/ftp/cond-mat/papers/0604/0604540.pdf
However, as far as I'm aware this code is not yet open to the public.
Same functionality could be obtained from the
For this scenario, do you think it is best if I switch my coordinates
such that the x is along the length of the wire? Will then, all the
blocks be equivalent?
As far as I remember (please, recheck the manual) the unit cell is not
partitioned in X, then partitioned into ProcessorY in Y and
It is OK to see the output of p orbital for H since this is the orbital
that you used in basis (DZP) i.e. 2 zetas for S and 1 P polarization
orbital.
If you plot it, you will see that it's DOS would be very close to 0
saying that this orbital is indeed practically unpopulated.
n l m z
Dear Victor,
I knew that one can make more than one polarized orbital but didn't
think that one might ever need to generate three polarized orbitals per
each angular momentum.
How do you generate them?
Could you show your basis block to make things more clear?
Sincerely,
Alexander
What is DZP-DZP for Fe (...)
Double zeta plus polarization for both 3d and 4s.
The polarization orbital (only one) is applied atop of other orbitals in
the basis, i.e. you cannot have P for both s and d. That's why DZP-DZP
sounds weird.
other complicated things. How to take into
What do
you think if I just choose the same energy shift for carbon atom
as the one used in diamond solid system calculation?
I think that if you'll take a single atom surrounded by ghost atoms, all
having the same basis as in bulk, that would give you the total energy
lower (i.e. more
This is an electronic temperature used for Fermi-Dirac smearing of the
electron population. Since you have big gaps between filled and empty
energy levels in atoms, molecules and semiconductors - the electronic
temperature would not affect your results since Fermi level is in the
mid of the
I did like that:
E=E(Fe-bulk)-E(Fe-isolated)
and I obtained 4.80 eV
This doesn't mean anything yet. It's the same as fitting pseudopotential
to obtain better lattice constant.
So I think
that in this case the BSSE correction is negligible.
Try it before saying it. The single atom vs atom
I would like to know how to calculate the bulk cohesion energy of a
crystalline solid (in my case, Fe) with Siesta. I guess it is the
difference between the total energies of the atom in the bulk and the
isolated atom.
Yes. But you have to do it with identical grid.
And also don't forget
Ok, but, see, the bcc Fe unit (primitive) cell has only one atom. If I
include ghost atoms in the calculation of the isolated atom, I will have
a basis set in this case larger than that used for the bulk calculation.
You have to take a unit cell of ~10A for an isolated atom,
carefully adjust
You do cell relaxation together with relaxations within the unit cell.
Don't ever do that at the same time.
You have AtomicCoordinatesFormat Ang
i.e. not expressed relatively to unit cell.
On place of SIESTA I wouldn't just know how to fix the atomic
coordinates but to change the unit cell.
You can also natively compile SIESTA under Windows (i.e. without cygwin)
although you would have to recreate all the dependencies in the project
(i.e. manually recreate what is written in the makefile).
I did this in Visual Studio with Intel Fortran, PGI now offers free
Fortran with GUI for
DFT-D is not implemented in SIESTA.
For most weakly bound systems LDA can give a good idea about what could
you expect with vdW (LDA even overestimates the binding).
In simple cases you can add vdW from empirical forcefields, before or
after actual DFT calculations.
What is your system of
David is correct, you have to decrease the electronic temperature.
If you encounter any difficulties send me a message.
Thanks, I've changed to 50K and that helped indeed.
But this is probably still too much of a broadening if I want to select
the single levels with 30 meV spacing.
While I
Hi,
I'm trying to study the small nanocrystal and need to visualize separate
energy levels of it, which are spaced by ~50meV (only Gamma point is
used obviously).
Here is a part from EIG file:
-4.477
-4.39768
-4.35551
-4.33129
here is the gap
-3.83119
-3.80327
-3.76139
-3.72478
The problem
pe Gallium
tm2 2.1
Ga pbr
0.0 0.0 0.0 0.0 0.0 0.0
64
40 2.00 0.00
41 1.00 0.00
42 0.00 0.00
43 0.00 0.00
2.10 2.50 3.00 3.00 1.10
The requirements for the forces are also much higher when doing
calculations in frozen phonons approximation.
So set the relaxation to lets say 0.005 eV/A and also don't take too big
deviations when calculating force matrix. The deviations should be as
small as possible liek less than 0.1 Bohr.
Your compilation options for rho2xsf should be exactly the same as for
SIESTA.
Particularly this relates to the use of single or double precision,
otherwise rho2xsf cannot decode correctly the binary format of RHO.
i.e. it just looks for the values in the wrong place (that's what you
see - it
You have to vary the basis (for example H) and see how it affects the
total energy of the system of interest (e.g. C6H6).
For isolated atom, the longer the basis the better, which is not the
case for a complex system where overlap with other bases may occur.
Longer basis is also required for
Indium has 3d electrons, so your pseudo is quite hard and the energy
cutoff should be quite high even with LDA. Did you check the convergence?
thank you. However, in this case, I think we will have some undesirable
contribution to the formation energy thanks to the basis set
superposition error, don't you think so?
You have to calculate E(63) with the ghost atom in place of the vacancy.
P.S. E(64)/64 should be equal (or very close)
Ghost orbitals can (and should) contain some charge on them, that's why
you add them.
Although charge of 6 is definitely too much.
Did you try to check the RHO itself - do you see any changes for +8 vs
-8 case?
Check in the output whether SIESTA indeed made the calculations with -8.
I did
Basis is created based on the shape of pseudo-wavefunctions derived from
pp calculation.
So, yes, you should use the same pseudopotential for ghost as for normal
atom (just make a copy of the O.vps and call it O_Ghost.vps (normally I
use symbolic links for all pseudopotential files))
Hi,
I modified the code a little but to do this.
You'll get the idea here
http://voznyy.elinity.com/blog/2008/01/bader-analysis-with-siesta/
Some code is there, but it's not final. I'll be glad to finish it with
someone's help.
Michel Sassi wrote:
Hi everybody,
I would like to extract the
I would say that for LDA overestimation of lattice constant is not OK.
(Although I cannot guarantee that this is the case for SmTe)
Normally, LDA lattice constant should be 1% smaller than expt, while in
GGA up to 2% overestimation is OK.
Otherwise, try other set of parameters.
(1) Is there a way of calculating then plotting the transferred charge
using denchar?
I 'm not sure about Denchar, but you can do it with XCrysDen and
Andrei's tools.
You have to create 3 files for molecule, surface and molecule on
surface. Then you can manually combine them into 1 file
Did you try to converge the bulk Ni?
Your MixingWeight is quite small, it looks to me that you have problems
with your k-gird.
If k-grid is dense enough, the only thing to try is to increase the
electronic temperature or use MP mixing scheme.
I think DZ is not the best choice, try DZP.
Hi,
as far as I understand SIESTA can compute the dielectric function directly.
But for phenomenological analysis you don't really need SIESTA itself.
I believe you meant flat-band analysis (based on the amount of search
results in Google I suppose fat-band is just a mistype found in
several
At one of the web presentations of Fortran 10 they promised that the
bugs were fixed in Fortran 10 and that even -O3 should not affect the
calculation results.
I've got the advice to use -O3 with
-fp-model precise and -fp-model source
and see if numerics are stable and faster than -O2
Check this page.
http://www.xcrysden.org/doc/density.html
If that menu item is active (and if your XSF file is bigger than several
KB), it means that you have density data in your file.
I remember I was confused with XCrysDen as well by the view of the
checkboxes (they are read and blue and
SaveTotalCharge writes ionic+electronic, not valence+(pseudo)core.
SIESTA calculates pseudo-core charge but doesn't have an option to save it.
I've made very small changes in SIESTA to save core charge instead of
ionic (I don't have any idea what for ionic charge could be used) by
setting
Hi Sungjong,
yes, Mulliken populations are very basis-dependent and (ideally) should
not be used at all for the analysis of charge transfer.
Some fast results could be gained with Andrei Postnikov's sphere charge
integration, but again this one is very radius dependent.
More reliable values could
Hi Abraham,
I think you can use the bases on the SIESTA site, any of them should
give you better results than unoptimized ones.
But I would suggest using the one with the biggest cutoff radius.
You can check which one would work better for you by comparing the total
energy for calculations
One thing i want to ask Siesta gives fermi energy in -ve value .
What can i do to set it at Zero. Can i do it or it is default?
You can of course shift all energies by the value of Ef and get Ef=0 on
your plots.
The zero-energy level in Siesta is defined as
the mean value over the unit
Here are ours optimized for C6H6
%block PAO.Basis
H2 0.0
n=1 0 2 E11.36136 0.00928
7.72405 2.19949
n=2 1 1 E41.15301 0.00947
2.89938
C3 0.0
n=2 0 2 E39.65304 6.21693
7.40483 3.265
n=2 1 2 E
but the major of experymetal works I know disagree in bulk
moduli by even more than 17%
That's true, but we are talking about disagreement with another PW
calculations, and in such case there should be ideally no disagreement.
Marcos,
comparison of different basis sets is boring, I know,
From my experience, I don't believe in the accuracy of SZ calculations
at all.
17% difference is a lot.
I would suggest you to really perform same calculations for all basis
sets possible and then decide what basis set to use.
Would DZP give similar bulk modulus close to PW?
Often I hear that
I have used f77 compiler for siesta and intel fortran compiler named
ifrotbin for compiling sies2xsf.
Your XSF file doesn't contain RHO data, since it couldn't be read by the
utility.
Use the same compiler and same compiler options - the data in RHO is
binary, so to decode it correctly you
Hi,
I've installed XCrysDen on Windows under cygwin.
It works fine but it really had some problems installing.
I had to use compiled binary.
Also you have to have installed Tcl/Tk (this is a separate package for
Windows http://www.activestate.com/Products/activetcl/) and OpenGL
(Mesa) - I
Have you generated the pseudo with the partial charge as well?
Hi John,
SZ and DZ results would not be very compatible, since due to different
basis size (and BSSE) you'd get different lattice constants.
However the rest of the physics should be reproduced at least
qualitatively correct.
So, I'd suggest, that if this is a really big slab, you can go with
I was wondering if any one could possibly share any rule of thumb or
technique for choosing appropriate rc and rm values for each shell.
Hi Thomas,
- I'd recommend using 1 zeta for deep levels and 2 zetas for higher
valence.
- Then let SIESTA to generate the PAO.Basis - for the start
I wonder why the n channel is running from 1 to 4 in those lines. All
the rest in the OUT file looks normal. Have anyone else seen this? I
guess it is just an innofensive bug
Yes, it is. I reported it about 2 years ago and I remember Alberto
Garcia told that he fixed it or was going to fix.
As my simulation runs, I see that in the .ANI file some of the water
molecules simply run away from the simulation box
This might be OK. As far as I remember, if you initially give the
molecule outside the unit cell, SIESTA will continue to use them and
will not translate into the cell during
Well, it's hard to call a relaxation.
Have you tried CG instead of Broyden?
What about the structure, if you visualize all relaxation steps, can you
see that the displacements reduce?
Also try to increase the accuracy of compilation by using double precision
DEFS=-DMPI -DGRID_DP -DBROYDEN_DP
Usually, the smaller cutoff radius for pseudo, the better (but not too
small, i.e. not closer than the node of the real all-electron potential,
to make it possible to conform to norm-conservation).
Pseudopotential would be transferable, i.e. can be used in any
environment, and the smaller
maybe I was doing a
mistake when estimating the BSSE correction.
What are the interaction energies without BSSE corrections?
How do you do BSSE corrections?
Please, send your full input fdf.
Try to take optimized basis sets from SIESTA page. It should give much
better results than
You don't have any problems with forces.
They are naturally oscillating on first CG steps, since you 440 atoms
and so many degrees of freedom.
But all in all after 30 CG steps you get forces ~0.1 eV/A, which is not
that bad.
Many more CG steps are usually needed to achieve smaller steps even
I'm new with SIESTA. I tested an optimization of H2O with both serial and
parallel using the same .fdf and .psf file. Funny thing happened when they
give total different results. The serial version seems correct.
Hi, it looks very strange, since the total energies after the first CG
step are
I think the reason is the small
MaxSCFIterations 100
This is usually not enough to achieve the DM convergence of every CG
step (check the energies of SCF cycles), that's why the forces are
jumping from step to step.
To improve it, you can use use pulay steps ~8.
Also, not on topic of
but somehow i get beta phase stable than alpha phase
while in literature it is reverse
I'm not sure of this is the case for Sn, but for Fe, for example, the
reason is the use of LDA instead of GGA.
Also, check the convergence vs. energy cutoff and k-points and try the
optimized basis
So the question is, is it acceptable such a large value of
PAO.EnergyShift ?
It might be. The optimal basis does not have to be the shortest.
You need to check the total energies obtained with big and small
EnergyShift. The one with lower Etot is better by variational principle.
If 1eV shift
I am trying to run a constant temperature
NOSE MD at 1500K. But each time,
my job hangs at MD step 1.
Can anyone help?
Show your input and output files, otherwise it's hard to geuss where the
error comes from.
Also how is the atom projected
DOS computed? The .EIG file
contains the
S 3
n=3 0 2
6.0 3.205
n=3 1 2
7.594 3.543
n=3 2 1
3.25
pe Sulfur
tm2 1.7
S pbr
0.0 0.0 0.0 0.0 0.0 0.0
34
30 2.00 0.00
31 4.00 0.00
32
Mixingweight 0.5
Have you tried smaller MixingWeight? Usually I find that values should
be smaller than 0.1
I might agree with Adam, that if you have some partially filled states
in the gap you would have problems with SCF convergence, so that more
k-points would be needed, or try to
I cannot agree that observing the relative change in Mulliken charges
instead of absolute values would be correct. I had experience that DZP
showed transfer of charge from surface to molecule, while TZP showed
the opposite (though the charge transfer itself was very smal - about
0.1e)
Was
There is an implementation for Bader analysis which can be used with SIESTA:
http://theory.cm.utexas.edu/bader/
Some minor additional coding is needed for SIESTA, I did it and can share.
Though, it's very rough, since I discontinued to work with it since
Bader analysis gave me very similar
DZ is defnitely not enough. Try DZP.
Have you checked for convergence of other parameters - k-grid, and
energy cutoff?
I don't think that 3d can be left in core in GaN (depends on what you
need, of course).
Moreover, I think you would see huge difference between 3d and no 3d cases.
Spell check (syntax highlighting) can be done in your editor which you
use for FDF files.
Here you have a file for NEdit on Linux
http://www.uam.es/departamentos/ciencias/fismateriac/siesta/
For Windows I use Notepad++ for both .fdf and .out files
If anybody needs, I can give the syntax
Try to play with BlockSize as well, although I don't think it would help
too much.
CPUs Time(secs) per MD
126.692
228.781
321.638
419.430
819.034
What do you call MD? Do you mean 1 SCF step?
Anyhow, these times are quite small and you can happily work even with 1
I tried to calculate the PDOS of a Si nanowire (~1400 atoms, SZP).
Hmmm... I wouldn't even continue to waste my time with SZ - it is fr
from good description of the system.
Is there any need to use so much atoms?
The
self-consistent calculation only takes a few hours to finish, however it
It is actually MgB2. The lattice is almost P6/mmm but not quite a!=c,
and, I guess, this additional degree
of freedom requires an additional position relaxation to obtain zero force.
It is very possible. Try to relax using variable cell.
MeshCutoff was 100 Ry. If any
of these parameters are
Free energy is the free energy of electrons at some finite temperature.
Efree=Etot-TS
At zero T they are equal.
Since the forces are calculated as derivatives of the
free energy all the DFT relaxations should be done at T=0K to show
correct results (i.e. same as derivatives of Etot).
Since
I'm not sure what exactly the problem is,
the only energy value that has changed is
siesta: Ena = 203.897261
but you don't run parallel version!!
You have:
* Running in serial mode with MPI
you need to include -DMPI in preprocessor flags to compile a parallel
version:
FPPFLAGS=
You have:
* Running in serial mode with MPI
you need to include -DMPI in preprocessor flags to compile a parallel
version:
FPPFLAGS= -DMPI -DFC_HAVE_FLUSH -DFC_HAVE_ABORT -DGRID_DP
First, the net force on a neutral system must be zero?
yes.
But from the
output, I don't see this. I even tried to run one atom in a big periodic
cell. To my surprise, the force on this single atom is not zero?
How far is it from zero?
You need also a good grid, i.e. high EnergyCutoff to get
My test system contains 58 atoms.
I have not input any K point.
Besides, the cell size is 9.38x9.38x6.7 Ang^3. Is this a small cell?
Is this molecule of bulk or slab?
For molecule the cell is too small, while for bulk, I believe, you need
more k-points.
Somewhere around 4 processors should be
I adjusted ProcessorY and BlockSize. But it doesn't seem to speed up a
lot. In fact, when I increase CPUs from 1 to 8, the speed only doubles.
Could someone tell me what I might miss?
This is normal behavior. The speed increase will depend on the size of
the system and how well the amount of
* Running in serial mode with MPI
Is this a error in the compilation plz help me in this regard.
You have to specify that you need parallel version using -DMPI in your
compiler flags
FPPFLAGS= -DMPI -DFC_HAVE_FLUSH -DFC_HAVE_ABORT -DGRID_DP
Thank you for the help. Could you please tell me how siesta calculate
ZPE? I couldn't find it in the manual or through the forum.
Usually people don't do that in DFT, although I saw that Gaussian
calculates it. If you could find experiments with ZPE extracted from
energy that would be the
When I then ran the executable (using mpdboot and mpiexec), I got errors in
the bessel test, depending on the number of nodes used: e.g. with 2 or 4
nodes, the test finished successfully, but with 6 or 8 I got a segmentation
fault.
If the system is quite small (which seems to be the case) you
BSSE corrections are the key to eliminate this error - lone atoms would
get more complete basis, so their energy will become lower and the
atomization energy as well.
This is absolutely not hard to calculate them, you just use the same
input as for the whole molecule, but only use something
What's interesting is that Spin-polarized calculations gives worse
results. For example, the atomization energy for oxygen:
O2: 7.357513(siesta SP) 6.191579(siesta NonSP) vs 6.18794982eV(PBEPBE
6-311G**)
H2: 0.010741(siesta SP) 4.529457(siesta NonSP) vs 4.53528752eV(PBEPBE
6-311G**)
You
It was already on the list:
There is indeed a silly bug in the denchar code that checks for files.
The fix is to replace the file Src/m_denchar_init.f by the one given in
the issues web page:
http://fisica.ehu.es/ag/siesta-extra/issues.html#denchar
atom: Maximum radius for r*vlocal+2*Zval: 27.49097
KBproj: WARNING: KB projector does not decay to zero
KBproj: WARNING: Rc( 0)= 6.1334
KBproj: WARNING: Cut of radius for the KB projector too big
All this tells that you have the problems with your pseudo generation
(too big radii).
1) Why is GGA pseudo-potential gives worse result than LDA?
What do you mean worse?
A well converged calculation (i.e. converged in E cutoff, k-grid, and
converged vs pseudo cutoff radii as well)
LDA SHOULD give bonds (or lattice constants) 1% smaller than expt (while
binding energies are
Zn is not a good example of using SZ for semicore
Well, it's easy to check - just take SZ and then DZ and compare total
energy.
If the impact of DZ is similar to that of DZ for s and p states, then it
is worth using DZ.
Of course checking for exact properties of interest would be more
Hi everybody,
after extensive discussion on this list, since I have partially filled
states in the gap of GaAs surface, I'm trying to use MP smearing to
improve convergence vs. k-grid cutoff.
But what I observe that for any non-zero value of smearing (I used up to
0.7eV (which increases
I need only validate results obtained with SIESTA.
Then compare the energy levels or binding energies and not basis.
Be aware that SIESTA default DZP basis set is not optimal, so that
results might be actually improved for longer (optimized) and/or more
complete basis (TZP for example).
I
However, it seems that I have
difficulty finding a systematic way of optimization
The general idea is described in the main SIESTA paper
The Siesta method for ab initio order-N materials simulation
José M. Soler, Emilio Artacho, Julian D. Gale, Alberto García, Javier
Junquera, Pablo Ordejón
Dear Professor Garcia,
I'm thinking to implement atoms with partial charges in SIESTA and ATOM.
I'm working with molecules adsorption on GaAs(001) surface and without
partial charges I cannot describe the system correctly - the terminating
H atoms bring too many states in the gap and affect the
The Fermi level is normally calculated by setting the cumulative occupation
number of all bands to the number of valence electrons.
As I understand this means that Ef in semiconductor would always be
at the VBM and not in the middle of the gap?
How it could happen that Ef appeared somewhere
Hi,
till recently I though that checking the convergence of total energy vs
k-grid cutoff is enough.
However, now I've found that while total energy can be very well
converged, Fermi level position is not, and requires at least twice
denser k-grid (and ~4 times more time).
Here is my example
I was feeling like I told something incorrect, since I hadn't actually
tried to change the T in fdf for a continuation run, but actually the
opposite situation - I changed the XV manually for only several atoms.
So I've rechecked it with a test MD run.
My conclusion about the ignorance of T in
Andrei,
the temperature will start from initial temperature and will gradually
change to target T during MD.TauRelax time with a given time step (as
for usual MD).
I've tried to use it with MD.TauRelax 200fs and 1fs step.
For my system (molecules on surface or molecules packing) it didn't
I'm not completely sure I understood what you want, but I assume you'd
have surf_basis_1.fdf, surf_basis_2.fdf, mol_basis_1.fdf and
mol_basis_2.fdf, and then you'd like to make different combinations of
those basis.
Right.
The easiest workaround to this would be to use shell scripts to
Thanks for the suggestion about using links for pseudos.
Following my previous question:
I model different molecules adsorbed on same surface, plus similar
situation appears when using BSSE corrections or playing with different
basis sizes between calculations - I would like to have bases in a
There is a reported problem with SIESTA, which is actually related to
the problems in math libraries that DivideAndConquer doesn't work correctly.
Try to set DivideAndConquer False.
Is it your case?
Error in Cholesky factorisation in cdiag
Perhaps a more physical way to
look at things would be to use the maximally localized Wannier functions
(which are quite similar but have a more molecular meaning).
How exactly to do that?
Or you mean exactly looking at LDOS of some energy region?
I think the simplest way to see what hybridization you have is to look
at DRHO plot.
Looking at just LDOS of some energy window can mislead your conclusions.
For example, recently I've tried to do similar things for GaAs surface.
In some of my geometries (of something adsorbed on GaAs surface)
1 - 100 of 131 matches
Mail list logo