Messages

5161

General Questions and Answers / Re: Choice of basis

« on: April 26, 2009, 23:10 »

As a general rule, to add the above considerations, I would never use SZ for real calculations, unless I was really forced to by memory constraints. There is just too little freedom both radially and angularly to properly describe bonds except in very spherically symmetric cases.

For several elements (specifically, those with a full shell as the valence), ATK will in fact refuse to run a SZ calculation since it will converge very poorly, if at all.

My rule of thumb is: SZP as a work-horse when trying things out, and most of the time for C-based systems, and DZP whenever I can afford it otherwise. DZ can sometimes be a good alternative to SZP if the geometry is close to "elemental". The difference between DZDP and DZP is rarely important; other errors like exchange-correlation functional are probably larger.

5162

General Questions and Answers / Re: ATKError: bad allocation

« on: April 26, 2009, 22:12 »

Is the error really exactly like you wrote?
In that case, it might not be an out-of-memory error after all (which is St9bad_alloc), but rather something else (like over-lapping atoms).
On the other hand, if you are running it properly now, let's hope it was a spurious thing, perhaps. Let us know if it appears again, and regularly.

Good luck with your calculations!

5163

General Questions and Answers / Re: why are there so many electrons in the scatter region?

« on: April 24, 2009, 10:52 »

Although I haven't seen the geometry, my guess is that the electrode is too short. If it's only 1 period, or even 2, this is too short and will, even if the calculation converges (which is sometimes hard, when the electrodes are too short), lead to the wrong results.

It was observant of you to note the charges, but you would probably have seen it very clearly in the transmission spectrum later. This is an ideal system, but I would not be surprised if the transmission spectrum comes out less than perfect (not integer plateaus).

So, increase the electrodes (use 4 periods) and let's see what happens!

5164

General Questions and Answers / Re: ATKError: bad allocation

« on: April 24, 2009, 09:25 »

The error means "out of memory", so you did the right thing to reduce the calculation size. Is there any chance you are running several MPI processes in parallel on the same computer? That is, you run in parallel but only have, say, a quad-core, or you use more MPI nodes than physical CPUs? That will put a much smaller limit on the max size of the calculation.

Where does it fail, in the two-probe part or in the electrode calculation? If it's in the two-probe (or equivalent bulk, when included), then changing kz points will not make a difference. How many kx/ky points? How many atoms in total?

Perhaps best to post the script + output, if you are comfortable doing so. You can mask the atom list if needed (replace all atoms by Hydrogen, or something). We just need to see the parameters and get an idea of the size of the system.

5165

General Questions and Answers / Re: "Killed signal 9" error

« on: April 23, 2009, 21:05 »

Upgrading to MPICH2 1.0.8 is probably always a good idea, and I hope it solves your problem. Otherwise, these MPI errors are a bit mysterious, and sometimes not reproducible. And when they are, they are not easy to solve...

5166

General Questions and Answers / Re: why are there so many electrons in the scatter region?

« on: April 23, 2009, 09:41 »

Please post the complete output file, it makes it easier to see exactly where the electrons are (if you used verbosity 10), or at least if there is a substantial difference in the equivalent bulk run compared to the two-probe part.

5167

News and Announcements / LaTeX on the Forum!

« on: April 23, 2009, 09:21 »

You can now use [tex]\LaTeX[/tex] equations on this Forum! The syntax is simple, just enclose the LaTeX formula in

Code

[ tex ]...[ /tex ]

(without the spaces, they are inserted to prevent the tags from actually rendering TeX code!)

Some examples:

[tex]\blue\Large      e^x=\sum_{n=0}^\infty\frac{x^n}{n!}[/tex]

There is even an interactive editor for equations! Click the last icon in the second button row (the blue one, to the right of the "list" icon)!

More examples to come (for now, see http://www.yourmathstutor.info/forum/index.php?topic=4.0, which uses the same technique as we do)!

5168

General Questions and Answers / Re: "Killed signal 9" error

« on: April 23, 2009, 09:11 »

Strictly speaking this is an MPI error, not an ATK error. It might indicate a network problem, which may be spurious or not. So, try to resubmit the calculation and see if it fails in the same point again.

5169

General Questions and Answers / Re: Confused on optimization

« on: April 23, 2009, 09:10 »

There are many approaches to optimizing systems, and two-probes are particularly tricky since they are made up of several parts.

The simplest, but also most time-consuming way is to make us of the equivalent bulk, as discussed above. Another is to pre-optimize the individual parts and then assume that nothing much happens when they are put together. For instance, if your problem is a nanotube with a defect, between metal electrodes, you may want to simplify the problem by optimizing the defect locally, in a smaller nanotube setting, and then assume the rest of the tube + the metal surfaces are perfect.

Or, if the contact properties of a nanotube with a metal surface are under investigation, you may assume that the tube and the surface are intact themselves, but their relative position may change. In this case you have to move the tube around on the surface in a clever, systematic way, to find the optimum. Or, as is probably more relevant, simply move it around more or less at random and do many calculations of the current, and then do a statistical analysis of the results (the current, for instance), since the real position of such a tube/metal interface will probably not be absolutely stable in an experimental setting anyway.

And so on

5170

General Questions and Answers / Re: Pseudopotential transfer between ATK and SIESTA

« on: April 22, 2009, 20:54 »

There are some software tools out there which can convert different pseudopotential formats between each other. Google something like "upf psf convert pseudopotential".

5171

General Questions and Answers / Re: Pseudopotential transfer between ATK and SIESTA

« on: April 22, 2009, 20:51 »

Theoretically this should be quite possible, if you can match the individual grids, etc. Just try

The difference in the two files lies in the exchange-correlation:
PZ = LDA Perdew-Zunger
PBE = GGA Perdew-Burke-Ernzerhof

5172

General Questions and Answers / Re: a convergence problem

« on: April 22, 2009, 10:15 »

Ok, the evidence for problematic behavior is already evident from the equivalent bulk part. While the total energy converges (barely, and in very many steps), the density matrix actually does not converge!

Thus, your first goal is to stabilize the equivalent bulk run, or, as an alternative strategy, skip it completly and go directly to the two-probe calculation. For this, change line 376 in the input file to

Code

    initial_density_type = InitialDensityType.NeutralAtom

However, the real reason for the instability of the equivalent bulk run is most likely because the electrodes are too short. I would make them twice at long. At the same time, you probably do not need such a long central region, so you can keep the total size of the system fixed by removing 2 periods in the central part.

For safety, I would also increase the size of the XY unit cell to 25-30 Angstrom instead of the current 18. This doesn't really influence the stability, but it might leads to some inaccuracies in the results in the end.

Further steps towards better convergence would include reducing the mixing parameter (probably not necessary) and increasing the temperature to perhaps 1000 Kelvin.

But try to just increase the electrodes and the XY cell, and just rerun with current parameters as a first attempt. If that works, there's no reason to worry about the other things!

5173

General Questions and Answers / Re: a convergence problem

« on: April 22, 2009, 09:37 »

Could you attach the complete output log, please? Which version of ATK was used?

Also, are you running this script in parallel or just on a single machine?

5174

General Questions and Answers / Re: St9_badalloc error

« on: April 20, 2009, 20:21 »

Yes, it means out of memory (searching on this Forum would give you that answer too )

480 C atoms can fit into 2 Gb, but it depends a bit on the parameters. For instance, if you have a DoubleZetaPolarized basis set, it naturally requires more memory than SingleZeta. Also, a bit hidden, is the number of k-points.

However, in your case there seems to be a simpler way to reduce the memory used per machine. You have 6 PCs yet you run 12 MPI processes. That means each PC will receive 2 MPI jobs, and thus only 1 Gb is available for each process. If you instead run "mpiexec -n 6" each MPI process will have 2 Gb.

5175

General Questions and Answers / Re: How to calculate the Fermi surface?

« on: April 20, 2009, 12:30 »

This functionality is not implemented in ATK yet.