Messages

5101

General Questions and Answers / Re: Speed issue in Two probe configuration

« on: May 22, 2009, 18:29 »

Do you use the exact same parallelization strategy on the two cluster? I'm particularly referring to the loading of MPI nodes vs. CPUs w.r.t. the number of cores/sockets. You should avoid using more MPI nodes than physical nodes (i.e., if you have for instance 2 dual cores, you may not get very good performance using 4 MPI nodes on this system).

Also, sometimes on Linux the OS will not show the correct number of cores to thread on to the application. You can ensure ATK knows how many threads it can start on each node by setting MKL_NUM_THREADS by hand (see the manual!).

5102

General Questions and Answers / Re: Choice of basis

« on: May 22, 2009, 17:44 »

There is a fundamental difference between TB and the orbitals that are used in e.g. DFT. In TB you focus on the bonds between neighboring atoms, while the orbitals used in ATK are used to expand the electron density in the vicinity of each atom. In this case, the "bonding" comes about from overlap of such orbitals between atoms within a certain range. In ATK we use orbitals with a finite range (hard cutoff), and atoms within this range have a chance to "bond" in this way. The range is independent of the basis set type, but can be controlled by the "energy shift", as described in the manual.

For more information, see also http://quantumwise.com/documents/manuals/ATK-2008.10/ref.atomicorbitals.html.

5103

Scripts, Tutorials and Applications / Re: Script for calculate the bandgap.

« on: May 15, 2009, 13:16 »

The value for (7,0) seems fine, while the negative value for (6,0) probably is due to the fact that it's semi-metallic.

I suggest you look at a plot of band structure and try to relate it to the numbers you get. If you attach those plots (properly zoomed in around the Fermi energy), we can understand the situation better.

5104

General Questions and Answers / Re: question about Li-H2-Li two-probe system

« on: May 14, 2009, 08:38 »

Most likely the calculation has converged to a state with zero charge in the entire system. Au is trickier than Li and requires not only a bit higher temperature but most likely also a longer electrode.

5105

General Questions and Answers / Re: question on two-probe model setting up

« on: May 13, 2009, 13:37 »

It seems your pictures did not get attached to the post.

When it comes to selecting the Li atom, you will not see the actual atom get selected, that's true (and a bit confusing). You just need to have the mouse over it when you right-click to choose translate. Hope this helps.

5106

General Questions and Answers / Re: DOS did not correspond to the T(E)

« on: May 13, 2009, 13:13 »

I don't see any disagreement in the data, I think everything looks fine.

Except for one thing, maybe: the transmission of the pristine CNT should have integer plateaus, as you roughly have. But the "rounding" of the edges indicates that perhaps you have too short electrodes, or maybe even too small unit cell in the XY directions. Unless, that is, you simply have too few points on the black curve (fewer than for the blue and red). But it should really look like a step function, with the slanted parts.

I suggest you use 200 points or so for these plots, and check if the T(E) is really step-like for the pristine tube. But, otherwise in general, I don't see any problems.

5107

General Questions and Answers / Re: What's the physical insight of the variation of HOMO and LUMO with bias voltage?

« on: May 13, 2009, 13:01 »

What complicates matter in this case is that your electrodes are semi-conducting. Thus, I suspect you have no current in the low-bias calculations. However, as the bias exceeds the band gap, you start to get tunneling from the valence band of the left electrode to the conduction band of the right one, via the molecule. I imagine this somehow pins the HOMO to the valence band top of the left electrode, perhaps.

5108

General Questions and Answers / Re: DOS did not correspond to the T(E)

« on: May 13, 2009, 05:44 »

I'm having trouble viewing the pictures. I think they may be too large (not file size, but pixels). Can you reduce the size a bit and re-attach them, or email them to me. Also, it's better (for many reason) to use PNG pictures rather than JPG.

5109

General Questions and Answers / Re: Is the negative current meaningful?

« on: May 13, 2009, 05:03 »

Yeah, it happens...

This is clearly a quite tricky system you have, so it will require some clever tuning of the parameters to get it to converge. That's a bit of an art, sometimes, for difficult systems.

Did you increase the k-point sampling? Also, so far it seems we haven't discussed the importance of k-point sampling, and - most importantly, perhaps! - the width of the electrodes. If your electrodes are too thin, the result is often bad convergence. The same if the k-point sampling is too low.

I will be offline for a while due to travels, hope the points above bring some improvements.

5110

General Questions and Answers / Re: why cant I drag-and-drop models from crystal cupboard to atomic manipulator

« on: May 13, 2009, 01:16 »

This behavior varies a bit between tools and types of systems. The reason you cannot drop a crystal to the open AM is because bulk structures are assumed to be used for electrodes in the AM, and if the tool is already open, with a two-probe present, you are not allowed to replace the existing electrodes.

Glad you worked it out by yourself!

5111

General Questions and Answers / Re: wondering!

« on: May 12, 2009, 05:55 »

Well, you output reflects your input script, which is a bit messed up

First, you restore the crashed calculation and recalculation it (restart it). This is lines 1-16. But then you have a new definition of the whole system again, geometry and method, and then on line 213 you run one more self-consistent calculation, this time from scratch.

Finally, you compute the quantities you are interested, based on the last calculation. What you really want to do, is remove lines 18-218, and call the recalculated variable "scf" (instead of new_scf):

Code

scf = executeSelfConsistentCalculation(self_consistent_calculation=crashed_scf)

You will also want to call the new checkpoint file (line 13) something else, in order not to destroy "backup.nc", you may want to use it again later for some reason.

5112

General Questions and Answers / Re: setting the quantum number to calculate the HOMO and LUMO

« on: May 12, 2009, 04:01 »

Your way of counting assumes that the charge remain conserved on the molecule. However, in a two-probe system charge is not conserved. And, even if it usually almost conserved in the system as a whole (typical charge loss or gain in the central region is typically not more than 1e or so), the charge can easily redistribute itself onto or off the "molecule", which you have defined via the "surface_atoms" keyword.

The eigenvalues of the two-probe molecule spectrum are normalized such that zero is the (average) Fermi level. Thus states with negative energy are indeed occupied, and those with positive energy are not (well, more exactly, they are all populated according to the Fermi distribution).

5113

General Questions and Answers / Re: strange result

« on: May 12, 2009, 03:56 »

Ok, in that case, note that ATK does not function properly in parallel with any other MPI library than MPICH2 1.0.8 (or similar, like 1.0.5). Not only will you get weird output, like you got several times the forces, if you try other MPIs like hpmpich or openmpi, but also the program will not actually run faster compared to in serial.

5114

General Questions and Answers / Re: strange result

« on: May 11, 2009, 18:59 »

I guess you run this script in parallel, on 4 nodes? Perhaps you added the calculation to compute the forces by hand, and forgot to wrap the print statement in "if processIsMaster()"?

5115

General Questions and Answers / Re: the error of the .vnl result document

« on: May 11, 2009, 03:47 »

We have to admit that VNL doesn't manage memory properly in this case. I know, it's weird that a 100 Mb cannot be opened with 4 Gb of RAM, but that seems to be the case (for some types of grids). Later versions will of course resolve the problem.