Peaks in the average effective potential at device boundaries

[ziand]

I calculate the effective potential of a strange "device". It has two metal electrodes, separated by a large vacuum. The Veff is then averaged along x and y (z is transport direction) and plotted as a function of z (see attached figure). By looking carfully at the right boundary of the graph, we see a sharp (1 data point) peak. What may be the reason for that and is this something to worry about?

[kstokbro]

It looks strange, if you send the input file such that we can generate the data, we can take a look at it.
Which version did you use?

[ziand]

Ok, so this is not some unproblematic numerical artifact?
Here are some details:

There still version 11.2.3 at the cluster where I did those calculations. The calculations itself were done quite some time ago, too. It is just now, that I calculated the effective potentials from the saved SCF-states. However, I have attached a minimal script with my simulation parameters (the saved SCF-state would be much too large). I created the script by dragging the calculated configuration into the VNL-scripter. This should keep all the parameters. I did not yet include the part for taking the averages along the A- and B-direction but.

But note: Redoing the calculation is very time consuming given the big size of the system (the vacuum gap could be decreased but it would still take its time).

Some further explanations:
The previously posted effective potential was for copper. I tried the same for Aluminium, Palladium, Siver, Platinum and Gold.
Silver and Gold did not work at all (segmentations fault). I have absolutely no idea why this is so! (If you can tell me a nice way to send you a 200Mb file, then you can try it out on your own.) The other metals run and I get results.
There is alway such a spike at the right boundry, except for aluminium!!! This one is okay.
Well, I was quiet dissatisfied with the situation, but luckily I had saved non-selfconsistent device calculations, too (that is: only the effective-bulk calculations). I have quite some metal to the left and to the right, so that the screening condition should be matched very well, so I think I can use those and get comparable results.
Everything worked out fine (and this time there was no seg. fault when calculating the eff. pot. of gold or silver). The good news: results are very much the same compared to the self-consistent run. The bad news: The peak at the right boundary is reproduced, too.

One final note: I have the feeling this issue could be related to the Poisson solver? I use FFT for the electrodes and FFT2D for the devices. Is that okay???
But even if not: how would that explain why there is a peak in the non-selfconsistent results, too?

[Anders Blom]

Haven't had time to look at this. We will release 12.2 this week, then we have a bit more free time

[Anders Blom]

Finally returning to this. It's hard for us to test, if it's so time-consuming. You can send large files via http://letscrate.com/ if we should have a look.

But the Poisson equation is a good hint; if different methods are used I'm not sure what would happen... The normal approach is to use FFT2D also for electrodes, even if you set periodic boundary conditions in all directions.

We can perhaps agree that this doesn't really pose a serious problem, it's just one point and the rest looks absolutely fine - very fine, in fact; it's not easy to get the potential profile right with such a large vacuum gap, this is possible thanks to the double contour integration in ATK.

The segfaults are another story of course. Memory?

[Echo]

Dear all,
I was working to find the effective potential along a cut-line. I have contour plot of 2D plot. How should I proceed?

[kstokbro]

This has to be done by scripting, you may use:

potential = nlread('results.nc', ElectrostaticDifferencePotential)[0]
for x in numpy.linspace(0,12,25):
    print x, potential.evaluate(x*Ang, 1.*Ang, 4*Ang)

To average the potential along an axis check:
http://quantumwise.com/documents/manuals/latest/ReferenceManual/XHTML/ref.electrostaticdifferencepotential.html

(In the next version ATK 13.8 it is our plan to allow these operations from the GUI)