Small Transmission Coefficients and Low Current

[duygu]

Hi all,
I am calculating  a 1D structure and seeing small transmission coeffients such that the resulting current values are at microAmpere range at finite biases. I first checked charge but there is no charge missing. There are approximately 51.60272 spin-up charges and 52.72461 spin-down charges in the central region at 1V (# sc 47 : q =  104.32771 e  Etot = -397.37066 Ry  dRho =  9.6462E-04  dEtot =  3.2366E-04 Ry). Then I tought that it might depend on number of buffer layers, so I repeated the calculation with 10-10, 20-20 and 40-40 buffer layers. Unfortunately this caused another problem. I saw that transmission properties of the device changes depending on the number of buffer layers. This is unphysical. At last I've checked the voltage drop (by choosing electrode constraint as Density Matrix) and saw no abrupt change in the potential. Then I conclude that the scattering of the electrons from the contact cannot be the cause of this small transmission coefficients.
Any new ideas? 

ps. Plots of the voltage drop for the spin-up and spin-down electrons and transmission spectrums at 1V are attached successively (due to size limitations I attached them seperately).

[duygu]

plot#2,3&4

[Nordland]

I seems you have done all the right things, so I am just guessing here.

1) The electrode might be a bit too short for avoiding having artifical scattering at the interface, if this is the problem, then this would lead to "buffer" layer variation in the transmission. But since I don't know the element of the electrode, it is simply speculation. But you can try to see what happens in you increase the electrode by perhaps 50%.

2) In some sense you are correct about it being unphysical that the transmission depends on the number of buffer layers, but as the determining factor is the electric field - in a crude approximation the total length of the scattering region divided by the bias - and therefore you should check if the voltage drop looks the same, depending on the number of buffers layers. If it changes by adding more buffer layers, then we have the reason, that metallic 1d wires is the worst screening material that exist and hence it really requires alot of buffers layers to avoid giving trouble.

3) How does the transmission spectrum look like at 0 V ? Perhaps this can give us a clue on, if this is a problem of method or system....

[duygu]

Carbon linear chains are used as electrodes. I tried to increase the electrode by 50% as you said but I got the following message:
ATKError: inverse(DZMatrix const &) : Could not LU factorize!
So I increased the electrode by 25% (without getting any error messages). Unfortunately there is no improvement, actually I got the same zero bias transmission spectrum. The plot for comparision is attached, the zero bias trans. spectrum can be seen. When I inspected the the structure with VNL I noticed that the distance between right electrode and the central region is different from what I wanted. While building the two-probe geometry I was using the same configuration for both electrodes.  By taking the mirror image of the left electrode  configuration for right electrode, I've solved this problem. Now both contact distances are symmetric. But this time I got the following message for zero bias calculation:

ATKError: solveSymmetricGeneralizedEigenProblem : Failed to diagonalize matrix!

Is there any other way to say the ATK to put the electrodes at a specific distance away from the central region?
 
Another thing in my mind is electrode repetition. I am using the default value but I am including more than one atoms in electrode unit cell. Could it be the reason?

[Nordland]

What are the atoms in the central region?

[duygu]

Cr's

[Nordland]

The two errors you see is due to atoms being placed on top of one another - something is incorrect in your script, since it gives this error. I will try to redo the calculation again on my laptop - what lattice constant have you used for the chain?

[Nordland]

Another thing in my mind is electrode repetition. I am using the default value but I am including more than one atoms in electrode unit cell. Could it be the reason?

No. The eelctrode repetitions is a quick way for expanding the electrode unitcell in the XY-plane, and should only be used in rare cases. And there is no use for it in a system like this.

[duygu]

At first I optimized the central region atoms (in periodic atom configuration) then I used the optimized geometry as central region. The scripts for setting up the geometry and zero-bias calculation are attached. Structures built by 2probe_configuration.py runs without any problem, but device-right electrode distance is not 1.267 Ang. as I wanted. 2probe_configuration2.py fixes this problem (I cannot see any overlapping atoms in VNL) but its two-probe calculation gives "fail to diagonalize matrix" error

[duygu]

Due to periodic boundary conditions does the first atom of the left electrode overlaps with the last atom of the right electrode in the geometry created by 2probe_configuration2.py? If this is the problem how can I make the right electrode to start 1.267 Ang far from the central region?

[Nordland]

There is no reason for moving atoms around to align them correct as this is done automatically.
It is done by using be the equivalent atoms, and therefore the setup generated by the first 2probe_configuration.py is correct.