Smearing the I-V curve

[Hasan Sahin]

Hello,

I tried to calculate an I-V curve with 0.005 volt steps and I see that there are some peaks. Interestingly, there is no physical reason (any molecular state causing NDR etc.) for these peaks. I guess that I can solve these meaningless peaks by a smearing correction but I dont know how to do this.   

[Nordland]

If you increase the infinitisimal in the calculation of the current, then you will no longer see these peaks.

[Hasan Sahin]

Ok. Thank you

[bidisa]

I am also getting some well defined and some not so well defined peaks in my I-V curve for a two-probe system with a molecule in between. I am finding it difficult to interpret the peaks. The transmission spectrum though have very sharp peaks near Fermi energy.
Anyone has a clue why this may happen. Molecule in between the eletrodes are fixed so conformation change is not allowed.

I attach the I-V curves for three systems in a series.

Thankyou for any replies.

Bidisa

[Hasan Sahin]

I had similar I-V curve that includes unrealistic peaks and dips... So far, I have tried to decrease green_function_infinitesimal to 1*e-4 and decrease the electron temperature to 100 K. Decreasing these parameters gives more reliable results but problem still occurs. I hope we solve this problem soon...

[Nordland]

Have you tryed to increase the tolerance to 1e-9 or so? Do the problem still persist?

[Hasan Sahin]

No, I didn't. Now I am going to start the calculation with the parameters;

electron temp = 300  (it was 600 in my first trial)
infinitesimal=1*e-4
tolerance= 1*e-8

[Nordland]

If possible try to use the infintisimal 1*e-7 in both the SCF calculation and current calculation.

[Hasan Sahin]

I can set paramaters to these values but for scf calculation there is no green_function_infinitesimal parameter . Did you mean that real_axis_infinitesimal must be set to  1*e-7.

[Hasan Sahin]

bad news...  I have been tried various parameter sets as follows.

green_function_INFINITESIMAL === e-5, e-9
TEMPERATURE  === 600, 300 and 5 Kelvin
TOLERANCE      === e-5 and e-8

My I-V calculation still has strange peaks (as can be seen in attached figures)
What are the  paramaters directly affect I-V values.

[Anders Blom]

What is red vs black curve?

How different are the transmission spectra?

One should also note that the current integration routine in ATK is not extremely accurate numerically. So another point that would affect the current values at the level of accuracy you are looking at would be the number of points in the current integration.

[Hasan Sahin]

black and red curves correspond to UP and DOWN spin current.  I said that they are strange because there are peaks and dips in I-V curve while corresponding transmission spectra has zero transmission coefficients.

[Anders Blom]

If the transmission spectrum is zero then the current will be zero

But I guess T(E) is not exactly zero, just very small. In that case, I guess the current integration can have a large effect on the results. Can you post an example of the transmission spectrum, at two different biases? Perhaps in log-scale.

[Hasan Sahin]

If the transmission spectrum is zero then the current will be zero

But I guess T(E) is not exactly zero, just very small. In that case, I guess the current integration can have a large effect on the results. Can you post an example of the transmission spectrum, at two different biases? Perhaps in log-scale.

If transmission coefficient is zero at a certain energy value it has no contribution to the current (here we are agree) but current does not have to be zero (it may has a finite value due to previous transmission contributions ).
I attached transmission spectrum for up and down at zero bias ( for finite bias, negligible changes occur and we have almost same transmissison spectra)

I am sorry, I have no log-scale transmission.

As you see, there is only one peak (around fermi level) for up and down transmission, and I-V curves (see previos message) need to be corrected...

[Anders Blom]

Ok, it looks pretty much like I suspected

First, I'm wondering whether the actual current is that relevant in this case. The resonance is very sharp and has a reasonable value (not like 0.01), but probably does not change much at all with bias. Therefore, you may not need to spend a lot time computing the current extremely accurately; it seems to me the effect you are looking for is not in the I-V curve but the position of the resonance itself, its height, the difference between spin up/down, how the peak changes when you modify the structure, etc.

The fluctuations in the current are almost certainly due to poor accuracy in the integration. Taking the above point into account, if you are really concerned with the precise value of the current, you need to integrate the transmission spectrum more accurately. However, since the peak is quite sharp, it will be very inefficient to just increase the number of points to a huge value, since most of the points will be in the areas of zero transmission anyway. What you need an adaptive integration routine, that puts more points in the region of the peak, and fewer outside it.

Fortunately for you, I recently wrote such a routine :-)

I'll be happy to share it with you, perhaps, for now, off the Forum, however, since it's not 100% finalized (anyone else who is interested in it are most welcome to contact me!).

However, again: even if you were able to compute the current extremely accurately, I think it will just show a linear increase, which is the bias window "background". The real physics here lies, as I see it, elsewhere, and not in the I-V curve.