Query regarding TS.biasContour.NumPoints keyword

Asked by Sam on 2020-10-02


I have a query regarding TS. biasContour.NumPoints keyword in siesta/4.0. I wanted to understand the effect of this keyword on tests that are run at different biases.

I am running some tests on a Au-CO-Au structure using the following settings:

1. With TS. biasContour.NumPoints = 20 at 0.5eV and 2. With TS. biasContour.NumPoints = 25 at 0.5eV
3. With TS. biasContour.NumPoints = 20 at 1eV and 4. With TS. biasContour.NumPoints = 25 at 1eV

Tests 1 and 2 gave me very similar I-V results (i.e. 5.00000000E-01 4.51745599E-06 ; 5.00000000E-01 4.52008816E-06). Similarly, I got very similar results for tests 3 and 4 (1.00000000E+00 1.93833341E-05
1.00000000E+00 1.93752652E-05) .

Another test (#5) was run with TS. biasContour.NumPoints = 5 at 0.5eV. Calculation gave the following error: "Contour: too few points for real axis int"

I have the following questions:

1. In one of the questions asked in this forum, you had asked to set this value = V/0.01 (for older versions). However, since I am not observing any difference in my results, should I just stick to one value (say 20) for all my calculations at different biases (voltages from -0.1 to +5)

2. Can it be said that as long as we do not see any error/warnings in the output files related to contour points and if scf convergence criteria is met, I can continue using one value of TS. biasContour.NumPoints for all the biases (even if the structure is different).

3. Also, if TS. biasContour.NumPoints is set to one value in all the calculations, will it have any impact on the convergence?

Thank you,

Question information

English Edit question
Siesta Edit question
No assignee Edit question
Solved by:
Nick Papior
Last query:
Last reply:
Best Nick Papior (nickpapior) said : #1

1. You should probably look at the DOS rather than the current. The current is more "stable" since it is an integration over the bias window. Whereas, the DOS may be more energy dependent.
If you have done proper tests to ensure that the energy spacing is ok, then sure, you can do whatever. But generally for an energy spacing 1 eV / 20 = 0.05 eV you have to be sure that no states are in the window from say Ef -- Ef + 0.05 eV. This is why we recommend a spacing of ~ 0.01 eV since that is a relatively good compromise between accuracy and speed.

2. No, generally not. Again this refers to the above. If you change the applied bias but do not change number of points in the bias-contour, then you have two different integral spacings, meaning they are incommensurate.

3. Typically this flag does not really influence the SCF, but the final density matrix (thus density) heavily depends on it. So you can't guess whether you have a good number from the SCF cycle.

Sam (sam132019) said : #2

Thanks Nick Papior, that solved my question.