Unusal Behaviour in the Vacuum

[Dhirendra]

Hi,

I am simulating a NiSi2/Si(100) interface similar to  the one given in the tutorial. But I am using vacuum padding instead of the device configuration. The boundary condition are not changed so I think the boundary conditions are periodic in the Poisson. With this I am getting a weird behaviour in the vacuum in the effective potential. The effective potential is not flat in the vacuum. I tested the same structure with Quantum Espresso package. It gives flat effective potential in the vacuum. What am I missing in ATK?
XC Functional used is GGA-PBE.

I am attaching the .py file for the simulation and the image of effective potential.

[Jess Wellendorff]

You are using periodic boundary conditions, and what you see is an effective potential that is perfectly periodic across the boundaries along C, right? So far so good. Try to use Dirichlet BCs along C instead. See the Note box just above the section title of this link: http://docs.quantumwise.com/tutorials/work_function_ag_100.html#running-the-calculation-and-analyzing-the-result

[Dhirendra]

Thanks Jess for your answer. I tried Dirichlet-Dirichlet boundary conditions on both the sides in the C direction. I still see the gradient in the vacuum on both sides (left and right). I also tried Neumann-Dirichlet boundary conditions, Neumann BC on left and Dirichlet on right this makes effective potential on left hand side vacuum flat but there is still a gradient on the right hand side vacuum.

Sufficient vacuum included so as to separate the neighboring atomic structures in the artificial periodicity. And so flat effective potential should come naturally with the periodic boundary conditions. A gradient in the vacuum is suggesting some effective field acting upon the system, which is not the correct description of the system.

[Anders Blom]

No, a flat potential does not come naturally with periodic boundary conditions, because in the slab approximation of an interface you have an artificial surface dipole across the periodic boundary, which compensates the real dipole at the interface between the two materials. QE and other codes have a 3rd dipole (also artificial) which compensates for this, whereas in ATK the more physically correct approach is to use Dirichlet/Neumann to eliminate it. If  you have a slope still, it means by definition you do not have enough vacuum. Don't be afraid to add more - unlike a plane wave code, vacuum is essentially "free" in ATK, so you can put quite a lot, it will not increase the simulation time substantially.

But - the main question is why use the slab approximation when you can treat the interface properly using the device configuration...