Geometrical optimisation of Si nanosheet

[tfpper]

What is the best way to optimise the geometry of a nanosheet of Si atom. From the tutorial, I've got the following. Should I use the default ATK-DFT or ATK-Extended Hückel?

Code

calculator = LCAOCalculator()

bulk_configuration.setCalculator(calculator)
nlprint(bulk_configuration)
bulk_configuration.update()
nlsave('C:/Users/davng/ATK/Silicene/SiliceneSheet10_0_10_Buckle2_OptGeom.nc', bulk_configuration)


bulk_configuration = OptimizeGeometry(
        bulk_configuration,
        max_forces=0.05*eV/Ang,
        max_steps=200,
        max_step_length=0.5*Ang,
        trajectory_filename='C:/Users/davng/ATK/Silicene/SiliceneSheet10_0_10_Buckle2_OptGeom_traj.nc',
        disable_stress=True,
        optimizer_method=QuasiNewton(),
        )
nlsave('C:/Users/davng/ATK/Silicene/SiliceneSheet10_0_10_Buckle2_OptGeom.nc', bulk_configuration)
nlprint(bulk_configuration)

[zh]

To get a reliable results before production run for a specific system, the users must do some test calculations for the setup of k-point mesh and the mesh cutoff by themselves. Usually, the ATK-DFT is more reliable than the ATK-Extended Huckel.

[tfpper]

OK. I got this from another ATK tutorial. Do you mean to add this code into the script. What is the default k_point_sampling? Also, as a test calculation, what is a good sample size in the case of a nanosheet?

Code

numerical_accuracy_parameters = NumericalAccuracyParameters(
    k_point_sampling=(9, 9, 1),
    )

calculator = LCAOCalculator(
    numerical_accuracy_parameters=numerical_accuracy_parameters,
    )

[zh]

Yes, you should add such setup into your script.  The default k_point sampling is 1 x 1 x 1, see
http://quantumwise.com/documents/manuals/latest/ReferenceManual/index.html/ref.numericalaccuracyparameters.html
.
It is difficult to say the  good sample size for a general nanosheet, because the unit cell size and the electronic properties (metallic or semiconductor) of the nanosheet may vary case by case.

[Anders Blom]

Huckel cannot be used for optimizations since it has no repulsive part of the pair potential. You could use DFTB if you obtain access to the relevant parameters, it's at least a lot faster than DFT, and reasonably accurate.