bands_above_fermi_level

[yasheng]

I want to run ATK faster because I have around 300 atoms in my calculations. As stated in this page (http://docs.quantumwise.com/tutorials/advanced_performance.html), one of the method to run the calculations faster is setting bands_above_fermi_level value. I have several questions about this parameter:

How do we estimate the empty bands above the Fermi level based on the number of atoms and the basis set?

What will happen if we use insufficient number of empty bands? Does it effect the optimized geometry and total energy etc?


One more thing: When I put the AlgorithmParameters to ChebyshevExpansionSolver (it was bulk configuration), calculation stopped with error, because the key word 'ChebyshevExpansionSolver' is not recognized. 


Thanks for your advice,
Yasheng




 

[zh]

For the value of "bands_above_fermi_level",  you can estimate it from the number of valence electrons (assumed as N_ele),
"bands_above_fermi_level" = integer (N_ele  * 0.5  +  N_arb),
You can take value of N_arb from this range: 10 < = N_arb  <= 50.

"ChebyshevExpansionSolver" is  not recognized because of the version of atk you used isn't consist with the one mentioned in the tutorial in document.

[Jess Wellendorff]

The bands_above_fermi_level argument to the DiagonalizationSolver gives you the option to include only a limited number of unoccupied bands when diagonalizing the Hamiltonian. This can give both speed-up and reduce memory consumption. However, if you include too few bands (i.e. exclude bands that are not completely empty), you will not correctly diagonalize the Hamiltonian, and your calculations may be inaccurate, hard to converge, or both.