This thread is locked.Only browsing is available.
Top Page > Browsing
Big carbon nanotube structure
Date: 2006/04/28 00:01
Name: Vasilii Artyukhov


I've been trying to perform calculations with OpenMX on a carbon nanotube fragment (a junction of semiconductor and metallic tubes) with about 500 and more atoms. The frament structure was initially optimized using Brenner interatomic potential.

I have encountered the following problem: the SCF calculations do not converge. There are huge oscillations in total energy, and Mulliken populations on different atoms vary chaoticaly from nearly zero to nearly eight. NormRD remains on the order of 10-20 and more. The problem persists both when I put passivating hydrogens on the dangling bonds at the cluster edges and without hydrogens.

I've tried the following input parameters:

PBE functional
scf.PartialCoreCorrection on
cutoff energy - 90 Ry
PAO - C3.5 and H3.0 (H4.0) s1p1
DC or GDC solver with hopping ranges from 2.5 to 5 and number of hoppings 1
every mixing scheme
electronic temperature - 300K and 3000K

Can you have any idea why the SCF problem can be so ill-conditioned in this case? Can it have anything to do with the quality of initial structure (I guess, it shouldn't)? Can it be due to the small basis set and energy cutoff used? I've performed calculations of a C60 molecule with same parameters, and they were successful (in fact, succesful with even more modest cutoff and basis). Shall I increase the number of iterations (I've tried up to 100)?
Page: [1]

Re: Big carbon nanotube structure ( No.1 )
Date: 2006/04/29 01:58
Name: T.Ozaki


The problem you have is charge sloshing which tends to occur
in systems characterized by large, low-dimensional, small gap,
surface or edge structures.

This charge sloshing can be significantly suppresed by using
k-space mixing methods such as Kerker and RMM-DIISK.

Then, you will see that a larger "scf.Kerker.factor" is quite
effective to avoid the charge sloshing. In addition, the choice
of "scf.Min.Mixing.Weight" and "scf.Max.Mixing.Weight" is also
a crucial factor for the charge sloshing.

By the way, although you are using basis functions with short
cutoff radii (C3.5 and H3.0 (H4.0) s1p1), the cutoff seems to
be short for me. Also, please make sure that "DC or GDC solver
with hopping ranges from 2.5 to 5" is less accurate.



Page: [1]