Page 1 of 1

Structural optimisation of CNT (13,0)

Posted: Thu Apr 15, 2010 10:06 am
by mikeb86
Hi,

I am trying to optimise the sructure of the carbon nanotube (13,0) for a phonon calculation.

I am running VASP 4.6.36 on the british supercomputer HECTOR.

Unfortunately, despite 100's of iterations, I have not been able to get the forces down to below 0.01. The target is to get them down to below 0.0001.

I have been running it with the PAW basis set and the PBE functional (I have been told that LDA will lead to bond lengths that are too long).

INCAR file (tried with IBRION = 2 as well):
NWRITE = 2
ISTART = 0

PREC = HIGH
ENCUT = 400
EDIFF = 1E-05

EDIFFG = -1E-03
ISIF = 3
IBRION = 1
NSW = 500

ISMEAR = 0
SIGMA = 0.1

ALGO = V
LREAL = F
LVTOT = T
LWAVE = T
ADDGRID = T
LSCALAPACK = T
NPAR = 16


I have been running it with one unit cell (containing 52 atoms). I am very confident that this starting geometry is close to the minimum. Usually CNTs are easy to optimise, does anyone have an idea?

Cheers,

Michael

Structural optimisation of CNT (13,0)

Posted: Thu Apr 15, 2010 11:11 am
by forsdan
A couple of points:

1: You shouldn't use ISIF = 3 to simulate an isolated carbon nanotube, since there can be interactions through the vacuum region which drives the tubes towards each other. This relaxation can be extremely slow and can be one contributing issue for convergence. So use ISIF = 2.

2: Your EDIFF is way too high to get very accurate forces. Please decrease it to 1E-6 or even 1E-7 or 1E-8 if necessary. Your target force 1E-3 is however extremely tight. Usually 5E-3 should be sufficent for a well-determined geometry. I have not yet seen a non-neglectable frequency change in the phonon spectrum when going down to 2E-3 or 1E-3 (at least not for my systems).

3: PREC=High is not longer recommended. Please use PREC=Accurate instead if you want to do very accurate calculations.

Cheers,
/Dan






<span class='smallblacktext'>[ Edited Thu Apr 15 2010, 03:33PM ]</span>

Structural optimisation of CNT (13,0)

Posted: Fri Apr 16, 2010 9:05 am
by Danny
in addition set EDIFFG=-1E-5 to get forces converged to 0.0001meV/A

Structural optimisation of CNT (13,0)

Posted: Wed Apr 21, 2010 10:12 am
by mikeb86
I have run the simultion again and it worked. This time I used PREC=ACCURATE, ISIF=2 (instead of 3) and EDIFF = 10-7. I suspect it was EDIFF=10-7 that allowed me to get down to lower forces (10-5), although I have not tested this.

Thank you for your help,
Michael