# [Pw_forum] error on phonon calculation (root not converged)

Nicola Marzari marzari at MIT.EDU
Mon Apr 10 14:30:01 CEST 2006

```> there are 8 orders of magnitude between 10^{-10} (not-so-tight
> convergence) and 10^{-18} (execeedingly tight convergence).
> Try 10^{-12} and 10^{-14}, it shuod be more than enough for
> all purposes.

Agreed (caveat: if you have a long or large system, e.g.
a metallic slab or a nanotube, you might need to tighten
the convergence to -16/-18 - it also depends what do you need
the phonon for, and how accurately you need them converged.)

>>But the results (here 20-atom 2*2*1 supercell used) don't agree
>>well with those of calculation where 1*1*1 cubic perovskite
>>unit cell (5 atoms inside) are used. For example, the dielectric
>>constant of Srtio3, is for 2*2*1 supercell
>>
>>( 222.139158023 0.000000000 0.000000000 )
>>( 0.000000000 222.139158023 0.000000000 )
>>( 0.000000000 0.000000000 6.262166130 )
>>and for 1*1*1 unit cell
>>( 6.296436141 0.000000000 0.000000000 )
>>( 0.000000000 6.296436141 0.000000000 )
>>( 0.000000000 0.000000000 6.296436141 ).

To be on the safe side, use equivalent k-point sampling
for the two calculations. This means that if in the 1x1x1 unit cell
(5 atoms) you use, say, a 8x8x8 monkhorst-pack mesh, you should use
a 4x4x8 mesh in the 2x2x1 supercell. (This is not going to be
the source of your error, unless you are using an extremely poor
sampling (e.g. Gamma only). Note that dielectric properties
require more k-points to be converged, than say total energy or
forces.)

nicola

--
---------------------------------------------------------------------
Prof Nicola Marzari   Department of Materials Science and Engineering
13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
tel 617.4522758 fax 2586534 marzari at mit.edu http://quasiamore.mit.edu

```