ÿ On Sun, Apr 18, 2010 at 5:34 PM, Руслан Жачук <[email protected]> wrote:
> I use Siesta-2.0.2. I need to calculate surface formation energy very > precise. > But these fluctuations produce errors well below 0.1 meV/Ang2 so I am > satisfied. > You mean per atom? Because, in absolute values, you have 0.5 meV/Ang. > I will try to increase cutoff and follow how the numerical noise change, > just for curiosity. > From the graph I posted the calculation is converged at 30 Ang (15 Ang > vacuum space), isn't it? > That depends a lot on your criterion of convergence. Nevertheless, vaccuum comes practically for free in Siesta so you can set it to 70 Ang as well... Check the time it takes to finish 15 ang and 70 ang vaccuum, if the calculation times are not so different, why not place as large a vaccuum as you can? Marcos > > > Ruslan > > 2010/4/18 Marcos Veríssimo Alves <[email protected]> > > Ruslan, >> >> This is already off the limits of DFT precision... I'd say it's just >> numerical noise. If the fluctuations were much larger (say, of the order of >> a few tens of eV), then I might be tempted to wonder about it. I would say >> your calculation is already very well converged at about 70-80 Angstrom. By >> email correspondence, a friend told me that this convergence problem could >> be greatly diminished by increasing the mesh cutoff, which really leads me >> to think this could be a spurious numerical issue, maybe a spurious dipole >> moment. Try doubling the mesh cutoff, or even multiplying it by three, and >> check out what the effect is on the numerical results. However, if you're >> using siesta 3.0, have you tested the feature of dipole correction for >> slabs? (This is actually a question driven by curiosity, since I have not >> tested it myself but would love to know what its behavior is like) It could >> save you some cutoff increase - wild guess. >> >> Cheers, >> >> Marcos >> >> >
