Hi Magda, Yes, it is possible to achieve this, even though those are quite strict parameters for Siesta. You just need computer time and patience...
The fact that you get smaller forces for a very high cutoff is the eggbox effect that Alexander mentioned previously. You could see if your forces are as small with 300 Ry and some GridCellSampling. Unfortunately there is no rule of thumb, but I guess you could try some displacements based on the final forces you see, maybe setting the displacements along the directions of the residual forces when you use a lower mesh cutoff would help. If with a good grid cell sampling you get very very small forces in the end, it can be much more economical than using a very high mesh cutoff. Will demand you some time now, but could save you a lot in a longer run. Best of luck, Marcos On Wed, Jul 28, 2010 at 11:53 AM, Magdalena Birowska < [email protected]> wrote: > *Meshcutoff= 500 Ry is efficient to get accurate calculations. > What do you mean "accurate"?* > > I mean that I get for bulk calculations for GGA correct lattice constant, > bulk modulus. For 500 Ry I get zero forces on atoms for bulk GaAs, whereas > for 300 Ry I still have some not vanishing values. > > > * > So the question is, what exactly you are trying to study and how you model > it (based on what you prepare your starting geometry)?* > > I want to study a different configuration of one pair of Mn ions on the > surface of GaAs. I want to see the difference in the energy , electronic > structure, > I need to have very accurate calculations , forces should be very small > lake 0.002 eV/ A., DM.Tolerance 10-5, is it possible to reach such a > calculations on a slab? > > > Magda > > > 2010/7/26 Alexander Vozny <[email protected]> > > Do you think that I should also optimize the position of Hydroden atoms >>> or I shoudn't care about it? >>> >> I always did, but from my experience this should not matter much. So, in >> principle, you can just fix them in some predicted ideal/expected positions. >> (don't forget to fix the lowest Ga or As (actually both is even more >> correct) in the ideal bulk positions ("ideal" means ideal for your set of >> parameters)) >> >> I don't use any reconstruction, I just optimize atoms on the surface. >>> Optimazed surface is for me equal to reconstructed surface, am I right? >>> >> No. >> Ideal (or correct) reconstruction may even have a different stoichiometry, >> so whatever you do (even molecular dynamics at 1000K) will not help you to >> reach the correct geometry. >> So the question is, what exactly you are trying to study and how you model >> it (based on what you prepare your starting geometry)? >> >> > >> >> But I checked that for bulk calculation, cell containing 64 GaAs atoms, >>> Meshcutoff= 500 Ry is efficient to get accurate calculations. >>> >> What do you mean "accurate"? >> Even for bulk GaAs with 2 atoms in unit cell you won't get convergence >> even at 800Ry. >> This is what Andrei Postnikov always says (search for his older messages) >> - mesh cutoff is not a variational parameter (do I say it correctly?), so >> you should not check "convergence" for it (in a sense you do it for k-grid, >> for example). So mesh cutoff should be only as big as to provide the correct >> forces on atoms i.e. get rid of eggbox effect (try shifting your origin by a >> small value and see if the forces remain the same or simply check whether >> your forces on atoms in ideal bulk geometry are exactly zero (not sure >> though that this is equivalent to the above mentioned requirement)). >> GridCellSampling serves exactly for this. >> So (for GGA (for GaAs (not sure about Mn))) 300Ry is the maximum that I >> would use (even ~200Ry could be in principle enough). >> > >
