Hi Derek, Indeed this is a very good question. 0.04 eV/Ang is already quite a good convergence for the forces. However, the lower you go, the better - of course, within reason: there is no point in trying to get forces to values much smaller than this, most problem because of accuracy considerations. I guess that part of the whole thing is also related to a compromise between accuracy and computational effort. I remember that once I saw a paper on carbon nanotubes hat used something like 200 atoms or more, back in the late 90's - even with siesta, this could be a good effort, especially then. The authors converged their system down to 0.05 eV/Ang, and it was accepted for publication.
This kind of strictness in the convergence is good when, for example, you are going for phonon calculations. You want the initial positions to be as accurately described as possible, so that the numerical error that is already there influences the determination of the force constants as little as possible. Maybe in this case it would make sense to relax to even lower forces. It might be good to perform such strict relaxations also when you are investigating the energetics of alotropes and calculating structural phase transition pressures. This is because sometimes DFT predicts energy differences that are very small, so you might want to converge your results as well as possible. On the other hand, I would guess that if you want electronic structure - bands, DOS, etc. - it doesn't make much sense to relax below 0.01 eV/Ang. In this case, something between 0.04 eV/Ang and 0.01 eV/Ang would be good enough. Of course those are my personal impressions about the whole subject, and someone more knowledgeable on the list might have something to say about it as well. Please correct me if I'm wrong. As to Aaron's problem, maybe the problem is that the basis set you are using is non-optimized? Cheers, Marcos > Hi Derek, > > We currently use VASP to relax our molecules down to 0.01 eV/Ang. > However, > we do this for the experimental structure before engaging in both volume > and > uniaxial compressions. For the compressions experiments, we generally > like > to relax to 0.03 eV/Ang; however, at some point ionic convergence just > doesn't want to happen down to 0.03. In such cases we do everything we > can > to get convergence down to 0.05 - this we tend to deem publishable. Of > course, we are currently trying to match our Siesta results to our VASP > results with no success. If you end up successful in your endeavor, I > would > be very interested to know what parameters you had to adjust. Good Luck > with your research! > > Aaron > > >>From: "Derek A. Stewart" <[EMAIL PROTECTED]> >>Reply-To: "Siesta, Self-Consistent DFT LCAO program, >>http://www.uam.es/siesta"; <[EMAIL PROTECTED]> >>To: SIESTA-L@listserv.uam.es >>Subject: [SIESTA-L] Defining an acceptable force tolerance for Siesta >>Date: Mon, 6 Aug 2007 11:07:58 -0400 >> >>Hi everyone, >> >>I wanted to get the Siesta community's perspective on what is an >>acceptable tolerance for the maximum force during a structural relaxation >>tolerance. I know the default for the program is 0.04 eV/Ang, but I have >>seen several papers that use lower force tolerances. Would this imply >>that some papers using the default 0.04 eV/Ang tolerance are not converged >>enough? Are there specific systems where more care is required? >> >>Thanks, >> >>Derek >> >>################################ >>Derek Stewart, Ph. D. >>Scientific Computation Associate >>250 Duffield Hall >>Cornell Nanoscale Facility >>Ithaca, NY 14853 > -- Dr. Marcos Verissimo Alves Post-Doctoral Fellow Condensed Matter and Statistical Physics Sector International Centre for Theoretical Physics Trieste, Italy -------- I have become so addicted to vi that I try to exit OpenOffice by typing :wq!
