Correction to my previous email : Then, find the *smallest *value of mesh-cutoff that fits this criterion (while making sure of course that larger cutoff values also fit the criterion). *This smallest value is your converged mesh-cutoff.*
On Wed, Aug 12, 2015 at 7:34 PM, Nicolas Leconte <[email protected]> wrote: > Hi Stephan, > > In short, it is completely up to you to decide when convergence is reached. > > In long, yes, it is quite an arbitrary choice. However, you should let > literature guide you what are usual convergence thresholds for the > parameter of interest. Depending on the code, plane waves, numerical > orbitals, etc, you can achieve better or worse accuracy (e.g., meV, > micro-eV). > > So, how to proceed? > > Do your calculations for different mesh-cutoffs, just as you did before. > > Do one extremely converged calculation, i.e. a very large value where you > don't see any "sensible" change anymore in the numerical values. This > concept of "sensible" is related to the code-dependent accuracy that I > mentioned above. > > Decide on your convergence criterion, absolute or relative, in comparison > with this one calculation. These criteria can for instance be "0.1% error > compared to that one calculation" or "within 10 meV error compared to that > one calculation". > > Then, find the value of meshcutoff that fits this criterion (while making > sure of course that larger cutoff values also fit the criterion). > > And mention this convergence criterion in your publication! > > Note that convergence studies should be done for the parameter of interest > for your study. It's not because convergence is reached for the total > energy that it will be the case for your bond-length, or your band-gap, or > your phonon spectrum, etc... > As a rule of thumb, energy usually converges much faster than the rest. > > Good luck, > Nicolas > > > > > > On Wed, Aug 12, 2015 at 7:12 PM, Mayuri Vaghela <[email protected] > > wrote: > >> when you take any convergence if you draw mesh-cutoff vs energy graph , >> you will get minima in the graph put that value of meshcutoff as your >> converged mesh cut off value >> >> On Tue, Aug 11, 2015 at 10:21 PM, <[email protected]> wrote: >> >>> Hi Stephan, >>> it depends on how big is your unit cell and what you want to study. >>> >>> > Hi, >>> > >>> > I've a question concerning the convergence criterion. >>> > >>> > Let's say I want to examine convergence of the total energy a system >>> with >>> > respect to the meshcutoff. >>> > >>> > When is convergence reached? Is it reached when there is no change in >>> the >>> > first digit after the comma >>> > >>> > like in the following example: >>> > >>> > meshcutoff total energy >>> > >>> > 150 -42994.1019 >>> > >>> > 170 -42994.116 >>> > >>> > 190 -42994.127 >>> > >>> > 210 -42994.133 >>> > >>> > >>> > >>> > or is it usefull to continue the test till there is no change in the >>> > second digit after the comma? >>> > >>> > What is a sensible criterion for convergence? >>> > >>> > >>> > >>> > Thanks and regards >>> > >>> > >>> > >>> > Stephan Ludwig >>> > >>> > >>> >>> >>> ********************************************************* >>> Dr Zacharias G. Fthenakis >>> Research Associate >>> Institute of Electronic Structure and Laser (I.E.S.L.) >>> Foundation for Research and Technology Hellas (FO.R.T.H.) >>> Vassilika Vouton P.O. Box 1527 71003 Heraklion Crete Greece >>> Phone +30 2810 391824 >>> FAX +30 2810 391305 >>> webpage: http://esperia.iesl.forth.gr/~fthenak >>> ********************************************************** >>> >>> >> >
