[Pw_forum] Stress tensor from pw.x and cp.x
Dear Nicola, Thank you very much for giving me some ideas about the issue! Indeed, I am using ultrasoft pseudopotentials. And therefore, I've set the nrb's carefully using the formula, "nr1b=2*rcut*nr1/Lx", and make it a bit larger than values calculated with this equation. However, you remind me that I may not have a high enough ecutrho. My ecutwfc is 35 Ryd and 150 for ecutrho. This might not be enough... I will try a higher value on ecutrho to see if there are any difference. If there is still the problem, I will then have to try the QE-4 package. Thank you very much! -- S.H. Lee M.Phil Physics Division The CUHK Graduate School On Thu, 2008-05-22 at 02:36 -0400, Nicola Marzari wrote: > > Dear Lawrence, > > > the only suggestions that comes to mind is this - the stress tensors > converges to its "true" values more slowly with energy cutoff then > other quantities. Usually the error at standard cutoffs is > fairly "rigid", i.e. due to contributions that are not sensitive to > the chemical environment, and so you can even correct for that > empirically. > > CP and PWscf have a different way of treating augmentation charges, > so that's the only major difference that comes to mind - but it would > apply to ultrasoft pseudo (that's why posting your input would have helped). > > Bottom line - let's use QE-4.0 (i.e. download it now) and retest > with increasing cutoff, and if you are using ultrasoft pseudopotentials, > throw in for good measure a high charge density cutoff (10-12 times the > wavefunctions). In CP, test also for the box grid size (nr1b...), > increasing values. And then maybe post again, at least the results > and an input file. Of course, if it is ultrasoft, and your nr1b is not > really that good, that could explain it right away. > > > nicola > > > > Lawrence Lee wrote: > > Dear all, > > > > Since my calculation is really simple, I am not going to attach my input > > file here. I use the QE-3.1.1 code, and perform scf calculation using > > pw.x on a 64-atom cell, which has a perfect fcc structure, using only > > the gamma point. I obtain a stress tensor then. > > > > Later on I use exactly the same structure and cell and cut-off's, but > > using cp.x to perform a damped dynamics on the electrons, keeping the > > atoms fixed. I also obtain a stress tensor. > > > > However, the results between pw.x and cp.x is different, apart from the > > difference in units. The cp.x results are generally about 5~10 GPa above > > that of pw.x. > > > > I find this quite confusing. Is the method of calculation of stress > > tensor by these two programmes different? > > > > Thank you very much for solving my doubt. ___ ?x?? - Yahoo! Messenger ???]?W?o???W?r???f http://messenger.yahoo.com.hk
[Pw_forum] Stress tensor from pw.x and cp.x
Dear all, Since my calculation is really simple, I am not going to attach my input file here. I use the QE-3.1.1 code, and perform scf calculation using pw.x on a 64-atom cell, which has a perfect fcc structure, using only the gamma point. I obtain a stress tensor then. Later on I use exactly the same structure and cell and cut-off's, but using cp.x to perform a damped dynamics on the electrons, keeping the atoms fixed. I also obtain a stress tensor. However, the results between pw.x and cp.x is different, apart from the difference in units. The cp.x results are generally about 5~10 GPa above that of pw.x. I find this quite confusing. Is the method of calculation of stress tensor by these two programmes different? Thank you very much for solving my doubt. -- S.H. Lee M.Phil Physics Division The CUHK Graduate School ___ ?x?? - Yahoo! Messenger ???]?W?o???W?r???f http://messenger.yahoo.com.hk
[Pw_forum] Stress tensor from pw.x and cp.x
Dear Lawrence, the only suggestions that comes to mind is this - the stress tensors converges to its "true" values more slowly with energy cutoff then other quantities. Usually the error at standard cutoffs is fairly "rigid", i.e. due to contributions that are not sensitive to the chemical environment, and so you can even correct for that empirically. CP and PWscf have a different way of treating augmentation charges, so that's the only major difference that comes to mind - but it would apply to ultrasoft pseudo (that's why posting your input would have helped). Bottom line - let's use QE-4.0 (i.e. download it now) and retest with increasing cutoff, and if you are using ultrasoft pseudopotentials, throw in for good measure a high charge density cutoff (10-12 times the wavefunctions). In CP, test also for the box grid size (nr1b...), increasing values. And then maybe post again, at least the results and an input file. Of course, if it is ultrasoft, and your nr1b is not really that good, that could explain it right away. nicola Lawrence Lee wrote: > Dear all, > > Since my calculation is really simple, I am not going to attach my input > file here. I use the QE-3.1.1 code, and perform scf calculation using > pw.x on a 64-atom cell, which has a perfect fcc structure, using only > the gamma point. I obtain a stress tensor then. > > Later on I use exactly the same structure and cell and cut-off's, but > using cp.x to perform a damped dynamics on the electrons, keeping the > atoms fixed. I also obtain a stress tensor. > > However, the results between pw.x and cp.x is different, apart from the > difference in units. The cp.x results are generally about 5~10 GPa above > that of pw.x. > > I find this quite confusing. Is the method of calculation of stress > tensor by these two programmes different? > > Thank you very much for solving my doubt. -- - 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
