Il giorno 22/nov/2010, alle ore 19.23, Dhruv Singh ha scritto: > And as expected, the termination threshold of 1.0e-3 gives a rather > lousy answer for forces and energy/atom of -11.39486969 Ry > significantly different from the correct value of ~ -11.395192 Ry. I
I wouldn't say "lousy"... you got what you asked for, that is to a say an accuracy of 1e-3 Ry on the total energy. > believe a finer kpoint grid should resolve the convergence issue to a > certain extent (correct me if I am wrong) - but I am trying to > understand if a well converged asnwer can be obtained even on a 2x2x1 > kpoint grid. if you use a n x n supercell, you should get the same accuracy of the primitive unit cell by reducing the k-point mesh accordingly (e.g. k_x/n k_y/n) GS > > Thanks, > Dhruv > > On Mon, Nov 22, 2010 at 1:18 PM, Dhruv Singh <dhruv.singh at gmail.com> wrote: >> Thanks very much for the help regarding this topic.... I guess the >> hexagonal supercell is a better choice then and I will need to use a >> finer k point grid with tighter scf convergence criterion for better >> accuracy in forces. As an advice though, would you suggest that for >> subsequent force constant calculations - forces of the order of 1.0e-5 >> would not lead to unstable modes ? (I am using a supercell to >> calculate force constants rather than the ph.x codes because I later >> need to calculate third derivatives) >> >> As regards the 72 atom supercell scf calculations not converging, the >> corresponding input file is attached here. I increased the scf >> convergence threshold to 1.0e-3 in order to terminate the computation >> earlier (to check the results). I am attaching two output files-one >> with threshold 1.0e-8 and the other with 1.0e-3. Both of them use a 72 >> atom supercell and a 2x2x1 kpoint grid. Please let me know if you >> think tweaking a few parameters could help convergence. >> >> Thanks very much >> >> Dhruv >> >> On Mon, Nov 22, 2010 at 8:26 AM, Paolo Giannozzi <giannozz at democritos.it> >> wrote: >>> >>> On Nov 22, 2010, at 14:23 , Gabriele Sclauzero wrote: >>> >>>> Hence, if you want more precision you probably need to use >>>> better k-point sampling and C-C distances. >>> >>> and better (i.e. smaller) self-consistenty convergence thresholds. >>> The error on the energy is a quadratic function of the error on the >>> charge density due to imperfect self-consistency; the error on forces >>> is instead linear. >>> >>> P. >>> --- >>> Paolo Giannozzi, Dept of Chemistry&Physics, Univ. Udine >>> via delle Scienze 208, 33100 Udine, Italy >>> Phone +39-0432-558216, fax +39-0432-558222 >>> >>> >>> >>> >>> _______________________________________________ >>> Pw_forum mailing list >>> Pw_forum at pwscf.org >>> http://www.democritos.it/mailman/listinfo/pw_forum >>> >> >> >> >> -- >> Dhruv >> > > > > -- > Dhruv > _______________________________________________ > Pw_forum mailing list > Pw_forum at pwscf.org > http://www.democritos.it/mailman/listinfo/pw_forum ? Gabriele Sclauzero, EPFL SB ITP CSEA PH H2 462, Station 3, CH-1015 Lausanne -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.democritos.it/pipermail/pw_forum/attachments/20101123/ac4d8b2d/attachment.htm
