Dear Prof. Paolo Giannozzi: Thanks for your reply.
But, according to the change of number of pools, I also changed the number of cores; therefore, plane waves, fft grid, etc. are distributed equally in both cases. Also, each diagonalization is carried out within the same number of cores. At SCF step, most relevant values over k points are merged by use of mp_sum, mp_max, etc. At least to me, this difference seems strange. Regards. H. Lee On Thu, Oct 9, 2014 at 10:02 PM, Paolo Giannozzi <[email protected]> wrote: > On Thu, 2014-10-09 at 20:29 +0200, H. Lee wrote: > > > Is this just an inevitable numerical noise? > > very likely. Some numerical noise is always there, it > is usually almost invisible, especially at scf convergence. > In some known cases of special sensitivity to truncation > and roundoff errors: noncolinear, spin-orbit, DFT+U, it > is more visible while self-consistency is reached. No > easy solution is known (to me at least) > > > P. > > -- > Paolo Giannozzi, Dept. Chemistry&Physics&Environment, > Univ. Udine, via delle Scienze 208, 33100 Udine, Italy > Phone +39-0432-558216, fax +39-0432-558222 > > _______________________________________________ > Pw_forum mailing list > [email protected] > http://pwscf.org/mailman/listinfo/pw_forum >
_______________________________________________ Pw_forum mailing list [email protected] http://pwscf.org/mailman/listinfo/pw_forum
