2017-10-03 23:13 GMT+02:00 Muhammad Shadman <[email protected]>: > Dears > > Thank you for your attentions, > > 1- we have a periodic system with 164 atom. all calculations have been > done by 3.2. > So, we used "SpinPolarized T" for our goal. The ubuntu and single core of > cpu have been used for all simulations. You can find the "fdf" and "out" > files of one of them in attachment which have zipped and named 3.2. > It should be noted that in files of 3.2, we used "SpinPolarized T" and you > can observe in output file, the following internal auxiliary supercell: > > "superc: Internal auxiliary supercell: 1 x 1 x 4 = 4" > > 2- after our investigation, we find that the "SpinPolarized T" is not > proper for us while we need to use "Spin non-collinear" > So we change 3.2 to 4.1-b3 and start to same prevois run only with change > "SpinPolarized T" to "Spin non-collinear". You can find the "fdf" and > "out" files of one of them in attachment which have zipped and named > 4.1-b3. However, this run has not completed yet. here, we find in output > internal auxiliary supercell change to follow: > superc: Internal auxiliary supercell: 3 x 3 x 5 = 45 > > Let us to know for 2 questions > A) why does changing the "SpinPolarized T" to "Spin non-collinear"change > Internal auxiliary supercell form 4 to 45? > 2) in 4.1-b3, when we use "SpinPolarized T", Internal auxiliary supercell > is 3 x 3 x 5 = 45!!!!!!!!! > As you see, it also changes the auxiliary supercell for spinpolarized calculations.
Again, the auxiliary supercell is NOT the reason you experience the slow convergence. It is because you are now performing a non-colinear calculation. Secondly, several defaults have changed from 3.2 to 4.1-b3. For instance it now defaults to mix the Hamiltonian, rather than the density matrix. In fact, I would suggest you to re-perform the spin-polarized calculation with 4.1-b3 to assert the results are qualitatively the same between 3.2 and 4.1-b3. > > In summery, we think that "Spin non-collinear" is very hard to calculate, > but we think here Internal auxiliary supercell makes difficulty to finish > a cycle of DFT > Am I wrong? > Yes. The inconsistency you find is merely due to the change between spin-polarized and non-colinear calculations in conjunction with changes in the default options. > Sincerely Yours > -------------------------------------------------------------------------------- > > Muhammad Shadman > -------------------------------------------------------------------------------- > > Assistant Professor of Physical Chemistry (PhD), > Group of Physical Chemistry, > Department of Chemistry > Faculty of Science, > University of Zanjan, > Zanjan, > Iran. > P.O.Box: 313, > Postal Code: 45371-38111 > Mobile no.: +98-912-671-4198 <+98%20912%20671%204198> > Tel & Fax: +98-24-3305-2583 <+98%2024%203305%202583>, 2477 > web page: > http://www.znu.ac.ir/members/shadman_muhammad > -------------------------------------------------------------------------------- > > email: > [email protected] > [email protected] > [email protected] > ------------------------------------------------------------ > -------------------- > > > On Tue, Oct 3, 2017 at 23:35, Nick Papior <[email protected]> wrote: > > Dear Muhammad, > > In order to investigate this it would help if you elaborate your problem. > > 1) It is unclear which differences you experience between 3.2 and 4.1-b3. > If you could please elaborate the differences you experience. I.e. attach > an output of siesta for 3.2 and for 4.1-b3 using the same number of cores > and same input file. > 2) It is not clear whether you only did non-colinear using 4.1-b3, if so, > then you cannot compare against a spin-polarized calculation in terms of > performance. A non-colinear calculation is much more difficult to converge > beside the fact that the diagonalization is twice the size of a > spin-polarized calculation. > 3) The size of the auxiliary supercell has little influence on the > performance and/or the convergence. So I do not suspect this to be of > importance. > > To summarize, please attach more specific differences for us to help you. > > > 2017-10-02 20:25 GMT+02:00 Muhammad Shadman <[email protected]>: > > Dear exclusive developers of SIESTA > > At first, we must thank you for your code. Up to now we used the 3.2 > version of it. After releasing the last version siesta-4.1-b3 it was our > interest to calculate <spin non-collinear> instead of <spin polarized T> > for our crystallize materials. In version of 3.2 we have not any problem > even with single core calculation, and our runs done for maximum of 1 day. > Now with new version we observed strange state as follow: > > superc: Internal auxiliary supercell: 3 x 3 x 5 = 45 > superc: Number of atoms, orbitals, and projectors: 7380 91080 112410 > > we find that any run with omitted spin shows the above super cell = 45 > I am afraid this is probability be a bug!!!! > > As you know, this makes very long time simulation calculation even with > mpi method. > for the best investigation and response to us you can find our fdf file in > attachment. > Let us know if you can check it out and solve it. > > > > > > > > Sincerely Yours > -------------------------------------------------------------------------------- > > Muhammad Shadman > -------------------------------------------------------------------------------- > > Assistant Professor of Physical Chemistry (PhD), > Group of Physical Chemistry, > Department of Chemistry > Faculty of Science, > University of Zanjan, > Zanjan, > Iran. > P.O.Box: 313, > Postal Code: 45371-38111 > Mobile no.: +98-912-671-4198 > Tel & Fax: +98-24-3305-2583, 2477 > web page: > http://www.znu.ac.ir/members/shadman_muhammad > -------------------------------------------------------------------------------- > > email: > [email protected] > [email protected] > [email protected] > ------------------------------------------------------------ > -------------------- > > > > > -- > Kind regards Nick > > -- Kind regards Nick
