Ari, If you let me intervene a little bit; how did you(?) compile the code? > Some (most?) recent Intel compiler suites are known to be very buggy, > gcc/gfortran is much more reliable. Did you try first to create a serial > executable, should your MPI have an issue?
I tried GCC 5.4.0 and 4.8.5 with OpenMPI 2.0.2 and IntelMPI 2017. I have used this version of OpenMPI to compile other packages which pass tests. I think Paolo answered my questions. Thanks! Paolo, These are exactly the tests that give some small discrepancies. It is quite > possible that some references have not been updated. I am no longer > following what is happening in testing and hope that somebody will > eventually take over this aspect. Okay, that is the information I was looking for. Thank you so much! I grabbed b3lyp-O.in and ran this numerous times with different settings >> (conv_thr and electron_maxstep), different numbers of processors (1-8), and >> larger cutoffs (doubled ecutwfc and ecutfock). >> if you change the cutoffs, you get different results, for sure. > > Just for reference, I ran the code 8 times when I changed the cutoffs. I was comparing test-to-test not test-to-reference. Hope this helps in some way. Thanks again, Barry On Sat, May 27, 2017 at 4:14 PM, Paolo Giannozzi <p.gianno...@gmail.com> wrote: > On Tue, May 23, 2017 at 3:26 PM, Barry Moore <moore0...@gmail.com> wrote: > >> >> pw_b3lyp b3lyp-O >> pw_dft dft8 >> pw_noncolin noncolin >> noncolin-constrain_atomic >> noncolin-constrain_total >> pw_relax relax-bfgs_ndim3 >> pw_pawatom paw-vcbfgs >> pw_vdw vdw-d >> vdw1 >> vdw2 >> vdw5 >> vdw6 >> pw_xdm xdm >> > > These are exactly the tests that give some small discrepancies. It is > quite possible that some references have not been updated. I am no longer > following what is happening in testing and hope that somebody will > eventually take over this aspect. > > I grabbed b3lyp-O.in and ran this numerous times with different settings >> (conv_thr and electron_maxstep), different numbers of processors (1-8), and >> larger cutoffs (doubled ecutwfc and ecutfock). >> > > if you change the cutoffs, you get different results, for sure. > > I get a different result every time AND on some numbers of processors the >> SCF doesn't converge... I also noticed that the oxygen atom is slightly >> offset from center and moving it to the origin helps (the convergence >> issues on differing processors remains). I was hoping if you could comment >> on these tests too. >> > > if you have a case in which two different runs for the same data give > significantly different results, please provide an input an the related > outputs. > > Paolo > > >> - Barry >> >> On Mon, May 15, 2017 at 3:10 PM, Paolo Giannozzi <p.gianno...@gmail.com> >> wrote: >> >>> On Mon, May 15, 2017 at 8:16 PM, Barry Moore <moore0...@gmail.com> >>> wrote: >>> >>> > Thank you. Is there any documentation where I can quickly see which >>> tests >>> > this issue would creep in? >>> >>> no, it's quite unpredictable >>> >>> > I will check the other failing tests with better conv_thr. >>> >>> you will need to re-run reference output as well >>> >>> Paolo >>> -- >>> Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche, >>> Univ. Udine, via delle Scienze 208, 33100 Udine, Italy >>> Phone +39-0432-558216, fax +39-0432-558222 >>> _______________________________________________ >>> Pw_forum mailing list >>> Pw_forum@pwscf.org >>> http://pwscf.org/mailman/listinfo/pw_forum >>> >> >> >> >> -- >> Barry E Moore II, PhD >> E-mail: bmoor...@pitt.edu >> >> Assistant Research Professor >> Center for Simulation and Modeling >> University of Pittsburgh >> Pittsburgh, PA 15260 >> >> _______________________________________________ >> Pw_forum mailing list >> Pw_forum@pwscf.org >> http://pwscf.org/mailman/listinfo/pw_forum >> > > > > -- > Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche, > Univ. Udine, via delle Scienze 208, 33100 Udine, Italy > Phone +39-0432-558216 <+39%200432%20558216>, fax +39-0432-558222 > <+39%200432%20558222> > > > _______________________________________________ > Pw_forum mailing list > Pw_forum@pwscf.org > http://pwscf.org/mailman/listinfo/pw_forum > -- Barry E Moore II, PhD E-mail: bmoor...@pitt.edu Assistant Research Professor Center for Simulation and Modeling University of Pittsburgh Pittsburgh, PA 15260
_______________________________________________ Pw_forum mailing list Pw_forum@pwscf.org http://pwscf.org/mailman/listinfo/pw_forum