Hi, I am getting a very strange behavior, when I compile with (lapack_double.o in vasp.4.lib) - and I need to do that to make that insidious ZHEGV error to go away. So when I compile my 4.6 with lapack_double.o, my energies are about x8 from where they were when I used external LABPACK. I got test files from collegues with the bench energies and my external LABPACK energies agree, but build in LAPACK yiels about 8 times higher energies
thanks
erroneous energies with lapack_double.o compilation
Moderators: Global Moderator, Moderator
-
- Newbie
- Posts: 11
- Joined: Thu Nov 28, 2013 3:25 pm
- License Nr.: 1266
erroneous energies with lapack_double.o compilation
Last edited by jasius on Sat Feb 15, 2014 5:23 pm, edited 1 time in total.
-
- Administrator
- Posts: 2921
- Joined: Tue Aug 03, 2004 8:18 am
- License Nr.: 458
erroneous energies with lapack_double.o compilation
please first of all try to find out
1) which of the results are the correct ones
2) if the serial and the parallel executables of vasp show the same error (or if eg the factor in the energy scales with the number of CPUs?)
--> please run eg. one of the vasp-workshop (sessions 1-4) examples
http://cms.mpi.univie.ac.at/wiki/index. ... lculations
to check which of the 2 lapacks yields the correct energy (the numbers may not be consistent up to the very last digit, but certainly they will be close enough to one of your results so you can check which one is the correct one).
1) which of the results are the correct ones
2) if the serial and the parallel executables of vasp show the same error (or if eg the factor in the energy scales with the number of CPUs?)
--> please run eg. one of the vasp-workshop (sessions 1-4) examples
http://cms.mpi.univie.ac.at/wiki/index. ... lculations
to check which of the 2 lapacks yields the correct energy (the numbers may not be consistent up to the very last digit, but certainly they will be close enough to one of your results so you can check which one is the correct one).
Last edited by admin on Thu Feb 20, 2014 2:33 pm, edited 1 time in total.