Author Topic: Turbomole 7.4.1 released  (Read 4890 times)

uwe

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 558
  • Karma: +0/-0
Turbomole 7.4.1 released
« on: November 25, 2019, 09:26:29 AM »
Dear Turbomole Users,

Turbomole version 7.4.1 has been released, this is a bugfix version for 7.4 and mainly resolves technical issues with Intel MPI.

Initially it was assumed that there is a bug in riper when using ECPs, but this turned out to be not correct, see Marek Sierkas findings:

Quote
After running extensive test and comparing with other programs using the
same ECP and basis sets, it turned out that the ECP implementation in
Riper is correct. The reason for the strange results is the parameterization
of the default Stuttgart ECP [D. Andrae, U. Haeussermann, M. Dolg,
H. Stoll, H. Preuss, Theor. Chim. Acta 77 (1990) 123.]. These ECP were
not derived to describe periodic systems, in particular metals and metal
surfaces. They were fitted to reproduce calculations on atoms and very
small molecules.

see the full post here: http://www.turbo-forum.com/index.php/topic,1166.0.html


The changes included in 7.4.1 are:

  • usage of Intel MPI 2019.5 avoids errors that occurred on most  Infiniband systems
     
      Typical error messages in 7.4 have been:

      MPIDI_OFI_send_handler(685).....: OFI tagged inject failed

      OFI addrinfo() failed
       (ofi_init.h:689:MPIDI_NM_mpi_init_hook:No data available)

      (../../../../../src/pm/i_hydra/mpiexec/mpiexec.c:2024):
         downstream was killed by signal 11 (Segmentation fault)

      -> all those errors have been solved in 7.4.0 by setting Intel MPI
         environment variables (like FI_PROVIDER), but with 7.4.1
         this is not necessary any more. There is no need to apply
         any additional settings.

      In some cases the parallel version using MPI showed poor performance
      with increasing number of processes, this was fixed.

  •   bugfix: the charge keyword in the xtb section was not correctly treated

  •   bugfix: plotting of periodic molecular orbitals with riper was faulty
              in former Turbomole versions

  •   new feature: new custom k-point sets can be provided in the input

  •   documentation for xtb was missing in 7.4.0, this has been added

  •   new feature: gcp can be used also for other cases than PBEh-3c and HF-3c

  •   new basis set for X2C NMR (x2c-SVP-s, x2c-TZVP-s, x2c-SVPall-s,...)

  •   bugfix: scaling in local hybrid gradients was modified

  •   bugfix: freeh used wrong units for cutoffs of imaginary frequencies
              when using the free rotor option

  •   bugfix: linearity check in aoforce has been improved

Bugfixes for TmoleX:

  • UV/Vis spectra using TDDFT failed to run in job templates with
       an error message about faulty $soes settings
  • xtb as preoptimizer in job templates followed by an RI-DFT
       geometry optimization did not switch on RI for the second step
  • import of scan or batch jobs from old TmoleX versions failed to
        run because the import option for directories was missing
  • several settings in the periodic boundary condition panel
        worked unreliably