Author Topic: Turbomole 6.1 released  (Read 7597 times)

uwe

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 560
  • Karma: +0/-0
Turbomole 6.1 released
« on: October 26, 2009, 04:07:18 PM »
Release Notes of Turbomole 6.1

 
  • New features:
    • N4 spin scaling SOS-RI-MP2 (module ricc2)
    • faster MP2-F12 (module ricc2)
    • (One-electron) transition moments between excited states in CC2
    • Douglas-Kroll-Hess Energies in C1 
    • OpenMP versions of ricc2 and dscf  (currently only Linux/PC 64bit binaries available)
    • new parallel version of ridft and rdgrad, especially for SMP systems (currently only Linux/PC 64bit binaries available)
           BETA version - passes all tests, but needs to be watched in  production runs !
           including:
           
      • parallel linear scaling exact HF-exchange for DFT hybrid
                  functionals (energy and gradients)
      • parallel RI-K routines for energy calculations, faster
                  pre-step for parallel RI-MP2 and RI-CC2 calculations
      • no master process needed any more
            There is a new website which describes how to test the new parallel version and what to do if an error occurs:

               http://www.cosmologic.de/parallel-faq.html   
     
  • Efficiency:
    • up to 30% faster binaries for current AMD CPUs
    • better speed up of new parallel versions compared to the standard MPI version, especially on SMP systems (ridft, rdgrad, and ricc2)
    • standard MPI version with reduced CPU usage for the master process


 
  • Usability:
    • TmoleX 2.2 with several bug fixes for running Turbomole jobs on remote systems


  • Bugfixes:
    • DFT-D aoforce calculations (for some structures 6.0 gave wrong  frequencies)
    • external E-Field crash on AMD CPUs fixed (ridft)
    • wfn output files containing f functions fixed with $wfn option
    • initial start velocity of molecular dynamics set ups now oriented to what the user has entered as temperature (mdprep)
    • fixed crashes on some CPU types of Intel and AMD when using highly  optimized linear algebra routines by using the latest Intel MKL library (Intel: spin-orbit calculations, AMD: aoforce)
    • NumForce and TmoleX fixes for parallel runs or remote jobs when having tcsh as default shell on target machines
    • Ir basis set in Turbomole 6.0 had a missing s function in def-TZVP
    • new keyword ($lastdiag) increases the accuracy of the output of orbital energies for very small HOMO-LUMO gaps - helpful for TDDFT calculations which complain about the violation of the Aufbau principle