Author Topic: Different QMMM results obtained on two machine.  (Read 3749 times)

biomd

  • Newbie
  • *
  • Posts: 4
  • Karma: +0/-0
Different QMMM results obtained on two machine.
« on: July 21, 2013, 02:55:37 am »
Dear all,

I have optimized one protien-ligand system by QMMM using TURBOMOLE and DL_POLY. The QM method is D3LYP/6-31G(d,p). In I7, three cpu were used for the calculation and the final QMMM energy is -1451.846262 (a.u.). In another Opteron 6174 AMD , 22 core were used for the calculation, the final QMMM energy is  -1451.802698 (a.u.). The energy gap is 0.043564 a.u., (27.336823858kcal/mol ), it seems the energy difference is quit large . The same version of TURBOMOLE and same input files were used. A litttle difference is that ChemShell 3.5 and 3.3 version was used seperately for the calculation.  Any comment would be much appreciated

biomd
« Last Edit: July 21, 2013, 03:29:08 am by biomd »

uwe

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 405
  • Karma: +0/-0
Re: Different QMMM results obtained on two machine.
« Reply #1 on: August 02, 2013, 03:48:13 pm »
Hi,

it is hard to say what happened, but Turbomole must give the same results if the input is identical, no matter on which hardware it runs. You could run the test suite (cd $TURBODIR/TURBOTEST; TTEST) on both machines to check if there is a general problem on one of them.

But, as a matter of fact, the same binaries with the same input can indeed differ in a sense that optimized linear algebra libraries like Intel's MKL or AMDs ACML are used which internally run different code for the same task, depending on which CPU they are. It is quite unlikely but not impossible that they give slightly different results - but only within the given convergence thresholds...

Did you compare the two control and basis set files? The name 6-31G(d,p) is not part of the official Turbomole basis set library, but 6-31G** is - which should be the very same basis set. So it might be that different basis sets have been used...

I'd recommend to send your input files to the Turbomole support (turbomole@cosmologic.de) to find out where the difference really comes from.

Regards,

Uwe