Author Topic: Strange dependence of aoforce on "$scratch files grad"  (Read 2269 times)

Hauke

  • Full Member
  • ***
  • Posts: 37
  • Karma: +0/-0
Strange dependence of aoforce on "$scratch files grad"
« on: February 08, 2013, 01:25:57 am »


I  found out that when I'm doing an aoforce run I get totally (several 100 cm^-1) different frequencies if I put the following two lines into the control file to  the path for this scratch file explicitly or if I omit them.

Quote
$scratch files
     grad ddens /scr/hauke/tm/ddens_grad

The actual path seems to be not of importance, of course you should change it to some directory that exists, is writable and empty before each run. The same was made sure for the in both cases specified $TMPDIR .

I tested it with (RI)-DFT using TM 6.3.1(SMP parallel) on different molecules and basis sets, functionals... Even in a very simple calculation of ethen I got the difference, see the attached files (I know that these settings are not suitable to get reliable results but just to see the effect fast). As the forum doesn't allow attachments without ending I had to put ".txt" at all filenames before I uploaded them. Of course this ".txt" endings have to be removed again.

Without the above stated two lines
Quote
mode               7        8        9       10       11       12
frequency         800.87   913.37   924.53  1025.29  1189.97  1330.68
and with the two lines
Quote
mode               7        8        9       10       11       12
frequency        1166.52  1288.92  1297.98  1329.34  1384.57  1437.16

Maybe one of you can test it on their machines and give a comment. I have no clue what could cause such an difference. As I don't see any error this is especially dangerous.


By the way, is there a reason why in the Thread of how to run TM efficiently in parallel mode for writhing out the ddens different filenames where used
Quote
  dscf ddens /localscratch/ddens_user
  grad ddens /localscratch/ddens
Or is it save/recommended [would for example save disk space] to writhe them to the same destination (of course different calculations running at the same time should not use the same scratch file overwriting each others state)?