[molpro-user] Two runs, same HF energy, different correlation E

[Mikael Johansson]

Hello All!

I ran a CCSD(T)/aug-cc-pV5Z calc on CCl4, and got some discomforting results. I 
started two jobs at the same time, as the jobs are prone to crashing due to the 
abysmal disk on the system I'm using. The reference HF energy is the same, but 
all the correlation energies are different:

           run1            run2
HF:      -1875.93858508  -1875.93858508
MP2:     -1876.99833304  -1876.99831334
SCS-MP2: -1876.98701710  -1876.98697088
CCSD:    -1877.03033563  -1877.02940796
CCSD(T): -1877.08949505  -1877.08842777

The only difference in the input files was the run1 (for some reason) had 
"memory,990,m" and run2 had "memory,900,m" (and different names for 
"file/punch")

Now my questions are, is there any way of figuring out which of the energies 
are correct? Can some check be performed to see if the calculations have 
produced good or bad numbers? What could be the reason for the two different 
results? I'm running the job a third time, but even if I wouldn't get a third 
different results, I'm not that comfortable deciding on the correct energy by 
majority vote...

The runs were done in parallel on 12 cores/nodes with Molpro 2009.1 patchlevel 
26.

Any help or comments appreciated, have a nice day,
      Mikael J.
      http://www.iki.fi/~mpjohans/