I think I may have found an error. I have been running RI-MP2 single point energy calculations on ~30 atom ribonucleosides. I'm using NWCHEM 6.0 on TACC's Lonestar machine (
link). Everything seemed ok with the aug-cc-pvdz and aug-cc-pvtz basis sets. But when I tried aug-cc-pvqz, I got really crazy total energies. Here is an example with cytidine:
Conformation 1 (aug-cc-pvqz)
conf1.nw :
SCF energy -886.425706868216
Correlation energy -30.263926310512
Total MP2 energy -916.689633178728
Conformation 2 (aug-cc-pvqz)
conf2.nw :
SCF energy -886.423257496456
Correlation energy -4.233108269289
Total MP2 energy -890.656365765745
The difference in correlation energy leads to a ~16,000 kcal/mol energy difference between the conformations. Just for comparison, here are the correlation energies at aug-cc-pvtz:
Conformation 1: Correlation energy -3.606311785092
Conformation 2: Correlation energy -3.606330351042
Here are some things I've tried, with no success:
- Latest development version of NWCHEM
- Removing COSMO
- Using a direct calculation (vs semidirect)
- with or without "freeze atomic"
- Using a different computer
Thanks for any help you can give,
--Niel
PS
I reported this problem:
http://nwchemgit.github.io/Special_AWCforum/st/id62/Odd_correlation_energies_for_R...
Might it be related?