Posted by Roland Lindh on June 02, 2015 at 11:03:58:
Hi,
during the last months comparisons of the MOLCAS CASPT2 code against two independent machine generated version of the CASPT2 algorithm revealed an error in the original Lund produced CASPT2 module. This is an error (a change of sign for a term on a single line of code) which has been present ever since the code was developed in the early 1990's. A report of this will be published soon. I would like to take this opportunity to report this to you before publication. The error has been verified and corrected by the Lund group, updates for both 8.0 and 8.1 versions are now committed to the git repositories. The Uppsala and Lund groups have assessed the significance of this error for both the spectroscopy of organic compounds (some 300+ excitation energies) and for some transition metal compounds. We have found that the average error in the excitation energies is of the amount of about 0.001 eV, with a maximum error of 0.02 eV. This error is well below the error of the CASPT2 method it self, which is about 0.2-0.3 eV.
If you have some questions about the details contact Per-Åke Malmqvist for a preprint of the report.
Best Regards,
Roland